Igors lab released another (german) video today where he talked about the capacitor topic again and to the wrong names he said that technically they arent POSCAPS but in the industry it is a common habit to call them that. Igor said it stands for "piece of shit caps" haha😂
Before Buildzoid: That Zotac Trinity looks kinda pretty this year After Buildzoid: Can't wait to see oscilloscope graphs comparing 220, 330 and 470 polymer tantalum smd's to multilayer ceramics
Omg so true. For me. Before buildzoid. RTX 3080FE looks gorgeous and is stupid fast. After buildzoid: I wonder if its possible to deliberately fluxuate voltage using a carefully designed chip to guess and match blips due workload spikes an attempt to smooth out power delivery.
EVGA: "During our mass production QC testing we discovered a full 6 POSCAPs solution cannot pass the real world applications testing. It took almost a week of R&D effort to find the cause and reduce the POSCAPs to 4 and add 20 MLCC caps prior to shipping production boards, this is why the EVGA GeForce RTX 3080 FTW3 series was delayed at launch. There were no 6 POSCAP production EVGA GeForce RTX 3080 FTW3 boards shipped." COLORFUL: The manufacturer was the first to report on a issue unofficially to the press. Review samples that were already sent out, were later recalled. ASUS ASUS has changed the design in pre-production. No official statement was made. MSI MSI acknowledged the issue during the MSI Insider Livestream, suggested it could be a driver issue. ZOTAC Igor’sLAB reports: PC Partner, Zotac’s mother company, seems to have recognized this and is obviously changing its cards.
First blackballing gamers nexus for thorough reviews that instead of responding with launching a Rev2 board like an adult, now this. MSI effing up lately.
@@vyor8837 Yeah, you bet drivers will cause MSI some issues. Toyota, Honda, Ford, and Beemer driving PC Master Race will be cueing up round the block to run the fecks down as they exit the office!
We need to start a campaign: Get LMG, JayzTwoCents, GN - anybody who has cards - to send Buildzoid a selection of cards to "stab with an oscilloscope".
Yes, watched Jayz first, just to see how he would try to spin this issue and found out he doesn't know what to call the capacitors, even after post editing!!!! GN work with Buildzoid, so that would seem the natural route. Don't know if you follow NAAF, but Paul posted fairly quickly, it could be a power supply (Main), proplem, not for the same reason as Buildzoid, but a question mark has hovered over these GPU's before any release? Above my pay grade, but why are they rated , and does a noisy power supply play a part, even if adequate in wattage? Buidzoid needs a 3080, fast!!!
You forgot the rest You: Yeah Nvidia: go to the last crossroads of the darkest road you can find. Sell us your soul and all of your information, then you can play a game of dice to possibly win a blob of ones and zeros that might work. You(&me): wtf, should've went AMD damn it.
In his (german) video, Igor explains that the big ones usually are either aluminium or tantalum, so i guess he's well aware of the different CAPs used, just didn't know (or didn't look up) the names for the different series
He is aware. But what BZ is not aware of is this: everybody in the industry calls them POSCAPS. It is commonly accepted to say POSCAPS. Engineers at the AiB even joke about them being the Piece-of-Shit CAPS. EVGA also refers to POSCAPS in their statement because of this.
@@tanmaypanadi1414 No, it has nothing to do with insufficient research on Igors part, many people just call them POSCAPS in a general way, even if they are not specifically refering to the Panasonic capacitor. There are many more brand names that over time turned into generic names for a specific item, such as kleenex, Jet ski, bubble wrap, Tupperware etc etc. These are all brand specific copyrighted terms people today use to descibe a specific item. Imagine if I gave you a lecture everytime you descibed plastic dishes as "Tupper ware"... that's essentially what AHOC is doing. AHOC is wasting several minutes nitpicking on this non-issue, this unfortunatly wrongly leads some poeple to believe that he just made a major revelation and Igor made a massive mistake...
There's a funny quote from Igor's article where he explains why engineers tend to call all of those capacitors POSCAPS even though they're not the ones specifically made by Panasonic. It's because the engineers like to call them piece-of-shit caps.
Igor: It's all because cheap POSCAPs! YT Reviewer: POSCAPs is a LEMON! Customer: I don't want that POSCAPs on my card! Panasonic: What the fuck is this?!
Not now, it’s been like that since Turing as we were all willing ready to pay 1500$ for beta access to a feature that still doesn’t work properly on Turing and barely works on Ampere.
@Masterr Laster With new model yes there is always a risk. Beta testing is meant to catch most of these issues and make the product acceptable to use for an end consumer. Some bugs may remain after beta testing that may be generated from a much smaller probability and its harder to catch. However this time the hype around 3080 was so much that people would bought it regardless whether any of the cards were reviewed or not. This is like iphone bending where a product that sold in massive numbers had a fatal design flaw that it would bend in real world usage. The fact that such a major problem was caught this early and replicated easily by most people means beta testing was non existent for the non founder cards. This is especially a problem since this cannot be solved with a firmware update, it's an actual hardware deficiency that cannot be fixed. Why was the beta testing skipped ? Mostly to cater to the hype behind the cards themselves and get the product to the end user asap regardless. Only those who went all the way in using the best components all around probably knew what they were doing or played it safe by over engineering.
@@solomonshv I don't have one friend does I normally wait for these irregularities to iron themselves out, good card otherwise sorta poorly planned/tested launch. But there have been worse much worse...
@@DanielWW2 i dont O.o There fired most of the old marketing. Thats why we dont have any of the amd rebelion crap. And almost no leaks compared to Vega. The new team isnt as cringy but the fortnite thing leaves a bad taste in my mouth. And people roasted them enough for gamecache 😂 But If true im interessted in what they call the huge cache on rdna2.
@@SuperUltimateLP I know they fired them and I am happy with that. Still I am a bit hesitant because so far I have seen some questionable stuff again. Not nearly as stupid as Vega or whatever the hell Nvidia is doing right now, but I was also not a fan of the Fortnite cooler unveil, nor the twitter marketing teases. That is the type of stuff that can backfire massively. They better deliver because they are being quite cocky. Especially now that Nvidia is making such a mess for itself.
After watching Jayz video, this is so helpful. Buildzoid goes into so much more detail than Jay, and yet remains much clearer and understandable - a sign of a competent and good teacher!
Jay is a clown, he's proof that production and budget are more important than having anything important or interesting to say, or AHOC would have 10 million subscribers, literally no one else will give you a detailed breakdown of a VRM from an EE perspective.
@ZPinch Yup... and that's how you end up with 6 SPCAPS and 0 MLCC on a PCB because someone didn't really listen to the "sperg" and was just like "let's be cheap". In extremely technical and complex fields, details matter... A LOT!
First time watch AHOC. 42 minutes of great capacitor discussion and comparing cards and capacitor configurations, and then a blazingly fast self-promo schpeel at the end that if you blink you'll miss! LOVE IT! Thanks for a entertaining and educational discussion on the capacitor types and configurations! Very much enjoyed it!
Humble request, it would be really interesting to see you cap mod one of the lower tier 3080s, like a ZOTAC Trinity non-OC. And, like, do some benching before and after with an unlimited power mod to hit 2,000mhz+ Related, it was really cool to hear that you helped the dude from Frame Chasers figure out how to shunt mod his 3080 :D
You do not need a mod to hit 2k. It will boost to 2k on its own stock as long as you have good cooling. That is why all these people are reporting this as an issue. It is not a manual OC. If you have a water block on any of the cards you will hit 2k easy with GPU boost.
Assuming Moore's Law is Dead is right, NVIDIA contractually obliged AIB partners to sell at cost or higher, while FE cards are sold at a loss (the MSRP). In this way NVIDIA can sell small amounts of FE cards at above minimum spec to give a misleading perception of Ampere perf per $ compared to the AIB partner cards that most people will be able to buy, which will either be technically inferior or more expensive.
@@BruceLeedar MLiD is literally never right about anything. He claimed it was 'absolutely established' that the 30-series FE cooler would be tri-axial, for example.
@@BruceLeedar There's no misleading perception of Ampere Performance/$, there are partner models that cost the MSRP and perform the same as the FE, even if they might have less overkill parts, they still can be enough for what they are supposed to deliver, which is the stock performance. Also assuming that someone that missed most of the speculation they made until now is right about a speculation doesn't make much sense for me.
*me at the start of the video* : "wtf is a poscap? A really fancy popsicle?" *me after 10 minutes of buildzoid rambling about poscaps* : "that's OBVIOUSLY not a poscap, cmon, man! JEEEEEEZ, IGNORANCE ALL OVER THE TABLE! And that dude calls himself a JOURNALIST, UNACCEPTABLE"
and a real pro says: hey, it's not deal, poscap name is used by engineers as a general synonym for polymere-capacitor. So why talking about that much? It only impress a bunch of noobs, that for sure! I wish Buildzoid would make some real investigative work instead wasting time by judging about things by looking at pictures.
33:00 yep you're totally right I work in the industry, smaller components not only slow down the machine but create a lot more machine stops/alarms, also more problematic for video rejection of defects and people don't realize how much downtime costs in such a complex industry, it's also more wear etc...
