Hey BZ just wanted to say i really appreciate you going over this topic in great detail. There are too many people on forums spouting about high voltages/temperatures but cannot explain why. Thank you
A slight correction on electromigration. Current only flows on the surface of the conductor, not inside. Electrons jump from one atom to the next, from one end of the conductor to the other, leaving "holes" behind, and the direction of the flow of the current is actually the opposite of the direction of flow of electrons, so, basically, the direction of the flow of "holes". Every conductor has a certain resistance, which is actually the resistance to the electrons jumping from one atom to another, and one can think of it as the amount of hold an atom has over an electron. Also, current always flows in the direction of least resistance, which is now even more obvious. We can observe electromigration as migration of electrons that deviates from a wanted path due to changes that happen in either lower or higher resistance in certain areas of a conductor. If there's more than wanted resistance, electrons will take a different path, if there's less resistance, more electrons will flow on the same path over the same time. Voltage, temperature, impurities in the conductor and many other factors (including electromagnetism) can determine whether the electrons will behave as we want them to or not. So, too much resistance in a wanted path or lower resistance in a not wanted path will be the symptom of electromigration. The rest of the explanation that involves current, temperature and other factors is correct. O, and one more thing, electrons have insanely low mass, so it's not a bump as described, but they lose energy on overcoming the resistance, and are no longer able to follow a path that has more resistance than electron has energy, so they take a path of lower resistance. [edit] one more note, greater the resistance - more heat, which causes even greater resistance.
Very good explanation. That is the reason why I'm undervoting/underpowering my 1080 Ti down the point it does perform 5-10% less than stock settings (but plenty enough for me). But use SO much less power. 80-100W less. Having a GTX 1080 Ti use 150-160W is really nice. 60c max on air with very low noise. And the knowledge of much less degradation. Also always undervolting my CPUs when doing stock speed. Undervolting is possible on most chips, because the stock volts are normally set high(to be stable with even the worst binned silicons). Undervolting also saves you money on a bigger cooler and on the electric bill. But yeah... undervolting plus a bigger cooler is the best for longevity. I have a retro rig with a QX6850 (12 years old) that at stock freq, can be undervoltet quite a bit. So must still be in in good nic ;)
@@karl9791 whats the point of keeping an old processor alive for so much time? Like, idk where u from, but in Brazil we pay at least 3 times more in eletronics and we still dont want to keep the same cpu for 10 years.
@@Shadower2014 no one want keep it for 10 years. For me, the ideal would be swap every launch announcement. But is what it is. My point about my cpu is simple, intel have a healthy long life. And I use it almost 24/7. So.. Just that. And I have another pc. New one. Waiting just for the rtx 3080. And still no plans for rework in the old one, maybe some change in the storage... Maybe. Idk.
@@MovementCultivation All electronics degrade. Just the variable how quickly.. And actually the 1080 Ti might degrade faster than a smaller GPU die (like a 1070) especially if loaded with many thermal cycles. Any big die GPU chip generally degrade faster than the smaller ones with many thermal cycles. But As long as you use it within Specs (stock settings) it should have no problem lasting for years (5-7 yrs) Keep thermals and voltage as low as possible, and you gain less degredation and longer lifespan. A way to check chip degradation, is to try running it low as possible voltage vs highest clockspeed as brand new, and then 3 years after, try it again. It might need a bit more voltage to run at the same clockspeed it could when new. Meaning its degraded. Might still be okay to run perfectly within specs, but degraded compared to when new. Memory is a more fragile thing with modern graphics cards. Nvidia cards do not have memory temp reading and I know GDDR6(X) run hot. The new cards give you speed.. but not lifespan. Nvidia has no intention on making these last for 5+ years. (Id be surpriced if more than half of the RTX 30xx cards still works fine in 6-7 years) Nvidia knows the market demands Performance now. Not long term endurance. Not overclocking the memory will be a good idea. Downclocking it a bit will give it a cooler longer life. Especially if its going to be put at hard work.
Room temperature can also be like 5C when you're the extra ghetto type of overclocker who just opens the window in the winter for extra low ambient temperatures since chilled water cooling or dry ice is too expensive.
@@PainterVierax Norway here. Managed to get the water temp in my loop to 7C by opening the window. My 1080 Ti was chillin at 18C under full load overclocked. It's pretty dank
Hi Buildzoid, I'm both a trained physicist and chemist. If you have questions or want me to make a video going into more detail on Oxide Breakdown and Electromigration just shoot me a reply.
@RectalDiscourse You feign giving advice for someone else to be taken seriously on the internet, while simultaneous and ironically making an example of yourself for how not to be taken seriously at all. There's no room in hell for all the psychological baggage you seem to have.
Thank you for sharing. I really appreciate what your sharing. Some of us don't know, wouldn't know, or even consider the things you are explaining. You help make people smarter about things. TY!
Very good vid, it gave me the confidence to try to beat the next gen big brother of my current CPU. I did 4.2 all core OC at 1.4V on a Ryzen 5 2600X with idle temp of 27.50c/load over time temp 71.50c using the Wraith Max box cooler for FX 8390 CPUs. This allowed me to beat the CPUZ single core score of the stock Ryzen 3600 and match it in multicore score, validated and stored on CPUZ site. It gave me great joy to beat $190 CPU with one I got new for $89 on sale. Note: it is Dec. 2019 mature manufactured.
33:27 tried to tell a guy once that higher clocks take a bit more power, he insisted that only voltage matters. but like any electrical system power draw is a result of work being done, in the form of heat generation...and transistor switching. forgot the context of the argument but i think the guy was saying that an extra 500mghz on the core, as long as voltage stayed the same, power draw would remain the same
Fun fact: The material for PCB (sub) is always mostly glass and epoxy resin & copper for the tracing and via hole plating (of course some other substances, but mainly speaking). More higher end tier board, more fillers. Fillers in the material usually for heatspeading and to try to make the PCB more robust.
This was an incredible breakdown and really helped me with my overclocking adventures and will help my viewers stay safe and not damage their components! Thank you so much!
