You're correct, tantalums fail short when they fail. They're internally just a tantalum metal slug sourrounded by manganese dioxide and a metal leadframe cathode. The isolating oxide layer thickness is set by the dielectric strength of the applied electric field during manufacturing, which is just their nominal voltage rating plus some tolerance. If you over-volt a tantalum, the oxide layer grows even larger to meet the dielectric strength of the electric field, which ruins their capacitance. Eventually there's not enough material left and the field can just bridge the gap and it'll fail short. The main difference these days is that tant-poly uses a formulation that absorbs oxygen during breakdown, whereas the old tantalum (non-polymer) caps produced oxygen during breakdown, which is why they'd go off like rockets when they failed. Modern tant-poly will burn up a bit but won't fail with that firework-like flame of the old ones.
this reminds me of the NEC TOKIN caps in the PS3 and what people do to replace them. the TOKINs have very clean and stable power delivery at very high frequencies, and indeed at ALL frequencies, have extremely low ESR and have a 1000uf of capacitance. the PS3 used an array of 4 of them for the CPU and GPU since they had very strict power requirements and any variation on the capacitance at any frequency would cause computing errors. The problem is that the PS3 runs crazy hot and these degrade very fast at 85-90C. Modern replacements for a while was to use 24 of these tantalum capacitors, which worked for a short time, but would ultimately fail as well. Newer replacements also pair an array of ceramic 32 ceramic capacitors to smooth out the power delivery and improve longevity and performance at differeny frequencies.
I really love your long ramblings about VRM design and components! (And well, it actually is pretty interesting and useful to me since I somehow ended up designing and building switchmode power supplies as a hobby)
hey, buildzoid. i wanted to drop a note of thanks for helping me understand and solve the problems with my 14th gen processor gigabyte board problems. long story short the processor would be fine in everything execpt export video from adobe apps. which is the sole purpose i built the machine for. before i had to down clock it to get a video export, but after watching your ramblings and going thru settings, for me it was disabling voltage reduction tvb and it now works at on box clock speeds and exports vids. i am sad i have to accept the undervolt that gigabyte applies and cant tweak more but it is what it is. and i figured this out with your help, thank you.
Another infamous application of the proadlizers was the fat PS3 and its early slim models! Early 90nm cell cpu and rsx gpu ran so hot, many ps3s got the yellow light of death because of this.
@@LedNe0nDevil which broke because of constant heating and cooling cycles off of CPU's and GPU's where BGA connection would fail and you would get YLOD and it was not the lack of tin, it was lack of lead in the solder because leaded solder has a way higher melting temp compared to what was used to solder those CPU's and GPU's because we used leaded solder just fine till fiasco around those times regarding leaded solder and environmental risks around it
@@LedNe0nDevil not quite. Failing underfill on the RSX was (and many other Nvidia GPUs of the time). But proadlizers definitely DID fail on the hotter-running models. Look up the list of errors the PS3's syscon chip can provide over serial: a few are specifically related to power instability of the cell and/or rsx (errors 1001 and 1002). replacing the capacitors usually fixes them.
@@coccoborg Not quite. While the NEC/Tokins can be a problem, they're not the problem. That's not to say caps don't go bad, they can, but the yellow light of death is still primarily a BGA & Bumps failure. Comparison: Rate of NEC/Tokins fail: 11% Rate of BGA/Bumps fail: 51% From a data pool of 264 consoles Syscon code Data. (of course it was the underfill, where else would solder be used on a die?) we all clearly saw the same video from RIP Felix. Just say that. BTW PS: NEC/Tokins are amazing. So. I'm here to defend them.
You'd definitely see the inductive effects of that carrier if we were talking about decoupling right behind the GPU package, but this far out into the power plane it's going to be negligible.