Yeah but how much do RMAs cost because you tried to save a couple of bucks in manufacturing? Not to mention the hit to your reputation which is your most valuable commodity
@@longjohn526 If nvidia sad that this cheaper config works and it turns out it dosnt its still nvidia at fault here. I guess this has to do with samsungs inferior quality. That some cards work with this cheaper desing and some dosnt.
I've worked in heavy industry a long time ago. The tolerance on truck engine components are very small and even that can cause mayhem. I still don't comprehend how electronics can be manufacturerd like they are, so yeah makes total sense
8:13 is when he stops talking about poscaps and the differences between the type of caps on the card and the fact that Panasonic isnt the only capacitor maker.
Hahaha. I was in the grocery store. Went in, found eggs, grabbed beer, and checked out before he finished. I was like, dude, i mean... cool but chill. 🤪
@@FcoEnriquePerez I knew about the difference between a poscap and SPcap already. Its for those of us that get that they arent the same thing, and a large part of the info up until that point is redundant.
I haven’t seen Asus being this competent with their TUF cards compared to their competition in lord knows how long, I was really impressed with their cooler design and the PCB and then I thought Gigabyte and MSI would have better models but I haven’t seen any better models yet.
I feel like I dodged a bullet my TUF 3080 is arriving Monday and I woke up to this news, hopefully there's no isses with it but so far it seems like it's ok.
WayStedYou and that’s why I said they were incompetent for a long time, 5700XT TUF’s were really fucked up especially not having any memory cooling but 2.0s are not bad. I didn’t like them at first but you just can’t deny the numbers.
EVGA admitted that the FTWIII launch was delayed because of this capacitor bs. Pretty obvious then that ASUS Strix was delayed for the same reason. It's 100% confirmed because if you look at the launch day photos of the Strix you can see it also had those big caps on the back, where as now they're using six arrays of MLCC's. ASUS found out and had to delay their launch so they had time to fix the card.
@@Android-ng1wn If I was the unfortunate owner of one of these cards, the one thing you may be able to do to improve your situation, if you have balls of steel and know what you are doing, is to stack two X7R ceramic caps sideways on top of the tantalum caps, of the same value as the normal ceramic caps in the reference design. It's a low risk maneuver, IF you have decent soldering skills and a decently fine soldering pen and good tweezers. Just stack more caps on top of the ones already there. However if you do this heat may become an issue for the tantalum cap, and tantalum caps tend to explode when they overheat. So do this at your own risk. The better option would be to remove the tantalum caps from the board, and replace them all with X7R ceramics, but I don't think anyone will even accept your money to attempt that... Boards like this are not made to be soldered to with an iron, it's all done with wave soldering at the factory... The bulk of the capacitance usually is with much larger bulk caps near the VRMs. Bypass caps are there to filter noise from the GPU.
@@RngmonsterX nice find... Atleast if you pre orderd a card you can rest at ease. Actually im glad this was found early.. Saves alot of hassle in the future for buyers.
kougar7 Yep They did run problems and did last minute change. I wonder if They go for mixed configuration when They found out how Many ceramics is needed and how Many polyvaps They can actually use. Most likely tuf will get bigger cut and strix smaller because strix is clocked higher... we have to wait and see. The asus did not chose ar random to use those cheaper first. Making the change did cut their margins and They want to get those margins as fat as possible...
Except no. Bypass capacitors have nothing to do with storing energy. This video clearly wasn't made by an electrical engineer. I know because I happen to be one.
@Zzz A/ you didn't look at all my videos. B/ I don't have much interest in making videos about my job. You're making the mistake of trusting the guy who talks the most. Volume isn't a sign of expertise. Damn he spends the first 15 minutes ranting about capacitors and some curves he doesn't seem to understand. He throws words around like "impedance" but he doesn't know what they mean, which is obvious because he never links it to frequency even though that's all this is about. It's like a guy talking about cars engines but never mentioning RPM's.
@@TheNefastor Okay okay, take it easy. Let's say I believe you, care to elaborate what happened to the rtx 3080/3090 and with their capacitors behind the gpu core in language?
@@limingtogunawan4756 I'll try to keep it simple, even though it isn't. Let's start with describing the cause of the problem. Basically, any chip that uses power will inject interference on the power supply that feeds it. This interference is essentially crosstalk from internal high-frequency signals to power lines, as well as nearly unmeasurable voltage spikes resulting from the chip's operation (transistors, when they switch, will draw and release power because of their "parasitic" gate capacitance and it adds up when you have billions on a chip). It's a very complex phenomenon that's impossible to accurately model because it depends mostly on what the GPU is running at any given time. Video encoding and 3D gaming might generate entirely different interference. Anyway, this interference, if left unchecked, would then travel through the board to other chips powered from the same source and interfere with them too. At some point, all that mutual interference will be so much that some or all of the chips on the board won't be able to function (basically, won't be able to tell a 0 from a 1). To prevent that, you need low-pass filters that will shunt high-frequency voltage (noise) to ground, leaving only "clean" DC voltage. That's what those capacitors are for, and that's why they need to be so close to each chip : the traces from the chip to those capacitors are effectively antennas that will radiate the interference and can also receive interference. This is an RF thing : you're not just worrying about noise on the copper, but also on the air waves. So. How do you deal with this ? Like I said, you need to use capacitors to the ground that will act as low-pass filter, meaning they'll take high frequency spikes and shunt them to the ground, thus eliminating them. And like any filter, they need to be calculated for a specific *central frequency.* In practice it's more complicated than that because you'll have to deal with several different frequencies (i.e. the RAM controllers run on a different clock than the CUDA cores, etc...) but for the sake of simplicity let's just say you want to bypass interference at the GPU's core clock frequency. If that frequency is well-known, as is the case with a GPU, then you can calculate what's called the "decoupling network", a set of capacitors of different values and characteristics that will filter the frequency you don't want, and maybe its larger harmonics too. So you implement that. And lo and behold, your board works as it should. Then comes along an overclocker who takes a board meant to run at, say, 1.5 GHz and pushes it to 2 GHz. Uh oh... suddenly your decoupling network is *no longer fitting* the frequency your biggest chip is now running at. So the filter becomes less efficient, and more so the higher the frequency gets. To make matters worse, overclocking will cause the GPU to draw a lot more power... and that means your interference gets *amplified* by that much. If you're using 50% more power to reach an overclock, then your interference is also 50% stronger. You end-up in a situation with stronger interference and a filter that's no longer tuned for it. If I need to spell it out for you, that means you're screwed : you've added a turbo to your engine and now it uses more fuel than ever and runs faster, but you haven't changed your exhaust. Either the engine ends-up choking itself, or you're going to blow-up your exhaust. Either way, you're not getting a fast car, just a dangerous one. Graphics cards are designed and sold to run at a specific clock speed. The chips might be capable of more, but in the end you'll be limited by all sorts of peripheral parameters that you can't change, such as those decoupling capacitors. Unless you're crazy enough to calculate a new decoupling solution for your GPU, desolder the one it shipped with and solder in the new ones. You'd also need to do that for the RAM, by the way. That's just not a viable proposition. Now, what I've said so far is pure expertise, any digital electronics engineer will confirm. What I'm saying next is only speculation because I don't work for a GPU manufacturer : When you see a brand offer a board that says specifically "overclocked", "OC" or something similar on the package, my *assumption* is that they built it with different values for the decoupling capacitors. See, there are enough capacitors on each chip you can actually "widen" the frequency band that can be decoupled. A smart manufacturer could design a board with decoupling that's, say, 75% efficient over a really wide frequency band instead of 100% efficient over a "manufacturer-approved" frequency band. That means your "OC" board should remain stable at higher clock frequency than a non-OC boards, all other things being equal. Damn, that went on for longer than I expected. But this just tells you how complicated this topic is. That's 15 minutes of my life I'll never get back so please read it and try to understand. I'm still game if you have questions afterwards.
As soon as I saw this happening, and after Jayz2cents did a video on it, all I could think about was how it was such a perfect subject for you to cover. Definitely right up your alley. Even commented on the video mentioning it. Looks like it could potentially be modded manually by those more tecnically savvy and comfortable removing and soldering on the proper MLCC configurations. So the real question is, is it worth removing the SP caps and modding in the MLCC arrays? As long as the pads are present on the particular card, obviously. Edit - Also, it's, "vast majority of them AREN'T", not "isn't". Sorry for the grammar N*-ish, can't help myself.
"If its not stock, don't expect it to be stable" sure, but i expect the asus oc version to be stable, because i pay for it in that state... and that goes for any sold oc version.
Big thanks for actually explaining why exactly these are there and what exactly they do in the first place instead of just talking about what type is better. These extra bits of electronics knowledge may make for 40 minute videos where others make 4 minute ones, but unlike the short ones, this is not just informative, but educational. Really makes your videos stand out in the avalanche of tech channels talking about topics like this.