The oven 'trick' also works because of surface melting, which is a quasi-state where on the surface a layer of atoms melts. This will/can mend cracks forming in the solder balls and its connections, therefore restoring a good connection
-digs out my Atari 800XL¬ Ah, the days when CPU's had no cooling whatsoever (not even a heatsink) and ran at 5 volts - just like everything else in a computer. XD To be fair, there ARE stories of overheating from that era. But it's usually a design with extremely poor airflow and an unusually hot component. (one of the rare components hot enough to require a heatsink) In fact, the most likely thing to overheat in that era would be your computer's power supply... By the way, if you have an old 8 bit system for some reason... Don't trust those old power supplies. Although they may still work, for many computers they're so cheaply made that when they do blow up (when, not if) they'll take most of the circuitry of the computer with them. Given the age of the systems, that's not something you can reasonably fix...
@@williamforbes6919 In ATX the -5V rail was used for ISA bus, it's gone for long time. The -12V rail is still used for RS-232 or RS-422 communication but now almost everything is connected through Ethernet or USB.
@@williamforbes6919 this CPU itself (and its predecessor the 8085) only requires a single +5VCC to run. As I said, the negative voltage rails were used mostly for communication protocols outside the chip.
@@williamforbes6919 Some systems required more than that. But usually due to peripherals. The 800XL is 5 volts DC only (technically you can power it over USB) But the early c64 machines had both 5 volts DC and 12 volts AC. (blame the SID chip for that though; 6502 processors only need 5 volt DC) Interestingly you can still buy brand new 6502 family processors (z80 as well). The modern 65186 chips are fully static designs that are rated for 14 mhz (but many can easily run at 20 mhz or more), and will run happily at 1.8, 3.3 or 5 volts. (though if you use less than 5 volts the maximum clock speeds go down) It's kinda bizarre to think you can buy new replacements for processors that old...
Bender: "Like putting too much air in a balloon!" Fry: "Of course! It's all so simple!" Great video! I'm a simple man, so I appreciate when smart guys break it down Barney-style.
Hmm. Thanks for the info! I am actually having what sounds like it might be a thermal cycling problem, currently. The (438mm^2) GPU is fine when web browsing or playing less graphically-intense games, but as soon as I play a game with "shiny" graphics, I get a BSOD (or sometimes just sudden black screen.) I thought it might be something in the interaction between GPU and power supply, since OCCT's GPU "memtest" runs for at least 4 or 5 minutes (which is where I stopped it manually,) with the core reaching 60-70ºC on DIY hybrid cooling w/120mm rad. The "3D" test that loads the core (and draws more power) crashes almost immediately. If replacing my power supply proves ineffective, I may just resort to partially disassembling my GPU and sticking it in an oven at 170ºC.
This was very informative. Little bit chaotic and not very specific at some points, but you touched so many topics that more details would be even more confusing. Thank you, another great vid!
The thruth is that you are a hidden gem and I really glad that you share these videos with us. I study electronics, and basicly I can fill up those hole's you have in your mind about thge physics of these, but without the knowledge you have specific in overclockingI can't use it in terms of PC hardware and overclocking which I really love.
The electrons wont smash the atoms out of their way. Instead they push the electrons from one atom to another and causing the atoms migrating from one type (e.g. copper) to another "useless" type which is not longer part of the copper-structure and drifts out of the material.
on the topic of water, one of the big advantages of it is actually the thermal volume of water it's self; under a long-term heavy load, water and air cooling performance isn't that different, but where water really has an advantage is in sudden temperature spikes, which would mean a larger thermal difference between the chip and the substrate, are much better managed due to the higher thermal volume of the cooler.
thanks man!! I like the rambling.. this gives me a moment to not hear me rambling with myself and I on this stuff, and this way I can get a few other points of view to add to my own rambling.. and I probably add a shirt to soon..
I suspect that the oxide breakdown dependence on temperature has to do with the electron energy distribution being narrower at low temperature, so you have fewer high-energy electrons that can damage the oxide. It's the Fermi-Dirac distribution.
33:31 in most cases the current rises more than linear because with higher frequency the time were there is a shoot through or bough of the mosfets still changing will get longer because the MOSFET has to work faster.
The Fury X also has a benefit of the built in water cooling. The Fury or Fury Nano cards will be the ones that'll show the thermal aging affects way sooner than the X. The Fury X also throttles at 65'C because of water pump. heh
I recently had my computer report 8GB instead of 16.... I thought I had killed one of my RAM sticks. Turns out I had thermal paste in one of the sockets.
I don't know if I missed it but there is the fact that the smaller the lithography, the lower the voltage and over voltage tolerance. I remember I saw an interview on gamernexus where the interviewee said that the biggest cause of the degradation was by far the voltage due to how you said, the place that heats up is well below the chip and because of that, independent of the cooling system (without subzero) the heat generated there by the overvolt increases exponentially compared to the rest of the chip and that over time slowly ends up burning part of it and depending on the voltage used the heat reaches such a point that it simply does not have time to transfer from there to the rest of it which causes it to burn almost instantly. As mentioned, it is a problem that also tends to worsen a lot with the reduction of lithography since the chip becomes a lot more sensitive to changes in voltage and temperature as it gets smaller and the components tighter and closer.
Ive been hearing stories of +1.3v overclocks significantly degrading zen 2 ryzen chips. Not sure whether to believe the internet on that one. I swear most people that post on overclocking forums dont know what they are talking about.
I heard the same stories about the +1.3v overclocks, I am still curious if this could be true as I have my Ryzen 7 3700X Running on a 1.325V Overclock The max temps are around 67*C under full load, but still I am curious
Well I managed to regrade an 1800x at around 1.40 after months of running it OC'ed. I won't take an OC anymore, only will run stock. To be fair I was running it on a smaller AIO (H80i GT) which wasn't really up to the task. Killed the AIO as well, as it started getting galvanic corrosion issues due to the heat destroying the glycol.