About the workstation/Fire Pro cards, I'd say that in all likelihood the choice of capacitors has more to do with reliability then performance. I don't design video cards, but I do use workstations/mobile workstation (laptops) at my day job. I've sent back leased laptops around 6-9 months early because it was cheaper to eat the remainder of the lease then the user downtime over a 2 week period. In enterprise/workstations the hardware are cheap compared to the payroll and licenses.
LGA775s came out in the tail end of P4 and through the Core and 2 era. They were displaced by the first Nehalem. So, yes, between 15 and 20 years old. The oldest ones are 20, the newer ones more like 15.
The root pcb layout shape looks reamrkably similar to last gen of NEC/Token proadlizers (custom casing TR2) which never went on mass production cause of HF parasitic ringing.
i'd guess SMD Caps being packaged in Plastic greatly impacts the ease of offering a wide range of wacky Colors. vs Metal packaging that you can pretty easily Plate/Anodize/ w/e them with whatever you want.
I see a hint of something cool for cryogenically cooled LNAs, however, it isn't there. that metal cover is the wrong kind of metal and too far from the GPU die to be an effective low-impedance bridge, and worse if put in the wrong spot, the riser-PCB becomes a resonant LC tank-circuit in parallel with the pawer plane. They do kind of do that with high-power RF grounding straps with a slightly longer wire in parallel with the grounding strap, however, the wire length is tuned to a very specific operating frequency explicitly for the equipment being grounded (there is no generic "detuning strap"), lol. The extra capacitors, I can almost like that if it wasn't for all the marketing gimmicks around them.
Okay so it sounds like independent repair shops should start offering painting services, removing caps and painting them with a non conductive nail polish or something. It might be niche but it would be cool as hell, Id love to have a kingpin card with all the other parts shiny gold too.
don't even need to take the caps off. Can just paint them in place. Honestly even a metallic sticker could work(MSI did that on the inductors of the Nvidia 20 series and RX 5000 series).
Buhhh 775 was a very very fun platform! OC was easy and if you had the right CPU you could push the clock quite high (compared to stock). I switched from 939 with an 4800X2 to 775 (with different CPUs over the next few years). Ahhh good old times. At leats a few times per week someone from our Clan or others came online to get help regarding OC settings. A really fun time when Intel CPUs had sooo much unused OC potential!
Agreed. It also got interesting at the tail end of the socket when secondhand Xeon X3xxx CPUs nosedived in price. They often had better silicon than the Q9xxx CPUs but with the same 12MB L2 cache that made them really good CPUs for gaming. When I finally retired LGA775 I was running a $95 Xeon X3360 @ 4.4GHz on water, and it was usually ~5-10% faster than an AMD FX 9590 in games, which I thought was both equally funny and sad.
Fyi, just so you know you actually do want higher series resistance in caps for signal filtering. The design rule is to create as much resistance as you can tolerate, not keep it down. It's obviously a balancing act but you end up in a situation where the required capacitance to keep impedance flat is rising exponentially as you increase power. The equation is for the required capacitance is Cout = Lexcess/R^2.
Maybe in the original design they were using a specialized component like NEC TOKIN. They probably did this to replace a problematic component with something more reliable. And then marketing comes to play.
I have seen shorted aluminum polymer caps of various shapes and sizes as a pro audio tech, and definitely lots of burned up tantalums. But in that world gear gets used for decades, graphics cards become obsolete way sooner.
@@ActuallyHardcoreOverclocking I might argue that they all probably should use the liquid caps, but I do find them from time to time. The tantalums got used more in the late 80s / early 90s in early surface mount stuff when the aluminum caps were bigger. Though you also find them used as coupling caps in some old gear for lower esr into sensitive circuits.
you can't buy new proadlizers and de-soldering the old ones without melting them is really really difficult. So I haven't tried using them. I'd suspect you're better of just using MLCCs instead.
Everytime you say bandcamp I'm reminded of American Pie.. It's your voice. 🤣 I thought someone already did a video about Zotac use of power boost. I don't get why they use an extra PCB though. Couldn't they just paint a square on their backplate?