Yes, to a degree it is indeed Nvidia's issue that they put forth such loose specifications. Also the board partners problem due to not stress testing their product. In terms of cost, using a smattering of MLCCs and a bunch of larger polymer caps is generally a better over all solution. Since then we can get the benefit of high frequency responsiveness, and also cheap bulk capacitance. Would also have liked seeing more of the cards have some more bulk capacitance at the VRM itself. (Some after all have practically non.... Could we at least get 25% of the pads populated?) And yes, poking a card with a scope should be fairly indicative of the voltage regulation. Would do it myself if I had a card at hand, but I don't. Though, I am surprised that they didn't put in low ESL capacitors into the design. (looks like regular ceramic caps, but the connection is on the long edges of the cap, instead of the short ones.) They can after all be much bigger, both in terms of capacitance and size, while having phenomenally low ESL. (Though, I guess they are more expensive due to being niche) And your point about a poorly filtered power supply potentially lurking through the VRM is indeed a problem that can crop in. And why one generally should have some input capacitance on the card itself. Doesn't have to be much for it to be effective.
@@bustex1 The board partners did have the firmware for the cards before they shipped them out to the reviewers. THereby the board partners could do som burn in testing of their own, considering that the reviewers had the cards for a couple of days. It wouldn't be unreasonable to "ask" nVidia for at least a 48 hour burn in period before sending something out to said reviewers, if this is the case or not is a different question. But if nVidia doesn't want to do that, then frankly, nVidia is the only one to blame for enforcing poor engineering practices. And to my knowledge, at least one board partner did ask reviewers to hold off on the review due to having found problems with the card. A very admirable way to handle the situation to say the least. (Since one shouldn't ship a non working design.)
Today on the Bench yes one board partner did say they did have an issue. I saw jays two cents video he said they didn’t have drivers and that without drivers they wouldn’t have known about the crashes. Just what he said I am just restating. Idk enough about this if firmware would have caught a software crash? Yea they could power it on and test that it powers on but idk if that has anything to do with the crashes that are happening. I don’t believe it does but like I said I’m no expert just restating what someone more knowledgeable than me said.
@@bustex1 As an engineer myself, I wouldn't want something I design get pushed to market without the design having been stress tested first. If nVidia's board partners aren't willing to at least require enough time with the product before launch to fix potential issues, then nVidia isn't doing much else than forcing their board partners to rush their product to market. While nVidia retains the upper hand in having designed the GPU in the first place. Though, one can also see it as an "early adopter fee", but that isn't how one should look at poor design practices.
That's true, but apparently overclocking it makes the self boost go higher from what I heard. If correct, then GPU boost is the final blow, but overclocking the cards makes it more likely to crash.
It does, but I believe the reports are coming in even for the cards without manual OC, as in only with the OC the GPU Boost does. Cause GPU Boost doesn't know the config of the caps, just knows "Hey, I have a power and temperature headroom, let's go higher", only to then crash because the power is there, but not in such rapid burst.
From what I've seen most cards can boost to about 2GHz for a short time. Enough to get unstable and crash the 3d program. Especially in light-ish games.
Tres Waters sound a lot last minute change because of problems... intereresting to see if They reduce ceramics when They found out how much is really needed...
The two types have different characteristics and do filter out noise in different ways , Just adding more isn't going to make anything better nor will just using one type but a combination of the two will have a huge difference when designed correctly.
Thanks for getting on this Buildzoid and I have to say young man your knowledge has really expanded as you continue to test, challenge yourself and learn, extremely informative, very well done... It's NVidia's own drivers and the built in boost technology that all their cards have. Even though cards are listed or even say on the box in some cases that the max boost clock it 1710ghz or whatever the claim is for any given card. Nividia's built in Boost, sorry not sure what they call it but basically all their GPUs do this, in some cases up to 2k or more...
Buildzoid: none of the cards have POSCAPS !!! Zotac (left next to lower left gpu mounting hole): has a POSCAP ... To be fair he was focusing on the gpu area ... still I chuckled.
The problem is the gpu boost automaticly oc the gpu clocks to crash level on the cheaper configs. So you can get random issues from the gpu at stock settings which didn't happen before
When CIA Langley was being built, the HVAC contractor had a problem: CIA wouldn't tell them exactly how many people or other heat-sources were in the building...classified. When NVIDIA creates their monsters, they don't tell the PCB contractors peak power and how clean it needs to be...classified.
I'm not condoning Nvidia's malpractice with some of their board partners but man if all those leaks a few months ago were an indication I'd be weary and careful with my product designs too. We almost got a complete picture of Ampere even before the architecture released and that's a major contributing factor to the over-hyping of Ampere...all those leaky AIB employees spoiling the launch.
@@Gandalf721 Board partners didn't have enough time to test b/c Nvidia didn't give them 3000 series beta drivers until very late dummy. Nvidia treat their partners like shit and kept everything a secret, board partners had to design their board blind without any way to test it until it was too late.
The problem is, when you pay a premium to get a factory OC variant, it should work out of the box. If you pay that premium and then you have a problem which solution is "well downclock the core offset for lets say -50Mhz, while its not a lot, its not what you paid for. Now NVidia GPU Boost auto-OC the chip, so AIB have to make sure that that auto-OC will not trigger any instability. I mean the price we pay, some of that price goes into the testing and certification of those board. Seems like somethings not right here.
Poor quality PSU's being pushed hard does make sense as being a significant part of this issue. With the OC and less expensive capacitors pushing the GPU past stable OC. The graph spikes being nearly vertical point to this being a more complex issue. Cheers
ALT it is most likely combined problem. If you have highend psu it can compensate problems with capasitors, if you have better quality chip aka you did win in silicon lottery it can compensate not so good psu... using more ceramics can compensate both to certain level etc...
Actually Hardcore Overclocking : What Nvidia doesn't tell the AIB partners is the complete detailed specs sheet for the PCB circuitry due o it being a business secret so if Nvidia would give detailed specs as of why certain electrical components have to be that way, the the AIB-s wouldn't take shortcuts on that area. AIB-s earn very little today on Nvidia cards because Nvidia is selling the reference design to them very very expensively as if the AIB parters were end consumers so that AIB partners are taking little shortcuts anywhere possible is only natural. But the problems like catching fire VRM-s (because Nvidia said to AIB partners that they are so efficient that they don't need extra cooling) on the EVGA GTX 1070-s, 1080-s and 1080 Ti-s as well as overheating and artifacting GDDR6 that Nvidia again knew and intentionally forgot to mention specifically "hey AIB guys, be careful on the memory, it heats up like crazy, so construct finstacks for them, because they are made on the same node as GDDR5 but only have 50 % higher clocks". Those are just recent hardware flaws that occur due to AIB-s not getting the complete "why" detailed specifications for the PCB design from Nvidia and literally having only days to do too small number of a testing on their custom designs before Nvidia expects them to launch. This is completely Nvidias fault because Nvidia has designed one thing very good and that is almost impossible circumstances for them to have at least 3-4 months to design and properly test their PCB designs.
One issue MLCCs have is loss of capacitance with DC bias. This reduces the amount of energy they can store. The 47uF caps you looked at lose 40% at 2V. This also has the effect of introducing nasty harmonics into the supply rail. The most optimal configuration would be a mix of electrolytic/polymer and mlcc. Large electrolytic/polymer caps to store energy and supply the IC during a load transient until the smps catches up, and mlccs to bypass the high frequency component of the load transient so that the voltage excursion is smooth and kept to a minimum. It's the ESL that holds the polymer caps back, so really the ceramics only need to supply power until the current through the polymer caps ramps up to match the load. The FE probably has the most optimal configuration. You see this sort of thing on power stages. Mlccs placed by the switches to decouple the switching energy and electrolytics that decouple the load current.
If the bill of materials showing the different options is correct, then its the aib's that screwed up cheaping out. Nvidia clearly shows the top middle being mlcc's and the rest having the option for either type.
And all this capacitor stuff is so insignificant compared to VRM tuning at DC-DC controller that every vendor doing differently ;) In your MLCC plot talk - you also missing that MLCCs have much much higher impedance at lower frequencies, compared to polymer caps... Job as DC-DC engineer is to make impedance and ESR plots as FLAT and as low as possible, not just one huge notch at particular frequency, which is bad too. That is why you always see mix of huge polymer cap, smaller polymer cap and array of 0805, 0603, 0402 and sometimes 0201 MLCC everywhere, instead of just one type.
I think Nvidia had really low margins for the AIBs even compared to the last gen and therefore they kept the guidelines loose and AIBs just went a little to cheap on the cards.
Probably. All the silicon companies know what the business next door is up to. If they didn't, they would die in a year. Rumor a year or so ago was AMD sampling silicon for the consoles, results were "better than expected", which caused Nvidia to panic, pull in the Titan launch and so on. My Speculation: The fact TSMC just shrugged at Nvidias "we''ll have Samsung make it" is another corner stone of this. TSMC has the better process, Samsung is still on one of their fuckery-nodes, but Samsung has huge capacity.
@@MazeFrame Yeah I feel there might be California/Silicon-Valley politics tied up in this somehow too, with TSMC looking to establish a new fab in the US to possibly escape mainland Chinese hostile takeover, and California having sold out largely to the CCP, but who knows. I'm not sure at this point whether going with Samsung's 'loss leader' 8nm process was the 'right' choice, but at least we don't have the problem of too few open fabs. Would help imo if Intel would open up more, since their 14nm (+++...) is still fantastic
Imagine selling expensive graphic cards but you have not stress tested and played games on them to check if they are stable. Sucks if consumers has to install software and undervolt their brand new gpu to avoid crashes.
That's Nvidias fault though. Board partners couldn't test the cards before launch. Literally the only thing they were able to do, was to run an application from Nvidia that produces a computational load and then tells them if the test completed or not.