I've had my 3600 for a few months, have settled at 4.1GHz and 1.38 vcore for a while now without any obvious instability and I don't care much about stability for 24/7 stress testing with something like prime95. I am skeptical of all the info about how above 1.3 volts is so dangerous because the stupid thing is supposedly rated for 4.2GHz boost and achieving that on all cores at all times without crashing is impossible with just 1.3 vcore, at least on my chip. Also the marketing about 4.2GHz boost is misleading because leaving everything on auto with overkill custom loop watercooling for tons of thermal headroom the CPU has only a single core intermittently boosted up that high when settings are left on auto. If anything the incredibly shit stock wraith stealth cooler is the real problem. I basically had typical above 90C laptop type temperatures even on auto settings when running cinebench R20. Now under water it's more like 70C and in typical use like gaming the temps are around 50 to 55C with the manual overclock and supposedly high core voltage.
"You can't have negative current." The physicists grant you a pass, since you didn't specify current density. ... Is the Hall effect just a -laminar- turbulent flow of electrons? (Edit: _breathes in electrical engineer_ )
"That's just like the chip got punched in the face" LOL that's what how I describe it to my friends when I'm trying to explain how to OC and about the voltage needed. Just don't punch it in the face too much and it will be a happy camper! I'll definitely be linking this to them!!
27:01 TL:DR Search images for MOSFET Drain current vs Voltage You can see there is a "saturation" of the drain current. At that point is it technically close to linear at that point, but it had a different behaviour from 0 to a certain voltage. Very rough explaination: I am by no mean an expert but if it is helpful I studied physics and I'll share some info: In a MOSFET transistor there are usually 2 voltages at play: the one between the Gate and the Base, and the one between the Drain and Source. The former acts as a "switch" in digital electronics, it serves the purpose of "turning on" the flow of current between Drain and Source. Let's say we keep the Gate and Base at a certain operating voltage and slowly make the Drain-Source voltage go up, Drain current at first goes up very quickly, then it saturates and goes up way slowly. You can find a lot of graphs by searching MOSFET Drain current vs Voltage or here www.researchgate.net/figure/Drain-current-versus-drain-source-voltage-in-trigate-MOSFETs-with-various-values-of-gate_fig5_260522076 That's why twice the voltage (past a certain point) does not equate to twice the current. The reason behind it has to do with how this "gate" operates: it's pretty long to explain here 'cause one needs to know how fundamentally a MOSFET transistor works, but very roughly speaking: the gate "creates" a path for the current to flow between Drain and Source, when the Voltage between Gate and Base is over a certain threshold voltage, lets call it V_{Threshold}. Also typically Base and Source are shorted to the same voltage so we can speak of Gate-Base voltage as Gate-Source voltage: V_{Gate-Source} . When V_{Drain-Source} > V_{Gate-Source} - V_{Threshold}, in in the proximity of the Drain gate voltage is lowered by the drain voltage and drops under the threshold voltage needed to keep this path open, so the conductive section aka "the path", pinches, thus hindering the flow of the current. On one hand you have higher voltage difference that tends to make the current flow higher, but on the other the zone around the drain which is hard for electrons to flow in gets larger, this effects counteract one another, still there is a linear increase in current vs voltage at that point, but one that does not intersect the {0V ;0I} point.
My 3950X was hitting 1.5v or more at times. I set a -.1v offset in my bios. Lost like 10°C and my Cinebench scores went up. I’m half tempted to keep going down to find the happy spot, as this was just a zOMG STOP fix.
As I understand the papers with the psyics is the following. Thermal breakdown = Heating and cooling, for example when solder cracks happen in SMDs (RTX "Space Invader" memory issues) Volts = Oxide breakdowns, for example in Ceramic Capacitors loose their effectiveness over time. (Image ghosting on older electronics) Current = Electron Migration, for example in a CPU when a transistor gate is "open" or "closed" other transistors sitting close or next to each other can cause free electrons to migrate across the gates. Some common exploits manipulate this effect like "Row hammer" in RAM.
@@lawliot It is a Ryzen 1600AF, I can easily overclock it to 4.3Ghz but don't really need the performance right now, so I decided to run it at stock frequency with aggressive undervolt. Benchmark results are the same as with regular voltage but the power consumption is around 49W instead of 80W under load.
@@snowhawk4049 not "forever", but maybe a lifetime if you still can find replacements for the other parts around. Also after 50 years, softwares, OSes, communication protocols and ports will likely be all incompatible.
I appreciate the video. However, I still don't know how to figure out what/how to adjust things on my GPU. I recently started playing Borderlands 3. That game pushes my GPU harder than any other games I've played. I have a GTX 1080ti that is NOT overclocked. But, the game pushes it to 82-83 degrees C. I have been trying to figure out what is a safe voltage to bring it down to, while also increasing the fan speeds earlier to reduce the heat. Can't afford water cooling. And am disabled, so I can't take it apart to repaste it. Any help would be appreciated. Thanks again for the video. Razar.
About electromigration if cause for electromigration is electrons bumping into atoms, than twice voltage mean twice amperage so four times the "kinetic" energy of moving electrons
Ugh...This might not be related to this video but I just bought a new PC with an X570 mobo that has a bad chipset placement (blocked by GPU). I actually looked at your videos for mobo recommendations but somehow 'forgot' about chipset placement factor. My supposedly brand new 3900X PC is having more problems than my much older PC. Don't do my mistake people.
12:50 That's EXACTLY what happened with my Rebrandeon variant, the 280X. And just a month before warranty expired, funnily enough. Couldn't return it in time. I can definitely tell you that the Twin Frozr cooler, infamous for inadequate thermals on the AMD side coupled with insufficient maintenance would yield this fate for sure. The recent hot summers really got it. I wonder if I can still fix it.
Just a quick tip: shoot your videos in segments. That way, if you want to reshoot something, you only need to reshoot one segment. I don't think anyone cares about camera cuts and whatever.
Electromigration sounds a lot like what might be happening to Intel CPUs. I'm guessing most motherboards are running unlimited current and unlimited power, plus thermal velocity boosting to high voltages. Even server boards might be hammering the prefered/boosting cores with high load and getting them up there in volt/current/temperature.
My 7970 thermal cycled to death. I bought it the day it came out and it died about a year ago. I baked it twice and got to use it for about a month after. But baking it every month isn't fun so I finally got rid of it.
I wish there was an Alexa type device that was a wee mobile Buildzoid that all his awesome technical knowledge could be referenced, intact with dry wit. I'd pay double for the dry wit.