No it is more to have something shiny and unique. It could maybe cost cut a little because they can use the reference design at the back of the GPU and mount the power boost further away, but that small saved redesign cost is probably offset by installing the power boost module itself. 4 caps close to the GPU would probably do the same as the 6 caps further away under the shiny metal lid.
BZ, the only thing that unnecessarily adds to the length of your videos is the time you invariably spend apologetically whingeing about the length of the video! 🤣🤣🤣Never change, son: we love you.
You're correct, tantalums fail short when they fail. They're internally just a tantalum metal slug sourrounded by manganese dioxide and a metal leadframe cathode. The isolating oxide layer thickness is set by the dielectric strength of the applied electric field during manufacturing, which is just their nominal voltage rating plus some tolerance. If you over-volt a tantalum, the oxide layer grows even larger to meet the dielectric strength of the electric field, which ruins their capacitance. Eventually there's not enough material left and the field can just bridge the gap and it'll fail short. The main difference these days is that tant-poly uses a formulation that absorbs oxygen during breakdown, whereas the old tantalum (non-polymer) caps produced oxygen during breakdown, which is why they'd go off like rockets when they failed. Modern tant-poly will burn up a bit but won't fail with that firework-like flame of the old ones.
power boost modules sounds like some recurring tesla monthly subscription service
coming next year (ten years from now, if ever - but really... never)
Intel still is sort of floating the idea of rentable units - subscription-unlockable CPU cores and features like BMW's heated seats.
I have the same 1070 ti. I knew it was just a bank of capacitors but I never knew the origin story of it. kinda cool and interesting.
this reminds me of the NEC TOKIN caps in the PS3 and what people do to replace them.
the TOKINs have very clean and stable power delivery at very high frequencies, and indeed at ALL frequencies, have extremely low ESR and have a 1000uf of capacitance.
the PS3 used an array of 4 of them for the CPU and GPU since they had very strict power requirements and any variation on the capacitance at any frequency would cause computing errors.
The problem is that the PS3 runs crazy hot and these degrade very fast at 85-90C.
Modern replacements for a while was to use 24 of these tantalum capacitors, which worked for a short time, but would ultimately fail as well.
Newer replacements also pair an array of ceramic 32 ceramic capacitors to smooth out the power delivery and improve longevity and performance at differeny frequencies.
That´s what i call a Gimmick.
I really love your long ramblings about VRM design and components! (And well, it actually is pretty interesting and useful to me since I somehow ended up designing and building switchmode power supplies as a hobby)
How hard is it to build a fanless psu for a 360?
I assumed it would just be a big package capacitor. Wasnt expecting it to just be a cover 😅
hey, buildzoid. i wanted to drop a note of thanks for helping me understand and solve the problems with my 14th gen processor gigabyte board problems. long story short the processor would be fine in everything execpt export video from adobe apps. which is the sole purpose i built the machine for. before i had to down clock it to get a video export, but after watching your ramblings and going thru settings, for me it was disabling voltage reduction tvb and it now works at on box clock speeds and exports vids. i am sad i have to accept the undervolt that gigabyte applies and cant tweak more but it is what it is. and i figured this out with your help, thank you.
Just like MSI used to do on the Lightnings with the addon pcb behind the core.
Another infamous application of the proadlizers was the fat PS3 and its early slim models! Early 90nm cell cpu and rsx gpu ran so hot, many ps3s got the yellow light of death because of this.
Common misconception, it was actually the dies using lead-free solder.