I was just about to print this polaroid with my xerox of this delicious Kraft n Cheese I made before putting it in a ziplock to go in the fridge, but all I had was a tupperware container and I didn't want to mark that with a sharpie.
I hope AMD can go past offering slightly better value for the mid range and offer something new in terms of performance (e.g flat out faster than 3080) . Not expecting much though
I think the big reason why this wasn't caught in QA is a lot of AIBs have been saying they got drivers to test against on release day basically. Actually it's possible some parts of the press had drivers before the AIBs where given them. So yeah at least part of this issue is clearly Nvidia's fault for not giving the AIB's the support they needed.
Can't wait to see all the marketing from AIBs after they come out with design revisions that fix this issue. "Super Gaming Capacitor Technology!" "Gaming CapX!" "GPU Reactor Cap!" "XCapOC!" "I'm givin' her all she's got CAPtain!"
Hey Buildzoid, I wanted to hear your take on pricing. I was surprised that everyone is saying that ceramic capacitors are more expensive. I don't know the cost the board partners pay for their orders of 100K capacitors, but in qty 10k, the ceramics are the same price (or cheaper) Panasonic SP-CAP EEF-SX0D471ER $0.57 (cheapest SP-cap I could find) Illinois / CDE XMPL series (these are on the FE) 477XMPL002MG19A $0.71 Samsung CL21A476MQYNNNG (cheapest 47 uF 0805 MLCC I could find, they are 6.8V) are $0.059 So 10 MLCCs is $0.59, SP-caps are $0.57 and the Illinois / CDE XMPL are $0.71. By the way, I suspect the yellow color is from the dielectric X5R and X7R are about that color. They have a high capacitance per volume compared with other dielectrics, so you see them on the larger capacity ones. Some of the more critical pins will be bypassed with NP0 /C0G dielectric capacitors which have much better properties, but less capacitance. They are usually more grey.
This is not non-video as you talked about the capacitors not being poscaps and then said you do not have enough info. Phenomenal job as always. I am eager to see a video of the resolution; if that ever happens... Unlikely these AIBs send you models now.
@@rosen9425 still alot of the riff raff about this situation is being spewed out of total ignorance. And it would help so much if half of the people trying to get these cards will have seen Buildzoids video here.
@@skoopsro7656 I think people who have issues with their cards will be kicking up some major whine about it, and RMA when they find these videos out soon enough
Has anyone bothered to point out to Buildzoid that the 2ghz "overclocking" is from running the cards at stock? It's the card's out-of-the-box boost behaviour that's causing crashing, this isn't users manually OCing.
Isn't the PCB layout under the caps the same for all the cards? Owning a hot-air station, flux and replacement caps it should be possible to upgrade the card, right?
HI Buildzoid watched you video on the asus strx 2080 tii flashed the xoc bios on my silent bios and it did flash but the card peforms very poorly and gpu usage is waaay down in benchmars also vf curve in MSI AB is gone any idea whats wrong??? i used a xoc bios on my 1080ti whitout any issues.
BZ in general : chill
BZ with wrong cap names : SO YOU HAVE CHOSEN DEATH
Igors lab released another (german) video today where he talked about the capacitor topic again and to the wrong names he said that technically they arent POSCAPS but in the industry it is a common habit to call them that. Igor said it stands for "piece of shit caps" haha😂
10 minutes in I still have no idea why 3080s are causing crashes, but I sure as hell now know how to identify a POSCAP
A just skipped this part after 2 minutes the point was made no need to go for 7 more minute
The voltage will drop to much because the big caps can’t stabilize it at higher frequencies. That’s it in short.
Simon Legros it’s a 47 min rant vid I can see the rabbit hole he went down
@@Todestelzer 750 us for nothing lol
it needed to be said tho :D
Before Buildzoid: That Zotac Trinity looks kinda pretty this year
After Buildzoid: Can't wait to see oscilloscope graphs comparing 220, 330 and 470 polymer tantalum smd's to multilayer ceramics
Lol! Funniest thing ive read in a while!
Those are aluminium polymer SMD capacitors, not tantalum...
@@marceldiezasch6192 I am suitably ashamed
Omg so true.
For me.
Before buildzoid.
RTX 3080FE looks gorgeous and is stupid fast.
After buildzoid:
I wonder if its possible to deliberately fluxuate voltage using a carefully designed chip to guess and match blips due workload spikes an attempt to smooth out power delivery.
lmfaooo
SOME ONE GET THIS MAN A GRAPHICS CARD
alot of channel did...
EVGA:
"During our mass production QC testing we discovered a full 6 POSCAPs solution cannot pass the real world applications testing. It took almost a week of R&D effort to find the cause and reduce the POSCAPs to 4 and add 20 MLCC caps prior to shipping production boards, this is why the EVGA GeForce RTX 3080 FTW3 series was delayed at launch. There were no 6 POSCAP production EVGA GeForce RTX 3080 FTW3 boards shipped."
COLORFUL:
The manufacturer was the first to report on a issue unofficially to the press. Review samples that were already sent out, were later recalled.
ASUS
ASUS has changed the design in pre-production. No official statement was made.
MSI
MSI acknowledged the issue during the MSI Insider Livestream, suggested it could be a driver issue.
ZOTAC
Igor’sLAB reports: PC Partner, Zotac’s mother company, seems to have recognized this and is obviously changing its cards.
EVGA has also confirmed it being an issue.
First blackballing gamers nexus for thorough reviews that instead of responding with launching a Rev2 board like an adult, now this. MSI effing up lately.
@@Wahinies drivers could cause some issues.
@@vyor8837 Yeah, you bet drivers will cause MSI some issues. Toyota, Honda, Ford, and Beemer driving PC Master Race will be cueing up round the block to run the fecks down as they exit the office!
So like... if its not a POSCAP why would EVGA even say it was and not correct it? blarg
We need to start a campaign:
Get LMG, JayzTwoCents, GN - anybody who has cards - to send Buildzoid a selection of cards to "stab with an oscilloscope".
"From my LN2 tank's heart, I stab at thee!"
Yes, watched Jayz first, just to see how he would try to spin this issue and found out he doesn't know what to call the capacitors, even after post editing!!!! GN work with Buildzoid, so that would seem the natural route. Don't know if you follow NAAF, but Paul posted fairly quickly, it could be a power supply (Main), proplem, not for the same reason as Buildzoid, but a question mark has hovered over these GPU's before any release? Above my pay grade, but why are they rated , and does a noisy power supply play a part, even if adequate in wattage?
Buidzoid needs a 3080, fast!!!
And let's crowfund a better oscilloscope for the guy, he deserves it ;-)
@@urishima I want him to say that if he does get his hands on them
Or... they could just learn how to stab the cards themselves!
Don't try taking a shot every time he says Poscaps. you will die
I'll one up with taking a shot every time swallowed saliva 😂😂 That mic is too sensitive
Me trying to buy a new GPU.
Nvidia: Do you know how a GPU works?
Me: No.
Nvidia: Do you wanna?
Reminds me of a scene from bad boys 2
“You ever made love to a man?”
“No.”
“You want to?”
it not really how it work it just lowest spec allow to be made by partner company...
we all know it will crash... how much this thing cost agin?
You forgot the rest
You: Yeah
Nvidia: go to the last crossroads of the darkest road you can find. Sell us your soul and all of your information, then you can play a game of dice to possibly win a blob of ones and zeros that might work.
You(&me): wtf, should've went AMD damn it.
In his (german) video, Igor explains that the big ones usually are either aluminium or tantalum, so i guess he's well aware of the different CAPs used, just didn't know (or didn't look up) the names for the different series
Still it's a good article just not proper research .I guess he wanted to publish it before everyone else got to it .
Still it's a good article just not proper research .I guess he wanted to publish it before everyone else got to it .
He is aware. But what BZ is not aware of is this: everybody in the industry calls them POSCAPS. It is commonly accepted to say POSCAPS. Engineers at the AiB even joke about them being the Piece-of-Shit CAPS. EVGA also refers to POSCAPS in their statement because of this.
@@tanmaypanadi1414 No, it has nothing to do with insufficient research on Igors part, many people just call them POSCAPS in a general way, even if they are not specifically refering to the Panasonic capacitor. There are many more brand names that over time turned into generic names for a specific item, such as kleenex, Jet ski, bubble wrap, Tupperware etc etc. These are all brand specific copyrighted terms people today use to descibe a specific item. Imagine if I gave you a lecture everytime you descibed plastic dishes as "Tupper ware"... that's essentially what AHOC is doing. AHOC is wasting several minutes nitpicking on this non-issue, this unfortunatly wrongly leads some poeple to believe that he just made a major revelation and Igor made a massive mistake...
@@thomasschraubt7497 And so does Nvidia themselves call them POSCAPS, at least they did so in todays press release....
YES,
first thought i had when i heard about his was :
"I need BZ thoughts on this"
Ikr. Posted the question like 8h ago in the last video, as soon I saw jayz video about it.
I was like: buildzoid or never happened
Same.
Lorenzo Zanotti
Also... same. Lol
Same, as soon as capacitors where the topic, I knew we needed him lol
F I R S T
Lol 🤣
No I'm last
At least you weren't the first to buy a card.
