The Brewing Problem with GPU Power Design Transients
video description
Date: 2022-06-22
Related videos
Comments and reviews: 10
Irritated
Electrical Design Engineer here: Great video, nothing inherently wrong but would like to maybe bring up a focus that is often overlooked when people say transients. While the peak is important, another major factor is the rate to peak, math nerds would know as dV(I)/dT, where V, I, and T are voltage, current, and time respectively. Basically how fast a system goes from 0 to 100. Because a spike is short does not mean it's safe. Infact, the faster a voltage for example spikes from 0 to 100, the worse it is in terms of current draw (draws SIGNIFICANTLY more). Don't think it's just a short spike it's fine . If it was a smoother ramp up to peak, and if the components are rated to handle the peaks, the system would be significantly less likely to be damaged.
Also, never build to cover the max. Always give tolerance. So if your nominal is 750, provide a psu that is rated to 850 and etc. In the industry, we NEVER want to run anything at it's absolute maximums because that is just asking for things to degrade faster than necessary.
reply
Electrical Design Engineer here: Great video, nothing inherently wrong but would like to maybe bring up a focus that is often overlooked when people say transients. While the peak is important, another major factor is the rate to peak, math nerds would know as dV(I)/dT, where V, I, and T are voltage, current, and time respectively. Basically how fast a system goes from 0 to 100. Because a spike is short does not mean it's safe. Infact, the faster a voltage for example spikes from 0 to 100, the worse it is in terms of current draw (draws SIGNIFICANTLY more). Don't think it's just a short spike it's fine . If it was a smoother ramp up to peak, and if the components are rated to handle the peaks, the system would be significantly less likely to be damaged.
Also, never build to cover the max. Always give tolerance. So if your nominal is 750, provide a psu that is rated to 850 and etc. In the industry, we NEVER want to run anything at it's absolute maximums because that is just asking for things to degrade faster than necessary.
reply
caldark2005
I think I recently had this issue.
My setup was
AMD 5950x
AMD 6900XT
Asus TUF x570 motherboard
1000w platinum power supply cant remember make now but not a cheap one.
Anyway sometimes when playing some games would do exactly what yours did at the start of the video. I also got a UEFI bios error about the graphics card not being compatible on reboot.
So assumed it was a 6900xt coursing it.. swapped out to my old card a founders edition 1080ti and everything seemed good so thought yeah must be graphics card.. so picked up a 3080 as they are actually a decent price now.. so got it all installed and started get the same issues :(
In the end picked up a replacement motherboard a higher end MSI MPG board.. and the problem has now gone away.. I still need to pop my 6900xt back in to see if that works fine as well now hopefully will get around to that this weekend. but yeah if its good then I guess it was the motherboard so not sure if it was that having problems with transient spikes or just a duff board..
reply
I think I recently had this issue.
My setup was
AMD 5950x
AMD 6900XT
Asus TUF x570 motherboard
1000w platinum power supply cant remember make now but not a cheap one.
Anyway sometimes when playing some games would do exactly what yours did at the start of the video. I also got a UEFI bios error about the graphics card not being compatible on reboot.
So assumed it was a 6900xt coursing it.. swapped out to my old card a founders edition 1080ti and everything seemed good so thought yeah must be graphics card.. so picked up a 3080 as they are actually a decent price now.. so got it all installed and started get the same issues :(
In the end picked up a replacement motherboard a higher end MSI MPG board.. and the problem has now gone away.. I still need to pop my 6900xt back in to see if that works fine as well now hopefully will get around to that this weekend. but yeah if its good then I guess it was the motherboard so not sure if it was that having problems with transient spikes or just a duff board..
reply
OnyxTech
Thanks Steve and team for the very comprehensive overview. I'm very excited about the quality and direction of the content from you all over the last year and since the move!
Recently I heard some speculation from the leaker community that a large reason for the 3090ti was to test the board design for the high end 40 series. Taking that into account and looking at the data that you provided, does it make sense to test more 3090tis?
My speculation is that Nvidia has been feeling the pressure from PSU manufacturers and decided to come up with a new reference board design that significantly reduces the transients that are created by their GPUs. Your data on the 3090ti would seem to corroborate this, but with only one 3090ti in the data, it's hardly conclusive.
It would also be interesting to see a board design comparison between 3090 and 3090tis from the same manufacturer to see what mitigations at the board level are helping to prevent these transients.
reply
Thanks Steve and team for the very comprehensive overview. I'm very excited about the quality and direction of the content from you all over the last year and since the move!
Recently I heard some speculation from the leaker community that a large reason for the 3090ti was to test the board design for the high end 40 series. Taking that into account and looking at the data that you provided, does it make sense to test more 3090tis?
My speculation is that Nvidia has been feeling the pressure from PSU manufacturers and decided to come up with a new reference board design that significantly reduces the transients that are created by their GPUs. Your data on the 3090ti would seem to corroborate this, but with only one 3090ti in the data, it's hardly conclusive.
It would also be interesting to see a board design comparison between 3090 and 3090tis from the same manufacturer to see what mitigations at the board level are helping to prevent these transients.
reply
Skioles
I've had problem with a 1050ti that would just crash it's drivers while playing games. This exact model of gpu didn't have additional power connector. I rma'd it like 9 times and clearly they couldn't figure out what is wrong with it. The other gpu - 960 with extra 6-pin plug worked fine on the same system. I tested a bunch of different drivers over time, reinstalled whole os multiple times and nothing helped. Worth to mention I used it with a gen2 pcie board.