Would it be correct to think that chiplet design chips are more resistant against thermal cycling caused damage since the individual dies are smaller and thus expand less?
This is good stuff, I swear I must have degraded my 9900K. Used to run stable at 5.4, just stopped one day out of the blue and never would again since. Not that I’ve tried it recently. I’ve considered delidding, but the risk is too high for a chip that’s still good for 5.3 all core daily It’s been liquid cooled on a high end custom loop, so not quite sure exactly what happened.
@Zfast4you yes, that is correct, well, set at 1.4v, temps when under heavy load in the 80s, but it's a gaming rig, temps rarely exceed the 60s. I'm not upset about it though, it lasted for about almost a year at 5.4. I'll hang onto it for another year, maybe 2, then upgrade. If i start having trouble at 5.3, then it's time to go.
Well, to be perfectly honest, I’ve been over clocking for 15 years and I’ve never experienced this before. I’ve had a few graphics cards die on me, but never had any other components give me trouble. I have a pretty extreme liquid cooling setup, so it’s about as good as it gets without going chilled liquid. Well, I guess I did have a 980x I used to run at 4.0 when it was stable at 4.4, but I’ve been throwing caution to the wind since then. I killed a Kepler Titan, my fault, pushing it too hard for too long. But never a CPU. I did crack the TIM on a 4790K once, but delidding fixed that. Besides, why sit on it at lower clock speeds if it can do more? I don’t keep anything beyond 2 years, 3 max. Regarding performance, I’m obsessed with minimum frame rates, the super high clock speeds are typically what bring those up. And I also have a high end set of RAM OC’d and timings all tuned for max performance. 380w BIOS on both my 2080 Ti’s. So everything is balls to the wall.
I am a computing freak since 1986 and the degradation of micro processors is totally new to me. Even old computer from the 1980's are still working fine. There may other parts fail at some point but the micro processor is the least to fail. 👀
I plan to get a used Vega 56 and powerlimit the shit out of it. You can halve the power consumption for like 20% performance loss. It uses less power and still significantly faster than an RX 580.
Nicely explained and now have better understanding of these failures. I have a faulty HD 7950 with a crappy cooler from sapphire, which seem to explain the overheating and damage happened. I always wonder why we cannot have socketable GPU at this point from NVDIA/AMD ?
i set 1.37v for 4.32 ghz and thats how tight for the chip on ryzen 2nd gen chips....not sure about why people set it pass that ...i have no clue why...probably board is different on anyone...i not planning keep this long since cpu death is closer i fear ...so waiting for switch back to intel for long term usage ..
I don't like this video very much but it's not like I'll make a better one any time soon
We are going to like it anyway, no worries!
Thank you so much for explaining us normal mortals this very interesting stuff!
I watch a lot of channels that take a very technical approach, so it doesn't bother me any lol. Looks good man.
better than nothing
It's an awesome vid. Thanks for all the knowledge
1:12 thermal Cycling
18:15 oxide breakdown
28:52 electromigration
@@Neiva71 Except in order to have comprehension of each of those you need some comprehension of the others
@cas curse If you already watched it, and you're back here again, did you really comprehend the subject matter?
Thanks for releasing this, it was great, even if you don't like it, we do.
Basically every Bz video :)
100%
I learned a lot
Brabão né meu, aprendendo muito aqui
Woah, didn't expect to see you here casual gamers!
The fastest degradation of chips I've ever seen was when I bought some Doritos yesterday.
bruh
You wouldn’t have had that problem if you had just gotten Cool Ranch.
Yo! This is funny🤣
I will proudly say I chuckled at that
I agree, that stuff doesn't last at all, turns into shit within a day
As an Electrical Engineer with a masters, I approve this video. Pretty well explained.
Hey BZ just wanted to say i really appreciate you going over this topic in great detail. There are too many people on forums spouting about high voltages/temperatures but cannot explain why. Thank you
Watched it - understood it - like it. Thank you for this, please keep up with the good work!
Alter, kommst du urspünglich aus einem Ort in Norddeutschland, dessen Namen mit F anfängt und K aufhört?
A slight correction on electromigration. Current only flows on the surface of the conductor, not inside. Electrons jump from one atom to the next, from one end of the conductor to the other, leaving "holes" behind, and the direction of the flow of the current is actually the opposite of the direction of flow of electrons, so, basically, the direction of the flow of "holes". Every conductor has a certain resistance, which is actually the resistance to the electrons jumping from one atom to another, and one can think of it as the amount of hold an atom has over an electron. Also, current always flows in the direction of least resistance, which is now even more obvious. We can observe electromigration as migration of electrons that deviates from a wanted path due to changes that happen in either lower or higher resistance in certain areas of a conductor. If there's more than wanted resistance, electrons will take a different path, if there's less resistance, more electrons will flow on the same path over the same time. Voltage, temperature, impurities in the conductor and many other factors (including electromagnetism) can determine whether the electrons will behave as we want them to or not. So, too much resistance in a wanted path or lower resistance in a not wanted path will be the symptom of electromigration. The rest of the explanation that involves current, temperature and other factors is correct. O, and one more thing, electrons have insanely low mass, so it's not a bump as described, but they lose energy on overcoming the resistance, and are no longer able to follow a path that has more resistance than electron has energy, so they take a path of lower resistance.
[edit] one more note, greater the resistance - more heat, which causes even greater resistance.
Very good explanation.
That is the reason why I'm undervoting/underpowering my 1080 Ti down the point it does perform 5-10% less than stock settings (but plenty enough for me). But use SO much less power. 80-100W less. Having a GTX 1080 Ti use 150-160W is really nice. 60c max on air with very low noise. And the knowledge of much less degradation. Also always undervolting my CPUs when doing stock speed. Undervolting is possible on most chips, because the stock volts are normally set high(to be stable with even the worst binned silicons). Undervolting also saves you money on a bigger cooler and on the electric bill. But yeah... undervolting plus a bigger cooler is the best for longevity.