S E C O N D :3
Jesus... that was fast. lol
There's a funny quote from Igor's article where he explains why engineers tend to call all of those capacitors POSCAPS even though they're not the ones specifically made by Panasonic. It's because the engineers like to call them piece-of-shit caps.
Sometimes it's a shame that informativ comments are less upvoted than jokes, quotes and memes.
Panasonic: “Why I need to take a blame for something that I didn’t do?”
lol. Poor Panasonic.
😄😄😄
Panasonic engineering:
Panasonic marketing: "But you HAVE heard of us."
Igor: It's all because cheap POSCAPs!
YT Reviewer: POSCAPs is a LEMON!
Customer: I don't want that POSCAPs on my card!
Panasonic: What the fuck is this?!
@@rusTORK This is a REALLY funny comment 🤣
Find someone that pays attention to you the way Buildzoid pays attention to capacitor names.
Beta testing is now paid for and done by the consumer. I am grateful for their sacrifice.
Rest in pepperoni beta testers
Not now, it’s been like that since Turing as we were all willing ready to pay 1500$ for beta access to a feature that still doesn’t work properly on Turing and barely works on Ampere.
@@RadarLeon you can return your card if you don't like. and honestly i hope early ampere buyers do to teach nvidia a lesson.
@Masterr Laster With new model yes there is always a risk. Beta testing is meant to catch most of these issues and make the product acceptable to use for an end consumer. Some bugs may remain after beta testing that may be generated from a much smaller probability and its harder to catch.
However this time the hype around 3080 was so much that people would bought it regardless whether any of the cards were reviewed or not. This is like iphone bending where a product that sold in massive numbers had a fatal design flaw that it would bend in real world usage.
The fact that such a major problem was caught this early and replicated easily by most people means beta testing was non existent for the non founder cards.
This is especially a problem since this cannot be solved with a firmware update, it's an actual hardware deficiency that cannot be fixed.
Why was the beta testing skipped ? Mostly to cater to the hype behind the cards themselves and get the product to the end user asap regardless.
Only those who went all the way in using the best components all around probably knew what they were doing or played it safe by over engineering.
@@solomonshv I don't have one friend does I normally wait for these irregularities to iron themselves out, good card otherwise sorta poorly planned/tested launch. But there have been worse much worse...
Incoming GAMERCAPS™ marketing in next gen and RDNA2
You are an AMD marketing genius!!!
🤣
The worst part is that I might actually see them do that...
@@Android-ng1wn RGB capacitors which flash colors based on the amount of voltage they're holding. Actual blinkenlights.
@@DanielWW2 i dont O.o
There fired most of the old marketing.
Thats why we dont have any of the amd rebelion crap.
And almost no leaks compared to Vega.
The new team isnt as cringy but the fortnite thing leaves a bad taste in my mouth.
And people roasted them enough for gamecache 😂
But If true im interessted in what they call the huge cache on rdna2.
@@SuperUltimateLP I know they fired them and I am happy with that. Still I am a bit hesitant because so far I have seen some questionable stuff again. Not nearly as stupid as Vega or whatever the hell Nvidia is doing right now, but I was also not a fan of the Fortnite cooler unveil, nor the twitter marketing teases. That is the type of stuff that can backfire massively. They better deliver because they are being quite cocky. Especially now that Nvidia is making such a mess for itself.
After watching Jayz video, this is so helpful. Buildzoid goes into so much more detail than Jay, and yet remains much clearer and understandable - a sign of a competent and good teacher!
Outside of watercooling I don't expect anything detailed from Jay, besides sponsor spots.
Jay is a clown, he's proof that production and budget are more important than having anything important or interesting to say, or AHOC would have 10 million subscribers, literally no one else will give you a detailed breakdown of a VRM from an EE perspective.
Buildzoid and GamersNexus for the data I want, the others are just for entertainment
I watch Jay and have for years but Jay really doesnt know much about the actual hardware.
@ZPinch Yup... and that's how you end up with 6 SPCAPS and 0 MLCC on a PCB because someone didn't really listen to the "sperg" and was just like "let's be cheap".
In extremely technical and complex fields, details matter... A LOT!
*buildzoid starts to nerd out about capacitor naming schemes*
OH YAS, THAT'S WHY I AM HERE.
"Worst smelling capacitor's ever" is my favourite Buildzoid quote
It's just so very specific. Like, the "what the hell happened here?!" level specific.
First time watch AHOC. 42 minutes of great capacitor discussion and comparing cards and capacitor configurations, and then a blazingly fast self-promo schpeel at the end that if you blink you'll miss!
LOVE IT! Thanks for a entertaining and educational discussion on the capacitor types and configurations! Very much enjoyed it!
Douglas L. Holeman check out the rest of his videos when you have time. He shares so much knowledge
One of us!
Humble request, it would be really interesting to see you cap mod one of the lower tier 3080s, like a ZOTAC Trinity non-OC. And, like, do some benching before and after with an unlimited power mod to hit 2,000mhz+
Related, it was really cool to hear that you helped the dude from Frame Chasers figure out how to shunt mod his 3080 :D
You do not need a mod to hit 2k. It will boost to 2k on its own stock as long as you have good cooling. That is why all these people are reporting this as an issue. It is not a manual OC. If you have a water block on any of the cards you will hit 2k easy with GPU boost.
derbauer has a video showing how to shunt mod a 3080 also if anyone wants to do it.
Btw, I am not shitting on your post, just thinking he could very easily hit much higher than 2k with mods. 😁 👍
"My god this video got long"
- Buildzoid, everytime
We're half way to an electrical engineering diploma at this point 😂
Me listening to any Buildzoid video:
"Mhm. Yeah. Oh, yeah. I know some of these words!"
reference is 5 spcaps, the FE cards came with 4.
Nvidia knew.
Assuming Moore's Law is Dead is right, NVIDIA contractually obliged AIB partners to sell at cost or higher, while FE cards are sold at a loss (the MSRP). In this way NVIDIA can sell small amounts of FE cards at above minimum spec to give a misleading perception of Ampere perf per $ compared to the AIB partner cards that most people will be able to buy, which will either be technically inferior or more expensive.
Even Asus knew what's up.
Yeh but fe cards are crashing too
@@BruceLeedar MLiD is literally never right about anything. He claimed it was 'absolutely established' that the 30-series FE cooler would be tri-axial, for example.
@@BruceLeedar There's no misleading perception of Ampere Performance/$, there are partner models that cost the MSRP and perform the same as the FE, even if they might have less overkill parts, they still can be enough for what they are supposed to deliver, which is the stock performance. Also assuming that someone that missed most of the speculation they made until now is right about a speculation doesn't make much sense for me.
Thank you for pointing out about the poscap thing. I tried to comment that on Jay's video, but I got buried.
Poor Panasonic.
*me at the start of the video* : "wtf is a poscap? A really fancy popsicle?"
*me after 10 minutes of buildzoid rambling about poscaps* : "that's OBVIOUSLY not a poscap, cmon, man! JEEEEEEZ, IGNORANCE ALL OVER THE TABLE! And that dude calls himself a JOURNALIST, UNACCEPTABLE"
and a real pro says: hey, it's not deal, poscap name is used by engineers as a general synonym for polymere-capacitor. So why talking about that much?
It only impress a bunch of noobs, that for sure!
I wish Buildzoid would make some real investigative work instead wasting time by judging about things by looking at pictures.
BZ: super in-depth technical explanation
Me: huh, those capacitors form a face...
33:00 yep you're totally right I work in the industry, smaller components not only slow down the machine but create a lot more machine stops/alarms, also more problematic for video rejection of defects and people don't realize how much downtime costs in such a complex industry, it's also more wear etc...
Yeah but how much do RMAs cost because you tried to save a couple of bucks in manufacturing? Not to mention the hit to your reputation which is your most valuable commodity
@@longjohn526 couple of bucks x thousands is more than a couple of bucks.
@@longjohn526 If nvidia sad that this cheaper config works and it turns out it dosnt its still nvidia at fault here.
I guess this has to do with samsungs inferior quality.
That some cards work with this cheaper desing and some dosnt.
I've worked in heavy industry a long time ago. The tolerance on truck engine components are very small and even that can cause mayhem.
I still don't comprehend how electronics can be manufacturerd like they are, so yeah makes total sense
8:13 is when he stops talking about poscaps and the differences between the type of caps on the card and the fact that Panasonic isnt the only capacitor maker.
Lol thanks
Imagine him at the exams getting Popcaps.
Hahaha. I was in the grocery store. Went in, found eggs, grabbed beer, and checked out before he finished. I was like, dude, i mean... cool but chill. 🤪
So, why would you want to skip that? To stay ignorant like half of the reviewers and content creators? lol
@@FcoEnriquePerez I knew about the difference between a poscap and SPcap already. Its for those of us that get that they arent the same thing, and a large part of the info up until that point is redundant.
I haven’t seen Asus being this competent with their TUF cards compared to their competition in lord knows how long, I was really impressed with their cooler design and the PCB and then I thought Gigabyte and MSI would have better models but I haven’t seen any better models yet.
I feel like I dodged a bullet my TUF 3080 is arriving Monday and I woke up to this news, hopefully there's no isses with it but so far it seems like it's ok.