@@LedNe0nDevil which broke because of constant heating and cooling cycles off of CPU's and GPU's where BGA connection would fail and you would get YLOD
and it was not the lack of tin, it was lack of lead in the solder because leaded solder has a way higher melting temp compared to what was used to solder those CPU's and GPU's because we used leaded solder just fine till fiasco around those times regarding leaded solder and environmental risks around it
@@LedNe0nDevil not quite. Failing underfill on the RSX was (and many other Nvidia GPUs of the time). But proadlizers definitely DID fail on the hotter-running models. Look up the list of errors the PS3's syscon chip can provide over serial: a few are specifically related to power instability of the cell and/or rsx (errors 1001 and 1002). replacing the capacitors usually fixes them.
@@coccoborg Not quite. While the NEC/Tokins can be a problem, they're not the problem. That's not to say caps don't go bad, they can, but the yellow light of death is still primarily a BGA & Bumps failure.
Comparison: Rate of NEC/Tokins fail: 11% Rate of BGA/Bumps fail: 51%
From a data pool of 264 consoles Syscon code Data.
(of course it was the underfill, where else would solder be used on a die?)
we all clearly saw the same video from RIP Felix. Just say that.
BTW PS: NEC/Tokins are amazing. So. I'm here to defend them.
You'd definitely see the inductive effects of that carrier if we were talking about decoupling right behind the GPU package, but this far out into the power plane it's going to be negligible.
About the workstation/Fire Pro cards, I'd say that in all likelihood the choice of capacitors has more to do with reliability then performance.
I don't design video cards, but I do use workstations/mobile workstation (laptops) at my day job. I've sent back leased laptops around 6-9 months early because it was cheaper to eat the remainder of the lease then the user downtime over a 2 week period. In enterprise/workstations the hardware are cheap compared to the payroll and licenses.
LGA775s came out in the tail end of P4 and through the Core and 2 era. They were displaced by the first Nehalem.
So, yes, between 15 and 20 years old. The oldest ones are 20, the newer ones more like 15.
The root pcb layout shape looks reamrkably similar to last gen of NEC/Token proadlizers (custom casing TR2) which never went on mass production cause of HF parasitic ringing.
i'd guess SMD Caps being packaged in Plastic greatly impacts the ease of offering a wide range of wacky Colors.
vs Metal packaging that you can pretty easily Plate/Anodize/ w/e them with whatever you want.
I see a hint of something cool for cryogenically cooled LNAs, however, it isn't there. that metal cover is the wrong kind of metal and too far from the GPU die to be an effective low-impedance bridge, and worse if put in the wrong spot, the riser-PCB becomes a resonant LC tank-circuit in parallel with the pawer plane. They do kind of do that with high-power RF grounding straps with a slightly longer wire in parallel with the grounding strap, however, the wire length is tuned to a very specific operating frequency explicitly for the equipment being grounded (there is no generic "detuning strap"), lol. The extra capacitors, I can almost like that if it wasn't for all the marketing gimmicks around them.
Okay so it sounds like independent repair shops should start offering painting services, removing caps and painting them with a non conductive nail polish or something. It might be niche but it would be cool as hell, Id love to have a kingpin card with all the other parts shiny gold too.
don't even need to take the caps off. Can just paint them in place. Honestly even a metallic sticker could work(MSI did that on the inductors of the Nvidia 20 series and RX 5000 series).
@@ActuallyHardcoreOverclocking now that I think about it, its only a matter of time before companies are producing RGB board components.
@@celeriumlerium8266 Don't you go giving the fecks ideas now!😁
Buhhh 775 was a very very fun platform! OC was easy and if you had the right CPU you could push the clock quite high (compared to stock). I switched from 939 with an 4800X2 to 775 (with different CPUs over the next few years). Ahhh good old times. At leats a few times per week someone from our Clan or others came online to get help regarding OC settings. A really fun time when Intel CPUs had sooo much unused OC potential!
Agreed. It also got interesting at the tail end of the socket when secondhand Xeon X3xxx CPUs nosedived in price. They often had better silicon than the Q9xxx CPUs but with the same 12MB L2 cache that made them really good CPUs for gaming. When I finally retired LGA775 I was running a $95 Xeon X3360 @ 4.4GHz on water, and it was usually ~5-10% faster than an AMD FX 9590 in games, which I thought was both equally funny and sad.