I have a retro rig with a QX6850 (12 years old) that at stock freq, can be undervoltet quite a bit. So must still be in in good nic ;)
I use i3 530. Still rocking. No GPU and 3 Gb RAM overclocked. Stock cooler. 9 years almost 10. Playing arkham assylum :)
@@karl9791 whats the point of keeping an old processor alive for so much time?
Like, idk where u from, but in Brazil we pay at least 3 times more in eletronics and we still dont want to keep the same cpu for 10 years.
@@Shadower2014 no one want keep it for 10 years. For me, the ideal would be swap every launch announcement. But is what it is. My point about my cpu is simple, intel have a healthy long life. And I use it almost 24/7. So.. Just that. And I have another pc. New one. Waiting just for the rtx 3080. And still no plans for rework in the old one, maybe some change in the storage... Maybe. Idk.
Modern GPU's like the 1080ti doesnt degrade like CPU chip does. Linus made a video about this.
@@MovementCultivation All electronics degrade. Just the variable how quickly..
And actually the 1080 Ti might degrade faster than a smaller GPU die (like a 1070) especially if loaded with many thermal cycles. Any big die GPU chip generally degrade faster than the smaller ones with many thermal cycles. But As long as you use it within Specs (stock settings) it should have no problem lasting for years (5-7 yrs) Keep thermals and voltage as low as possible, and you gain less degredation and longer lifespan.
A way to check chip degradation, is to try running it low as possible voltage vs highest clockspeed as brand new, and then 3 years after, try it again. It might need a bit more voltage to run at the same clockspeed it could when new. Meaning its degraded. Might still be okay to run perfectly within specs, but degraded compared to when new.
Memory is a more fragile thing with modern graphics cards. Nvidia cards do not have memory temp reading and I know GDDR6(X) run hot. The new cards give you speed.. but not lifespan. Nvidia has no intention on making these last for 5+ years. (Id be surpriced if more than half of the RTX 30xx cards still works fine in 6-7 years) Nvidia knows the market demands Performance now. Not long term endurance.
Not overclocking the memory will be a good idea. Downclocking it a bit will give it a cooler longer life. Especially if its going to be put at hard work.
24:27 all of buildzoid's chips
"I can't believe you've done this"
10/10 ramblings.
Video uploaded 50min ago.
Comment is 1 week ago.
Way to go TH-cam.
Maybe he got access to the video before it was public?
@@penumbrum3135 It's all the magical powers of patreon.
Room temperature can also be like 5C when you're the extra ghetto type of overclocker who just opens the window in the winter for extra low ambient temperatures since chilled water cooling or dry ice is too expensive.
It depends where you live :D
@@PainterVierax Norway here. Managed to get the water temp in my loop to 7C by opening the window. My 1080 Ti was chillin at 18C under full load overclocked. It's pretty dank
5Ghz on a R5 2600 using an air cooler gang
Just put your computer outside bro.
Theres no way you hit 5G off an air cooler.....what voltage are you using to hit 5G & whats your fps gain if so?
Hi Buildzoid, I'm both a trained physicist and chemist. If you have questions or want me to make a video going into more detail on Oxide Breakdown and Electromigration just shoot me a reply.
@RectalDiscourse Dude buildzoid specifically asked for a comment if anyone could help
@RectalDiscourse "degenerate waste of human organs" I really need to use that insult, it's really cool
@RectalDiscourse nobody feels your pain. We put our phones down and... You know... Live nice lives. Try it.
@RectalDiscourse username checks out
@RectalDiscourse You feign giving advice for someone else to be taken seriously on the internet, while simultaneous and ironically making an example of yourself for how not to be taken seriously at all. There's no room in hell for all the psychological baggage you seem to have.
Thanks for the information, love the scientific vulgarization of the different processes and phenomena.
Thank you for sharing. I really appreciate what your sharing. Some of us don't know, wouldn't know, or even consider the things you are explaining. You help make people smarter about things. TY!
Dude, this is amazing info for us mere mortals! Great video!
Interesting and enlightening. Thank you for your support.
Very good vid, it gave me the confidence to try to beat the next gen big brother of my current CPU. I did 4.2 all core OC at 1.4V on a Ryzen 5 2600X with idle temp of 27.50c/load over time temp 71.50c using the Wraith Max box cooler for FX 8390 CPUs. This allowed me to beat the CPUZ single core score of the stock Ryzen 3600 and match it in multicore score, validated and stored on CPUZ site. It gave me great joy to beat $190 CPU with one I got new for $89 on sale. Note: it is Dec. 2019 mature manufactured.
This was a VERRRYYY good video and has greatly educated me on-chip degradation and safer OC habits
I just want to re iterate: Thanks for your efforts. 👍
33:27 tried to tell a guy once that higher clocks take a bit more power, he insisted that only voltage matters. but like any electrical system power draw is a result of work being done, in the form of heat generation...and transistor switching. forgot the context of the argument but i think the guy was saying that an extra 500mghz on the core, as long as voltage stayed the same, power draw would remain the same
🤦♂️
The dumbass thinks CPUs are literally resistors, lmao.
Fun fact: The material for PCB (sub) is always mostly glass and epoxy resin & copper for the tracing and via hole plating (of course some other substances, but mainly speaking). More higher end tier board, more fillers. Fillers in the material usually for heatspeading and to try to make the PCB more robust.
This was an incredible breakdown and really helped me with my overclocking adventures and will help my viewers stay safe and not damage their components! Thank you so much!
The oven 'trick' also works because of surface melting, which is a quasi-state where on the surface a layer of atoms melts. This will/can mend cracks forming in the solder balls and its connections, therefore restoring a good connection
-digs out my Atari 800XL¬
Ah, the days when CPU's had no cooling whatsoever (not even a heatsink) and ran at 5 volts - just like everything else in a computer. XD
To be fair, there ARE stories of overheating from that era.
But it's usually a design with extremely poor airflow and an unusually hot component. (one of the rare components hot enough to require a heatsink)
In fact, the most likely thing to overheat in that era would be your computer's power supply...
By the way, if you have an old 8 bit system for some reason...
Don't trust those old power supplies. Although they may still work, for many computers they're so cheaply made that when they do blow up (when, not if) they'll take most of the circuitry of the computer with them.
Given the age of the systems, that's not something you can reasonably fix...