Probably because they fucked up the EVO 5700s and had to make a revision 2.0
Hoping the gigabyte aorus master isn’t affected
@@ezzahhh Hopefully you did. update us on Monday.
WayStedYou and that’s why I said they were incompetent for a long time, 5700XT TUF’s were really fucked up especially not having any memory cooling but 2.0s are not bad. I didn’t like them at first but you just can’t deny the numbers.
I picked up my puppy today . I named him ' Poscap' lol
😂
Get a cat and name her Mosfet
omg, this is pure gold. poscap the dog and mosfet the cat xDDD.
Those are actually cool ideas
You will get sued.
I understood basically nothing in this. but damn if buildzoid tech gablygook isn't enough to calm someone down.
the doubletime at the end was impressive BZ!
Suddenly waiting for those TUF and Strix cards seems like a good idea
Or waiting for rx6000
Or the ftw3
Yes...for $300 more.
EVGA admitted that the FTWIII launch was delayed because of this capacitor bs. Pretty obvious then that ASUS Strix was delayed for the same reason. It's 100% confirmed because if you look at the launch day photos of the Strix you can see it also had those big caps on the back, where as now they're using six arrays of MLCC's. ASUS found out and had to delay their launch so they had time to fix the card.
It's not just that people with tuf cards have also been having issues
LOL the amount of time BZ says "like" would put a Valley Girl to shame
Maybe not as much as "the thing is..."
Lmao thanks for ruining all future BZ videos, Im hearing it every time now. Never noticed it before
Buildzoid is actually a Valley girl with an advance voice and face and body changer.
Buildzoid is the deepfake master.
I personally love his, "As far as I'm concerned.."
@kulayeb godammit AHOC community ruining my life, how can I unhear this now
AHOC official AMPERE CAP modding service will have a lot of customers ;p
Buildzoid, you mentioned you wanted to move to a bigger flat on the last video. I think you just found an extra source of income :)
@@Android-ng1wn No, you can't put bypass caps on a remote board... They have to be as physically close to the GPU pin as possible.
@@Android-ng1wn If I was the unfortunate owner of one of these cards, the one thing you may be able to do to improve your situation, if you have balls of steel and know what you are doing, is to stack two X7R ceramic caps sideways on top of the tantalum caps, of the same value as the normal ceramic caps in the reference design. It's a low risk maneuver, IF you have decent soldering skills and a decently fine soldering pen and good tweezers. Just stack more caps on top of the ones already there. However if you do this heat may become an issue for the tantalum cap, and tantalum caps tend to explode when they overheat. So do this at your own risk. The better option would be to remove the tantalum caps from the board, and replace them all with X7R ceramics, but I don't think anyone will even accept your money to attempt that... Boards like this are not made to be soldered to with an iron, it's all done with wave soldering at the factory...
The bulk of the capacitance usually is with much larger bulk caps near the VRMs. Bypass caps are there to filter noise from the GPU.
03:00 The moment you realize that Buildzoid actually has a massive pet peeve! xD
Buildzoid also went full Eminem at the end! :D
More like dr. Seuss🤣
Angry buildzoid is awesome.
Man thanks a lot it's great to hear your opinion about this subject
EVGA have spoken about this, apparently they caught it when stress testing the cards and have gone to replacing two of the caps with MCC array's.
do you have a source?
Sounds good at least they are replacing them and not just sending out a crappy bios update to cap the frequency.
@@heikoguckes5693 forums.evga.com/Message-about-EVGA-GeForce-RTX-3080-POSCAPs-m3095238.aspx its hard to search forums i guess
zotac is dead silence.. i have zotac 3080.🙁
@@RngmonsterX nice find... Atleast if you pre orderd a card you can rest at ease. Actually im glad this was found early.. Saves alot of hassle in the future for buyers.
Newegg's ASUS TUF and Strix photos show 6 polycap designs, same as EVGA's. Both changed the designs before shipping to consumers.
kougar7 Yep They did run problems and did last minute change.
I wonder if They go for mixed configuration when They found out how Many ceramics is needed and how Many polyvaps They can actually use.
Most likely tuf will get bigger cut and strix smaller because strix is clocked higher... we have to wait and see. The asus did not chose ar random to use those cheaper first. Making the change did cut their margins and They want to get those margins as fat as possible...
FINALLY, someone with the expertise did the talking.
Except no. Bypass capacitors have nothing to do with storing energy. This video clearly wasn't made by an electrical engineer. I know because I happen to be one.
There is a video by Igor, who definitely also has the expertise, however he only makes them in german.
@Zzz A/ you didn't look at all my videos. B/ I don't have much interest in making videos about my job. You're making the mistake of trusting the guy who talks the most. Volume isn't a sign of expertise. Damn he spends the first 15 minutes ranting about capacitors and some curves he doesn't seem to understand. He throws words around like "impedance" but he doesn't know what they mean, which is obvious because he never links it to frequency even though that's all this is about. It's like a guy talking about cars engines but never mentioning RPM's.
@@TheNefastor Okay okay, take it easy. Let's say I believe you, care to elaborate what happened to the rtx 3080/3090 and with their capacitors behind the gpu core in language?
@@limingtogunawan4756 I'll try to keep it simple, even though it isn't. Let's start with describing the cause of the problem. Basically, any chip that uses power will inject interference on the power supply that feeds it. This interference is essentially crosstalk from internal high-frequency signals to power lines, as well as nearly unmeasurable voltage spikes resulting from the chip's operation (transistors, when they switch, will draw and release power because of their "parasitic" gate capacitance and it adds up when you have billions on a chip). It's a very complex phenomenon that's impossible to accurately model because it depends mostly on what the GPU is running at any given time. Video encoding and 3D gaming might generate entirely different interference.
Anyway, this interference, if left unchecked, would then travel through the board to other chips powered from the same source and interfere with them too. At some point, all that mutual interference will be so much that some or all of the chips on the board won't be able to function (basically, won't be able to tell a 0 from a 1). To prevent that, you need low-pass filters that will shunt high-frequency voltage (noise) to ground, leaving only "clean" DC voltage. That's what those capacitors are for, and that's why they need to be so close to each chip : the traces from the chip to those capacitors are effectively antennas that will radiate the interference and can also receive interference. This is an RF thing : you're not just worrying about noise on the copper, but also on the air waves.
So. How do you deal with this ? Like I said, you need to use capacitors to the ground that will act as low-pass filter, meaning they'll take high frequency spikes and shunt them to the ground, thus eliminating them. And like any filter, they need to be calculated for a specific *central frequency.* In practice it's more complicated than that because you'll have to deal with several different frequencies (i.e. the RAM controllers run on a different clock than the CUDA cores, etc...) but for the sake of simplicity let's just say you want to bypass interference at the GPU's core clock frequency. If that frequency is well-known, as is the case with a GPU, then you can calculate what's called the "decoupling network", a set of capacitors of different values and characteristics that will filter the frequency you don't want, and maybe its larger harmonics too. So you implement that. And lo and behold, your board works as it should.
Then comes along an overclocker who takes a board meant to run at, say, 1.5 GHz and pushes it to 2 GHz. Uh oh... suddenly your decoupling network is *no longer fitting* the frequency your biggest chip is now running at. So the filter becomes less efficient, and more so the higher the frequency gets.
To make matters worse, overclocking will cause the GPU to draw a lot more power... and that means your interference gets *amplified* by that much. If you're using 50% more power to reach an overclock, then your interference is also 50% stronger.
You end-up in a situation with stronger interference and a filter that's no longer tuned for it. If I need to spell it out for you, that means you're screwed : you've added a turbo to your engine and now it uses more fuel than ever and runs faster, but you haven't changed your exhaust. Either the engine ends-up choking itself, or you're going to blow-up your exhaust. Either way, you're not getting a fast car, just a dangerous one.
Graphics cards are designed and sold to run at a specific clock speed. The chips might be capable of more, but in the end you'll be limited by all sorts of peripheral parameters that you can't change, such as those decoupling capacitors. Unless you're crazy enough to calculate a new decoupling solution for your GPU, desolder the one it shipped with and solder in the new ones. You'd also need to do that for the RAM, by the way. That's just not a viable proposition.
Now, what I've said so far is pure expertise, any digital electronics engineer will confirm. What I'm saying next is only speculation because I don't work for a GPU manufacturer :
When you see a brand offer a board that says specifically "overclocked", "OC" or something similar on the package, my *assumption* is that they built it with different values for the decoupling capacitors. See, there are enough capacitors on each chip you can actually "widen" the frequency band that can be decoupled. A smart manufacturer could design a board with decoupling that's, say, 75% efficient over a really wide frequency band instead of 100% efficient over a "manufacturer-approved" frequency band. That means your "OC" board should remain stable at higher clock frequency than a non-OC boards, all other things being equal.
Damn, that went on for longer than I expected. But this just tells you how complicated this topic is. That's 15 minutes of my life I'll never get back so please read it and try to understand. I'm still game if you have questions afterwards.
Your one of the few pc tech youtubers that actually knows anything about electronics.
As soon as I saw this happening, and after Jayz2cents did a video on it, all I could think about was how it was such a perfect subject for you to cover. Definitely right up your alley.
Even commented on the video mentioning it. Looks like it could potentially be modded manually by those more tecnically savvy and comfortable removing and soldering on the proper MLCC configurations.