Fyi, just so you know you actually do want higher series resistance in caps for signal filtering.
The design rule is to create as much resistance as you can tolerate, not keep it down. It's obviously a balancing act but you end up in a situation where the required capacitance to keep impedance flat is rising exponentially as you increase power.
The equation is for the required capacitance is Cout = Lexcess/R^2.
Expectation: its a Proadlizer
Reaction after watching the video: 😮🤯
Maybe in the original design they were using a specialized component like NEC TOKIN.
They probably did this to replace a problematic component with something more reliable.
And then marketing comes to play.
I figured it was extra capacitance, but figured it was one or two very large capacitor(s) like the with a possible benefit to having a heat-sink. Sad.
Love how they swapped out the ram on the RTX 4070 TI Super 256bit top ram modules and OC it and it performed better than the RTX 4080 Super.
I have seen shorted aluminum polymer caps of various shapes and sizes as a pro audio tech, and definitely lots of burned up tantalums. But in that world gear gets used for decades, graphics cards become obsolete way sooner.
audio gear uses solid aluminum polymers? AFAIK most of it still uses liquid electrolyte caps but I don't really do audio gear so IDK.
@@ActuallyHardcoreOverclocking I might argue that they all probably should use the liquid caps, but I do find them from time to time. The tantalums got used more in the late 80s / early 90s in early surface mount stuff when the aluminum caps were bigger. Though you also find them used as coupling caps in some old gear for lower esr into sensitive circuits.
Curious how much improvement modern cards would get from proadliser modding rather than regular cap modding
Have you tried proadliser modding before?
you can't buy new proadlizers and de-soldering the old ones without melting them is really really difficult. So I haven't tried using them. I'd suspect you're better of just using MLCCs instead.
It's 5 dollars worth of capacitors that we charge an extra 50 dollars for ......
Marketing in a nutshell.
Everytime you say bandcamp I'm reminded of American Pie.. It's your voice. 🤣 I thought someone already did a video about Zotac use of power boost. I don't get why they use an extra PCB though. Couldn't they just paint a square on their backplate?
how about MSI lightning, they also have a daughter card that can plug in behind their core.
So the Power Boost modules is a cost-cutting gimmick, instead of putting the caps directly on the PCB, closer to the GPU?
No it is more to have something shiny and unique.
It could maybe cost cut a little because they can use the reference design at the back of the GPU and mount the power boost further away, but that small saved redesign cost is probably offset by installing the power boost module itself.
4 caps close to the GPU would probably do the same as the 6 caps further away under the shiny metal lid.
better than I expected tbh
BZ, the only thing that unnecessarily adds to the length of your videos is the time you invariably spend apologetically whingeing about the length of the video! 🤣🤣🤣Never change, son: we love you.
this was enjoyable
Asus has also used the 'Power Boost' thingies in the past as well. My Asus GTX760 has one.
no thats different, thats a copper bridge. see the video "Measuring the effects of the ASUS Direct Power bridge"
th-cam.com/video/JDw1Ul-nBhc/w-d-xo.htmlsi=vFep9GaGujzd6biG
we are really going to need ULTRA POWER BOOST from zotac next gen otherwise it will be bad
illusion
Hey guys!
PCB lore
more gpu content
yes
The L is silent in “Soldering”
Sod-Er-Ing
Not “Soul-Der-Ing”
dictionary.cambridge.org/us/pronunciation/english/solder
English comes from England (crazy right?) The correct English pronunciation is English
The school is to "Teach kids".
Teach-kids.
Not "shoot-ing-ground".
Countries outside America? No such thing
@@greebj *S0DDERS UNIT3!*
G'day Buildzoid,
Yeah I found this Quite Interesting
Under 30 min it was a "SHORT" Buildzoid Rambling video 😁