5v single supply, a huge step forwards over requiring -5v and -12v rails.
@@williamforbes6919 In ATX the -5V rail was used for ISA bus, it's gone for long time. The -12V rail is still used for RS-232 or RS-422 communication but now almost everything is connected through Ethernet or USB.
@@PainterVierax I was referring to the 8086's power rail requirements.
@@williamforbes6919 this CPU itself (and its predecessor the 8085) only requires a single +5VCC to run. As I said, the negative voltage rails were used mostly for communication protocols outside the chip.
@@williamforbes6919 Some systems required more than that.
But usually due to peripherals.
The 800XL is 5 volts DC only (technically you can power it over USB)
But the early c64 machines had both 5 volts DC and 12 volts AC.
(blame the SID chip for that though; 6502 processors only need 5 volt DC)
Interestingly you can still buy brand new 6502 family processors (z80 as well).
The modern 65186 chips are fully static designs that are rated for 14 mhz (but many can easily run at 20 mhz or more), and will run happily at 1.8, 3.3 or 5 volts.
(though if you use less than 5 volts the maximum clock speeds go down)
It's kinda bizarre to think you can buy new replacements for processors that old...
Bender: "Like putting too much air in a balloon!"
Fry: "Of course! It's all so simple!"
Great video! I'm a simple man, so I appreciate when smart guys break it down Barney-style.
So he's Lela in this situation?
Great lesson for free, thank you so much Prof. BuildZoid!
Hmm. Thanks for the info! I am actually having what sounds like it might be a thermal cycling problem, currently. The (438mm^2) GPU is fine when web browsing or playing less graphically-intense games, but as soon as I play a game with "shiny" graphics, I get a BSOD (or sometimes just sudden black screen.) I thought it might be something in the interaction between GPU and power supply, since OCCT's GPU "memtest" runs for at least 4 or 5 minutes (which is where I stopped it manually,) with the core reaching 60-70ºC on DIY hybrid cooling w/120mm rad. The "3D" test that loads the core (and draws more power) crashes almost immediately. If replacing my power supply proves ineffective, I may just resort to partially disassembling my GPU and sticking it in an oven at 170ºC.
Happy 1st Birthday to one of the best Videos on YT!
This was very informative. Little bit chaotic and not very specific at some points, but you touched so many topics that more details would be even more confusing. Thank you, another great vid!
So basically, keep temps low, and voltage moderate.. Thanks for all the details and explenations in between
Love this video! Very informative and thought provoking.
Very interesting vid. I flashed my GTX770 yesterday to -100MHz core and made a custom fan curve because it is my child and I dont' want to loose it.
The thruth is that you are a hidden gem and I really glad that you share these videos with us. I study electronics, and basicly I can fill up those hole's you have in your mind about thge physics of these, but without the knowledge you have specific in overclockingI can't use it in terms of PC hardware and overclocking which I really love.
Buildzoid doesn't care about DX10 or lower. We found his true love.. tesselation!
#Buildzoid - Punching chips in the face since 2015. (24:23)
The electrons wont smash the atoms out of their way. Instead they push the electrons from one atom to another and causing the atoms migrating from one type (e.g. copper) to another "useless" type which is not longer part of the copper-structure and drifts out of the material.
on the topic of water, one of the big advantages of it is actually the thermal volume of water it's self; under a long-term heavy load, water and air cooling performance isn't that different, but where water really has an advantage is in sudden temperature spikes, which would mean a larger thermal difference between the chip and the substrate, are much better managed due to the higher thermal volume of the cooler.
Cool video! Love watching these kinds of explanations
So much for rambling, this is a crash course! good job as always bz.
thanks man!! I like the rambling.. this gives me a moment to not hear me rambling with myself and I on this stuff, and this way I can get a few other points of view to add to my own rambling.. and I probably add a shirt to soon..
It`s a great vid bro, very informative despite a few uncertainties.
I suspect that the oxide breakdown dependence on temperature has to do with the electron energy distribution being narrower at low temperature, so you have fewer high-energy electrons that can damage the oxide. It's the Fermi-Dirac distribution.
I love videos like this one.
A topic i’ve always wondered about, thank you
33:31 in most cases the current rises more than linear because with higher frequency the time were there is a shoot through or bough of the mosfets still changing will get longer because the MOSFET has to work faster.
The Fury X also has a benefit of the built in water cooling. The Fury or Fury Nano cards will be the ones that'll show the thermal aging affects way sooner than the X. The Fury X also throttles at 65'C because of water pump. heh
I recently had my computer report 8GB instead of 16.... I thought I had killed one of my RAM sticks. Turns out I had thermal paste in one of the sockets.
I don't know if I missed it but there is the fact that the smaller the lithography, the lower the voltage and over voltage tolerance. I remember I saw an interview on gamernexus where the interviewee said that the biggest cause of the degradation was by far the voltage due to how you said, the place that heats up is well below the chip and because of that, independent of the cooling system (without subzero) the heat generated there by the overvolt increases exponentially compared to the rest of the chip and that over time slowly ends up burning part of it and depending on the voltage used the heat reaches such a point that it simply does not have time to transfer from there to the rest of it which causes it to burn almost instantly. As mentioned, it is a problem that also tends to worsen a lot with the reduction of lithography since the chip becomes a lot more sensitive to changes in voltage and temperature as it gets smaller and the components tighter and closer.
I really enjoyed this video. I was learning throughout the video.
I had a laptop who's Nvidia video card died like this. After resoldering it worked 3 more weeks
Ive been hearing stories of +1.3v overclocks significantly degrading zen 2 ryzen chips. Not sure whether to believe the internet on that one. I swear most people that post on overclocking forums dont know what they are talking about.
The last part of your comment is spot on.
I heard the same stories about the +1.3v overclocks, I am still curious if this could be true as I have my Ryzen 7 3700X Running on a 1.325V Overclock
The max temps are around 67*C under full load, but still I am curious
Well I managed to regrade an 1800x at around 1.40 after months of running it OC'ed. I won't take an OC anymore, only will run stock. To be fair I was running it on a smaller AIO (H80i GT) which wasn't really up to the task. Killed the AIO as well, as it started getting galvanic corrosion issues due to the heat destroying the glycol.