So the real question is, is it worth removing the SP caps and modding in the MLCC arrays? As long as the pads are present on the particular card, obviously.
Edit - Also, it's, "vast majority of them AREN'T", not "isn't". Sorry for the grammar N*-ish, can't help myself.
As a EE that designs electronics I agree with your statement on design guidelines!
"If its not stock, don't expect it to be stable" sure, but i expect the asus oc version to be stable, because i pay for it in that state... and that goes for any sold oc version.
Ye.
But not a single card is advertising 2.0ghz speeds, even if they can boost to these speeds.
@@PlayfullWithFire they do it automatically.
Seems that the regular tuf is all ceramic too.
Big thanks for actually explaining why exactly these are there and what exactly they do in the first place instead of just talking about what type is better. These extra bits of electronics knowledge may make for 40 minute videos where others make 4 minute ones, but unlike the short ones, this is not just informative, but educational.
Really makes your videos stand out in the avalanche of tech channels talking about topics like this.
Yes, to a degree it is indeed Nvidia's issue that they put forth such loose specifications. Also the board partners problem due to not stress testing their product.
In terms of cost, using a smattering of MLCCs and a bunch of larger polymer caps is generally a better over all solution. Since then we can get the benefit of high frequency responsiveness, and also cheap bulk capacitance. Would also have liked seeing more of the cards have some more bulk capacitance at the VRM itself. (Some after all have practically non.... Could we at least get 25% of the pads populated?)
And yes, poking a card with a scope should be fairly indicative of the voltage regulation. Would do it myself if I had a card at hand, but I don't.
Though, I am surprised that they didn't put in low ESL capacitors into the design. (looks like regular ceramic caps, but the connection is on the long edges of the cap, instead of the short ones.)
They can after all be much bigger, both in terms of capacitance and size, while having phenomenally low ESL. (Though, I guess they are more expensive due to being niche)
And your point about a poorly filtered power supply potentially lurking through the VRM is indeed a problem that can crop in. And why one generally should have some input capacitance on the card itself. Doesn't have to be much for it to be effective.
I only read your first paragraph but the board partners really didn’t get drivers for the longest time. They weren’t able to test cards before launch.
@@bustex1 The board partners did have the firmware for the cards before they shipped them out to the reviewers. THereby the board partners could do som burn in testing of their own, considering that the reviewers had the cards for a couple of days.
It wouldn't be unreasonable to "ask" nVidia for at least a 48 hour burn in period before sending something out to said reviewers, if this is the case or not is a different question.
But if nVidia doesn't want to do that, then frankly, nVidia is the only one to blame for enforcing poor engineering practices.
And to my knowledge, at least one board partner did ask reviewers to hold off on the review due to having found problems with the card. A very admirable way to handle the situation to say the least. (Since one shouldn't ship a non working design.)
Today on the Bench yes one board partner did say they did have an issue. I saw jays two cents video he said they didn’t have drivers and that without drivers they wouldn’t have known about the crashes. Just what he said I am just restating. Idk enough about this if firmware would have caught a software crash? Yea they could power it on and test that it powers on but idk if that has anything to do with the crashes that are happening. I don’t believe it does but like I said I’m no expert just restating what someone more knowledgeable than me said.
@@bustex1 As an engineer myself, I wouldn't want something I design get pushed to market without the design having been stress tested first.
If nVidia's board partners aren't willing to at least require enough time with the product before launch to fix potential issues, then nVidia isn't doing much else than forcing their board partners to rush their product to market. While nVidia retains the upper hand in having designed the GPU in the first place.
Though, one can also see it as an "early adopter fee", but that isn't how one should look at poor design practices.
This video had me looking at SMD cap pdfs. Well done. But yes, it is asking basics if you just peek a little bit behind the curtain.
The problem is not overclocking, the problem is the card self-overclocking with GPU Boost even with the cheaper caps.
That's true, but apparently overclocking it makes the self boost go higher from what I heard. If correct, then GPU boost is the final blow, but overclocking the cards makes it more likely to crash.
It does, but I believe the reports are coming in even for the cards without manual OC, as in only with the OC the GPU Boost does. Cause GPU Boost doesn't know the config of the caps, just knows "Hey, I have a power and temperature headroom, let's go higher", only to then crash because the power is there, but not in such rapid burst.
you can under volt them...
@@campkira yes, but that's something you shouldn't have to do on a brand new GPU.
@@polabecak undervolting should be a comfort measure, not the god damn reanimation.
From what I've seen most cards can boost to about 2GHz for a short time. Enough to get unstable and crash the 3d program. Especially in light-ish games.
The strix 3090 review sample had all sp caps but the 3080 that der8auer took apart the other day had all ceramics
Tres Waters sound a lot last minute change because of problems... intereresting to see if They reduce ceramics when They found out how much is really needed...
BZ - ppl are crashing on stock GPU speeds. GFX has its own boost integrated that is pushing from 1710 mhz to 19xx mhz and that causes crashes
Came from Jayz 2 Cents Special Education edition of this news. Thankful for Buildzoid who immediately invalidates knowledge I had a tenuous grasp of.
Same.
JayzTwoBraincells
Well there's a solid knowledge gap between an A+ certified IT guy and electrical engenier
Came here from, igors video from Jay's video.. Now I'm stuck in the matrix. Incredible info sir! 🙏🏼
I dont understand how can someone have this much knowledge about something 99% of pc community know nothing about :D thats crazy, i love it
Well yeah, it would be nice if more of the people covering it would have this kind of knowledge. It's not like it's impossible for anyone to acquire.
@@sluxi seems pretty impossible for me :D
Watching Jay's description of Bypassing and noise filtering was so cringeworthy...
The two types have different characteristics and do filter out noise in different ways , Just adding more isn't going to make anything better nor will just using one type but a combination of the two will have a huge difference when designed correctly.
HecTech FPV Yes... that is the Main point!
"I'm not gonna read an article, it's a bit long", goes on to make a 45min video :D
Thanks for getting on this Buildzoid and I have to say young man your knowledge has really expanded as you continue to test, challenge yourself and learn, extremely informative, very well done...
It's NVidia's own drivers and the built in boost technology that all their cards have.
Even though cards are listed or even say on the box in some cases that the max boost clock it 1710ghz or whatever the claim is for any given card.
Nividia's built in Boost, sorry not sure what they call it but basically all their GPUs do this, in some cases up to 2k or more...
so black big square = bad
small yellow square = good
hahahaha funny but true
Joey's terms
you need like 10x yellow/gray/brown rectangles NOT squares. Flip note: mlcc do crack and they fail mode tends to be short - i.e. your card is dead.
Avtually the best would be right mix of those... but yeah...
it`s reassuring when Buildzoid gives opinion and confirmation
IgorsLab has such a great content
izidor it does!
Came here expecting a BZ rant about capacitors. Did not disappoint A++
Buildzoid: none of the cards have POSCAPS !!!
Zotac (left next to lower left gpu mounting hole): has a POSCAP
...
To be fair he was focusing on the gpu area ... still I chuckled.
The problem is the gpu boost automaticly oc the gpu clocks to crash level on the cheaper configs. So you can get random issues from the gpu at stock settings which didn't happen before
When CIA Langley was being built, the HVAC contractor had a problem: CIA wouldn't tell them exactly how many people or other heat-sources were in the building...classified.
When NVIDIA creates their monsters, they don't tell the PCB contractors peak power and how clean it needs to be...classified.
Nvidia did tell the AIB partners to at least use 1 MLCC and they change the Reference Design too but it's up to the partners if they want to change...
I'm not condoning Nvidia's malpractice with some of their board partners but man if all those leaks a few months ago were an indication I'd be weary and careful with my product designs too. We almost got a complete picture of Ampere even before the architecture released and that's a major contributing factor to the over-hyping of Ampere...all those leaky AIB employees spoiling the launch.
@@Gandalf721 Board partners didn't have enough time to test b/c Nvidia didn't give them 3000 series beta drivers until very late dummy. Nvidia treat their partners like shit and kept everything a secret, board partners had to design their board blind without any way to test it until it was too late.
The problem is, when you pay a premium to get a factory OC variant, it should work out of the box. If you pay that premium and then you have a problem which solution is "well downclock the core offset for lets say -50Mhz, while its not a lot, its not what you paid for. Now NVidia GPU Boost auto-OC the chip, so AIB have to make sure that that auto-OC will not trigger any instability. I mean the price we pay, some of that price goes into the testing and certification of those board. Seems like somethings not right here.
I am excited to watch this one. I too thought "I can't wait to here from Buildzoid on this." To bad you don't have a few to play with for testing yet.
i would really like to see the voltage fluctuations with an oscilloscope
I am very surprised he didn't go live for this...lol
Poor quality PSU's being pushed hard does make sense as being a significant part of this issue. With the OC and less expensive capacitors pushing the GPU past stable OC. The graph spikes being nearly vertical point to this being a more complex issue. Cheers
ALT it is most likely combined problem. If you have highend psu it can compensate problems with capasitors, if you have better quality chip aka you did win in silicon lottery it can compensate not so good psu... using more ceramics can compensate both to certain level etc...