I've had my 3600 for a few months, have settled at 4.1GHz and 1.38 vcore for a while now without any obvious instability and I don't care much about stability for 24/7 stress testing with something like prime95. I am skeptical of all the info about how above 1.3 volts is so dangerous because the stupid thing is supposedly rated for 4.2GHz boost and achieving that on all cores at all times without crashing is impossible with just 1.3 vcore, at least on my chip.
Also the marketing about 4.2GHz boost is misleading because leaving everything on auto with overkill custom loop watercooling for tons of thermal headroom the CPU has only a single core intermittently boosted up that high when settings are left on auto.
If anything the incredibly shit stock wraith stealth cooler is the real problem. I basically had typical above 90C laptop type temperatures even on auto settings when running cinebench R20. Now under water it's more like 70C and in typical use like gaming the temps are around 50 to 55C with the manual overclock and supposedly high core voltage.
@@jeffm2787 1.4v wasn't ever regarded as safe for 1st gen Ryzen though
Very interesting video, thanks for making this !
Hey, this is better than my chem lecture. More interesting too.
👍👍 Really helpful. My knowledge of chip degredation finishes at the n.... So this was a great insight.
this is pretty interesting, it helps us understand why our chips do degrade over time which is my main concern as a non-overclocking consumer.
27:43
Patrons, you know what to do. Let’s make history.
"You can't have negative current."
The physicists grant you a pass, since you didn't specify current density.
... Is the Hall effect just a -laminar- turbulent flow of electrons?
(Edit: _breathes in electrical engineer_ )
My takeaway from this is the longer the video the more Buildzoid breakdown we have.
???
This is an important topic. Thanks.
"That's just like the chip got punched in the face" LOL that's what how I describe it to my friends when I'm trying to explain how to OC and about the voltage needed. Just don't punch it in the face too much and it will be a happy camper! I'll definitely be linking this to them!!
That part actually made me bust out laughing, lol.
his 1's really look like 7's sometimes
No
27:01
TL:DR
Search images for MOSFET Drain current vs Voltage
You can see there is a "saturation" of the drain current. At that point is it technically close to linear at that point, but it had a different behaviour from 0 to a certain voltage.
Very rough explaination:
I am by no mean an expert but if it is helpful I studied physics and I'll share some info: In a MOSFET transistor there are usually 2 voltages at play: the one between the Gate and the Base, and the one between the Drain and Source. The former acts as a "switch" in digital electronics, it serves the purpose of "turning on" the flow of current between Drain and Source. Let's say we keep the Gate and Base at a certain operating voltage and slowly make the Drain-Source voltage go up, Drain current at first goes up very quickly, then it saturates and goes up way slowly. You can find a lot of graphs by searching MOSFET Drain current vs Voltage or here www.researchgate.net/figure/Drain-current-versus-drain-source-voltage-in-trigate-MOSFETs-with-various-values-of-gate_fig5_260522076
That's why twice the voltage (past a certain point) does not equate to twice the current.
The reason behind it has to do with how this "gate" operates: it's pretty long to explain here 'cause one needs to know how fundamentally a MOSFET transistor works, but very roughly speaking: the gate "creates" a path for the current to flow between Drain and Source, when the Voltage between Gate and Base is over a certain threshold voltage, lets call it V_{Threshold}.
Also typically Base and Source are shorted to the same voltage so we can speak of Gate-Base voltage as Gate-Source voltage: V_{Gate-Source} .
When V_{Drain-Source} > V_{Gate-Source} - V_{Threshold}, in in the proximity of the Drain gate voltage is lowered by the drain voltage and drops under the threshold voltage needed to keep this path open, so the conductive section aka "the path", pinches, thus hindering the flow of the current.
On one hand you have higher voltage difference that tends to make the current flow higher, but on the other the zone around the drain which is hard for electrons to flow in gets larger, this effects counteract one another, still there is a linear increase in current vs voltage at that point, but one that does not intersect the {0V ;0I} point.
How long would it take to kill a chip by keeping it stored, unpowered, not connected to anything, in its ideal conditions in the original retail tray?
My 3950X was hitting 1.5v or more at times. I set a -.1v offset in my bios. Lost like 10°C and my Cinebench scores went up. I’m half tempted to keep going down to find the happy spot, as this was just a zOMG STOP fix.
As I understand the papers with the psyics is the following.
Thermal breakdown = Heating and cooling, for example when solder cracks happen in SMDs (RTX "Space Invader" memory issues)
Volts = Oxide breakdowns, for example in Ceramic Capacitors loose their effectiveness over time. (Image ghosting on older electronics)
Current = Electron Migration, for example in a CPU when a transistor gate is "open" or "closed" other transistors sitting close or next to each other can cause free electrons to migrate across the gates. Some common exploits manipulate this effect like "Row hammer" in RAM.
47:30 "you know what, just take the CPU out of the motherboard and just keep it in the fridge. Only way to keep your cpu from degrading"
Great BZ! I was missing some ramblings
So my Ryzen runs on 0.68-1.0V and the temperature is in a range from 24-44°C. If I keep it that way it should basically run forever.
Sounds like good silicon, overclock the snot out of it.
@@lawliot It is a Ryzen 1600AF, I can easily overclock it to 4.3Ghz but don't really need the performance right now, so I decided to run it at stock frequency with aggressive undervolt. Benchmark results are the same as with regular voltage but the power consumption is around 49W instead of 80W under load.
Yeah ... but running a cpu forever is rather boring ... as luuzeri suggested, go and oc the snot out of it
@@snowhawk4049 not "forever", but maybe a lifetime if you still can find replacements for the other parts around. Also after 50 years, softwares, OSes, communication protocols and ports will likely be all incompatible.
My 2700X runs at 66 degrees Celsius. It's doomed. =\
Thank you so much BZ!
0:45 humble buildzoid :)
Actually a ton of good points that are nice to know and really helpful. Very good content, wouldnt know where to learn it, if not here :-)
I appreciate the video. However, I still don't know how to figure out what/how to adjust things on my GPU.