Actually Hardcore Overclocking : What Nvidia doesn't tell the AIB partners is the complete detailed specs sheet for the PCB circuitry due o it being a business secret so if Nvidia would give detailed specs as of why certain electrical components have to be that way, the the AIB-s wouldn't take shortcuts on that area. AIB-s earn very little today on Nvidia cards because Nvidia is selling the reference design to them very very expensively as if the AIB parters were end consumers so that AIB partners are taking little shortcuts anywhere possible is only natural. But the problems like catching fire VRM-s (because Nvidia said to AIB partners that they are so efficient that they don't need extra cooling) on the EVGA GTX 1070-s, 1080-s and 1080 Ti-s as well as overheating and artifacting GDDR6 that Nvidia again knew and intentionally forgot to mention specifically "hey AIB guys, be careful on the memory, it heats up like crazy, so construct finstacks for them, because they are made on the same node as GDDR5 but only have 50 % higher clocks". Those are just recent hardware flaws that occur due to AIB-s not getting the complete "why" detailed specifications for the PCB design from Nvidia and literally having only days to do too small number of a testing on their custom designs before Nvidia expects them to launch. This is completely Nvidias fault because Nvidia has designed one thing very good and that is almost impossible circumstances for them to have at least 3-4 months to design and properly test their PCB designs.
One issue MLCCs have is loss of capacitance with DC bias. This reduces the amount of energy they can store. The 47uF caps you looked at lose 40% at 2V. This also has the effect of introducing nasty harmonics into the supply rail. The most optimal configuration would be a mix of electrolytic/polymer and mlcc. Large electrolytic/polymer caps to store energy and supply the IC during a load transient until the smps catches up, and mlccs to bypass the high frequency component of the load transient so that the voltage excursion is smooth and kept to a minimum. It's the ESL that holds the polymer caps back, so really the ceramics only need to supply power until the current through the polymer caps ramps up to match the load. The FE probably has the most optimal configuration. You see this sort of thing on power stages. Mlccs placed by the switches to decouple the switching energy and electrolytics that decouple the load current.
If the bill of materials showing the different options is correct, then its the aib's that screwed up cheaping out. Nvidia clearly shows the top middle being mlcc's and the rest having the option for either type.
Nope, like he said in the video Nvidia has final approval on the design and specs of the cards. Everything points to nvidia being the culprit
And all this capacitor stuff is so insignificant compared to VRM tuning at DC-DC controller that every vendor doing differently ;)
In your MLCC plot talk - you also missing that MLCCs have much much higher impedance at lower frequencies, compared to polymer caps... Job as DC-DC engineer is to make impedance and ESR plots as FLAT and as low as possible, not just one huge notch at particular frequency, which is bad too. That is why you always see mix of huge polymer cap, smaller polymer cap and array of 0805, 0603, 0402 and sometimes 0201 MLCC everywhere, instead of just one type.
loved the 8 minute intro-rant on the poscap naming
Thank you for shedding light on this and providing really useful information. Thanks for the knowledge transfer :thumbsup:
When biuldzoid mentioned the smell of those caps I laughed out loud.
Those things smell like a curse when we replace them with hot air.
Good video but isn't GPU Boost an Nvidia feature? So even the "stock" AIB cards could have this issue without a user actually overclocking?
I think Nvidia had really low margins for the AIBs even compared to the last gen and therefore they kept the guidelines loose and AIBs just went a little to cheap on the cards.
They had to overspend on the heatsinks
Thanks, your input is very valuable here!
when Buildzoid says: "I'm not gonna read it, its a bit long!" LUL
I am never in a hurry to upgrade and after watching this BZ video, I learn lots about capacitors and feel better I AM never in a hurry...
Finally, someone with actual knowledge not some randrom screamy youtubers.
Ngl, a couple minute rant about poscaps was just what I needed today
I miss Taiyo Yuden CD-Rs.
Been waiting for your take on this one
Perhaps the suspicions the 3080 was rushed to market to preempt whatever AMD has coming might be somewhere near target ?
Probably.
All the silicon companies know what the business next door is up to. If they didn't, they would die in a year.
Rumor a year or so ago was AMD sampling silicon for the consoles, results were "better than expected", which caused Nvidia to panic, pull in the Titan launch and so on.
My Speculation: The fact TSMC just shrugged at Nvidias "we''ll have Samsung make it" is another corner stone of this. TSMC has the better process, Samsung is still on one of their fuckery-nodes, but Samsung has huge capacity.
@@MazeFrame Yeah I feel there might be California/Silicon-Valley politics tied up in this somehow too, with TSMC looking to establish a new fab in the US to possibly escape mainland Chinese hostile takeover, and California having sold out largely to the CCP, but who knows. I'm not sure at this point whether going with Samsung's 'loss leader' 8nm process was the 'right' choice, but at least we don't have the problem of too few open fabs. Would help imo if Intel would open up more, since their 14nm (+++...) is still fantastic
@@JasonOfTGA What?
Like, dafuq?
@@MazeFrame Oh you didn't know about TSMC wanting to build a fab in the US ? This is the best timeline dude.
@@JasonOfTGA Interesting. I mean, with random tarifs dropping on random products, might be the best move.
Cheers BZ, in-depth stuff like this is exactly why I watch your videos.
Imagine selling expensive graphic cards but you have not stress tested and played games on them to check if they are stable.
Sucks if consumers has to install software and undervolt their brand new gpu to avoid crashes.
That's Nvidias fault though. Board partners couldn't test the cards before launch. Literally the only thing they were able to do, was to run an application from Nvidia that produces a computational load and then tells them if the test completed or not.
I was just about to print this polaroid with my xerox of this delicious Kraft n Cheese I made before putting it in a ziplock to go in the fridge, but all I had was a tupperware container and I didn't want to mark that with a sharpie.
Good point. He's wasting way too much time on this.
I smell fear and despair for what coming in October.
😂
You fanboys are delusional.
Rtx3070 and rx6000 series will be interesting for sure
@@Sal3600 Well, you think? At least I do not have to pay 2000 dollars to play at 100 fps...
@@moldoveanu8 It might not even been better, but no one can deny that AMD had a huge impact in terms of technology development.
I hope AMD can go past offering slightly better value for the mid range and offer something new in terms of performance (e.g flat out faster than 3080) . Not expecting much though
I think the big reason why this wasn't caught in QA is a lot of AIBs have been saying they got drivers to test against on release day basically. Actually it's possible some parts of the press had drivers before the AIBs where given them. So yeah at least part of this issue is clearly Nvidia's fault for not giving the AIB's the support they needed.
Can't wait to see all the marketing from AIBs after they come out with design revisions that fix this issue.
"Super Gaming Capacitor Technology!"
"Gaming CapX!"
"GPU Reactor Cap!"
"XCapOC!"
"I'm givin' her all she's got CAPtain!"
Bust a cap in yo GPU
Aorus RGB Caps
Hey Buildzoid, I wanted to hear your take on pricing. I was surprised that everyone is saying that ceramic capacitors are more expensive. I don't know the cost the board partners pay for their orders of 100K capacitors, but in qty 10k, the ceramics are the same price (or cheaper)
Panasonic SP-CAP EEF-SX0D471ER $0.57 (cheapest SP-cap I could find)
Illinois / CDE XMPL series (these are on the FE) 477XMPL002MG19A $0.71
Samsung CL21A476MQYNNNG (cheapest 47 uF 0805 MLCC I could find, they are 6.8V) are $0.059
So 10 MLCCs is $0.59, SP-caps are $0.57 and the Illinois / CDE XMPL are $0.71.
By the way, I suspect the yellow color is from the dielectric X5R and X7R are about that color. They have a high capacitance per volume compared with other dielectrics, so you see them on the larger capacity ones. Some of the more critical pins will be bypassed with NP0 /C0G dielectric capacitors which have much better properties, but less capacitance. They are usually more grey.
So is it possible to make a usb killer stink bomb out of poscaps?
I would give this an award on reddit.
This is not non-video as you talked about the capacitors not being poscaps and then said you do not have enough info. Phenomenal job as always. I am eager to see a video of the resolution; if that ever happens... Unlikely these AIBs send you models now.
this wont get nearly the views it should.. so much fudd out there RN
Others will pick up from here. This is not over
@@rosen9425 still alot of the riff raff about this situation is being spewed out of total ignorance. And it would help so much if half of the people trying to get these cards will have seen Buildzoids video here.
@@skoopsro7656
I think people who have issues with their cards will be kicking up some major whine about it, and RMA when they find these videos out soon enough
YESSS! was waiting for you to weigh in on this! in Buildzoid we trust!
Has anyone bothered to point out to Buildzoid that the 2ghz "overclocking" is from running the cards at stock? It's the card's out-of-the-box boost behaviour that's causing crashing, this isn't users manually OCing.
Isn't the PCB layout under the caps the same for all the cards? Owning a hot-air station, flux and replacement caps it should be possible to upgrade the card, right?
BZ: "i have no interest in rtx 3080"
Nvidea: "we might change that, wait a sec let us quickly reload and shot the other foot as well"
HI Buildzoid watched you video on the asus strx 2080 tii flashed the xoc bios on my silent bios and it did flash but the card peforms very poorly and gpu usage is waaay down in benchmars also vf curve in MSI AB is gone any idea whats wrong??? i used a xoc bios on my 1080ti whitout any issues.