I recently started playing Borderlands 3. That game pushes my GPU harder than any other games I've played. I have a GTX 1080ti that is NOT overclocked. But, the game pushes it to 82-83 degrees C.
I have been trying to figure out what is a safe voltage to bring it down to, while also increasing the fan speeds earlier to reduce the heat.
Can't afford water cooling. And am disabled, so I can't take it apart to repaste it. Any help would be appreciated.
Thanks again for the video.
Razar.
I just had a CPU die on me after probably 4 or 5 years at 1.38v, thank you for explaining what actually happened!
About electromigration if cause for electromigration is electrons bumping into atoms, than twice voltage mean twice amperage so four times the "kinetic" energy of moving electrons
As a network engineer, voltages and oxidation are part of our everyday lives, atrophy.
The over trick works since the 8000x series ... nVidia screwed the underfill and layer1 connections fail very early at normal use
*over trick = oven trick
full ack on your argument!
Whoa, new GIMP looks like Blender & Photoshop had a baby.
This was brilliant for my brain, easily the best video that explains what you are trying do very well. Dumb explanation videos are good.
Ugh...This might not be related to this video but I just bought a new PC with an X570 mobo that has a bad chipset placement (blocked by GPU). I actually looked at your videos for mobo recommendations but somehow 'forgot' about chipset placement factor. My supposedly brand new 3900X PC is having more problems than my much older PC. Don't do my mistake people.
What are your problems with 3900X sir?
Appreciate it, good video
Put my 3600 in the fridge right after finishing this video, hopefully it'll last long enough to give me my money's worth now. Thanks Buildzoid!
I will only idle with my cpu for now on to minimize electromigration.
12:50 That's EXACTLY what happened with my Rebrandeon variant, the 280X. And just a month before warranty expired, funnily enough. Couldn't return it in time.
I can definitely tell you that the Twin Frozr cooler, infamous for inadequate thermals on the AMD side coupled with insufficient maintenance would yield this fate for sure. The recent hot summers really got it. I wonder if I can still fix it.
Just a quick tip: shoot your videos in segments. That way, if you want to reshoot something, you only need to reshoot one segment. I don't think anyone cares about camera cuts and whatever.
Electromigration sounds a lot like what might be happening to Intel CPUs. I'm guessing most motherboards are running unlimited current and unlimited power, plus thermal velocity boosting to high voltages. Even server boards might be hammering the prefered/boosting cores with high load and getting them up there in volt/current/temperature.
My 7970 thermal cycled to death. I bought it the day it came out and it died about a year ago. I baked it twice and got to use it for about a month after. But baking it every month isn't fun so I finally got rid of it.
I wish there was an Alexa type device that was a wee mobile Buildzoid that all his awesome technical knowledge could be referenced, intact with dry wit. I'd pay double for the dry wit.
Would it be correct to think that chiplet design chips are more resistant against thermal cycling caused damage since the individual dies are smaller and thus expand less?
This is good stuff, I swear I must have degraded my 9900K. Used to run stable at 5.4, just stopped one day out of the blue and never would again since. Not that I’ve tried it recently. I’ve considered delidding, but the risk is too high for a chip that’s still good for 5.3 all core daily
It’s been liquid cooled on a high end custom loop, so not quite sure exactly what happened.
@Zfast4you yes, that is correct, well, set at 1.4v, temps when under heavy load in the 80s, but it's a gaming rig, temps rarely exceed the 60s.
I'm not upset about it though, it lasted for about almost a year at 5.4. I'll hang onto it for another year, maybe 2, then upgrade. If i start having trouble at 5.3, then it's time to go.
Well, to be perfectly honest, I’ve been over clocking for 15 years and I’ve never experienced this before. I’ve had a few graphics cards die on me, but never had any other components give me trouble. I have a pretty extreme liquid cooling setup, so it’s about as good as it gets without going chilled liquid. Well, I guess I did have a 980x I used to run at 4.0 when it was stable at 4.4, but I’ve been throwing caution to the wind since then. I killed a Kepler Titan, my fault, pushing it too hard for too long. But never a CPU. I did crack the TIM on a 4790K once, but delidding fixed that.
Besides, why sit on it at lower clock speeds if it can do more? I don’t keep anything beyond 2 years, 3 max. Regarding performance, I’m obsessed with minimum frame rates, the super high clock speeds are typically what bring those up. And I also have a high end set of RAM OC’d and timings all tuned for max performance. 380w BIOS on both my 2080 Ti’s. So everything is balls to the wall.
I am a computing freak since 1986 and the degradation of micro processors is totally new to me. Even old computer from the 1980's are still working fine. There may other parts fail at some point but the micro processor is the least to fail. 👀
this confirms my lowtemp fetish :)
Why hello there
Are we also idling at 28-31C and 50C under load? ( ͡° ͜ʖ ͡°)
@@The_Man_In_Red get a room you two
I wonder how vega 56 and 64 managed to survive this long when they run HOT and their cores are huge.
Interposer maybe? I don't know if that has a knock on effect for this.
they aren't that old
I plan to get a used Vega 56 and powerlimit the shit out of it. You can halve the power consumption for like 20% performance loss. It uses less power and still significantly faster than an RX 580.
*Worrying Laughter*
@@TheGyuuula my rx 480 uses 130w maximum
Nice to listen to while trying to fall asleep
We have run chips at -190 C for decades, in specialized low noise amplifiers.
Now i know why my Zotac 780 AMP! started working again after baking but only for 2 months. Swapped it for a EVGA 980Ti SC+. thx
Nicely explained and now have better understanding of these failures.
I have a faulty HD 7950 with a crappy cooler from sapphire, which seem to explain the overheating and damage happened.
I always wonder why we cannot have socketable GPU at this point from NVDIA/AMD ?
I "fixed" my old GTX 580 with the heatgun method about 3 months ago and it's still working fine on a daily basis...
Would the Zen 4 7000 series degrade if it's constantly running 95c long term?
i set 1.37v for 4.32 ghz and thats how tight for the chip on ryzen 2nd gen chips....not sure about why people set it pass that ...i have no clue why...probably board is different on anyone...i not planning keep this long since cpu death is closer i fear ...so waiting for switch back to intel for long term usage ..
You are talking nonsense lol.