Could this be a issue with a not enough power from pcie port to accommodate a sudden spike of power requirement? I know it's impossible to diagnose a problem without seeing it, I am asking if this would be possible explanation. I don't own this gpu anymore and it seems like there is no problems with it anymore, but I also tested it on my old system for a while before getting rid of it and suddenly it worked fine.
reply
I've had problem with a 1050ti that would just crash it's drivers while playing games. This exact model of gpu didn't have additional power connector. I rma'd it like 9 times and clearly they couldn't figure out what is wrong with it. The other gpu - 960 with extra 6-pin plug worked fine on the same system. I tested a bunch of different drivers over time, reinstalled whole os multiple times and nothing helped. Worth to mention I used it with a gen2 pcie board.
Could this be a issue with a not enough power from pcie port to accommodate a sudden spike of power requirement? I know it's impossible to diagnose a problem without seeing it, I am asking if this would be possible explanation. I don't own this gpu anymore and it seems like there is no problems with it anymore, but I also tested it on my old system for a while before getting rid of it and suddenly it worked fine.
reply
StrikeFirst1171
I have a RTX 3090 with a 450 watt VBIOS, running through a corsair AX 850 titanium power supply I thankfully had none of those issues.
However I really didn't like the idea of a GPU drawing that much power so at the point I've power limited it to 70 percent and it still runs between 1800mhz-2000mhz depending on the title its running.
I'd rather take a 1-2 percent performance hit than deal with increased memory, hotspot temperatures on the GPU and improve reliability of it not tripping my power supply any day.
Thanks GN for this level of quality in your content investigating the issues because the 40 series power requirements downright terrify me where things are going at the moment.
These spikes make me wonder whether I'll be power limiting or undervolting every new GPU in the future if the wattage is going to keep leaping at such an alarming rate.
reply
I have a RTX 3090 with a 450 watt VBIOS, running through a corsair AX 850 titanium power supply I thankfully had none of those issues.
However I really didn't like the idea of a GPU drawing that much power so at the point I've power limited it to 70 percent and it still runs between 1800mhz-2000mhz depending on the title its running.
I'd rather take a 1-2 percent performance hit than deal with increased memory, hotspot temperatures on the GPU and improve reliability of it not tripping my power supply any day.
Thanks GN for this level of quality in your content investigating the issues because the 40 series power requirements downright terrify me where things are going at the moment.
These spikes make me wonder whether I'll be power limiting or undervolting every new GPU in the future if the wattage is going to keep leaping at such an alarming rate.
reply
Richard
Along with the GPU power design issues, let's not forget the SIZE of these graphic cards! For crying out loud... why are those newer graphic cards getting bigger and bigger? Are we going to see graphic cards going taking up 5, 6, 7 slots on the mainboard? It's almost like putting in a computer inside of a computer! The insanity has not stop! Graphic card manufacturers NEED to start implementing better AND smaller cooling systems that don't take up all of your precious mainboard slots. If the graphic card manufacturers can adapt their graphic card technology to the laptops (including those thin ones), which do NOT take up that much power draw, why can't they do the same for desktop computers? That's the one thing I don't get.
reply
Along with the GPU power design issues, let's not forget the SIZE of these graphic cards! For crying out loud... why are those newer graphic cards getting bigger and bigger? Are we going to see graphic cards going taking up 5, 6, 7 slots on the mainboard? It's almost like putting in a computer inside of a computer! The insanity has not stop! Graphic card manufacturers NEED to start implementing better AND smaller cooling systems that don't take up all of your precious mainboard slots. If the graphic card manufacturers can adapt their graphic card technology to the laptops (including those thin ones), which do NOT take up that much power draw, why can't they do the same for desktop computers? That's the one thing I don't get.
reply
DavidsonMA
Hey, so i've got a SFF build in a NR200P, i got a Corsair SF600 (Platinium Version) hooked up to a 3080/ R5 3600 on a B540i AORUS MOBO (1NVME Drive and 2 2.5 HDD and SSD), ive done no undervolting on the system, ive been using it for quite a while (scoring an early 3080 helped alot) and i've never get any random shutdown, i play alot of triple AAA titles (that i like to max out in order to get the maximum out of my system) so im pretty sure that the PSU is working hard, nevertheless no random shutdown on sight, idk if corsair PSU's are stronger, eitherway im happy with it
reply
Hey, so i've got a SFF build in a NR200P, i got a Corsair SF600 (Platinium Version) hooked up to a 3080/ R5 3600 on a B540i AORUS MOBO (1NVME Drive and 2 2.5 HDD and SSD), ive done no undervolting on the system, ive been using it for quite a while (scoring an early 3080 helped alot) and i've never get any random shutdown, i play alot of triple AAA titles (that i like to max out in order to get the maximum out of my system) so im pretty sure that the PSU is working hard, nevertheless no random shutdown on sight, idk if corsair PSU's are stronger, eitherway im happy with it
reply
itech
I have a curious question. Gamers for the most part want performance Now if we used cars for example, if you want performance, not only does it cost more, requires more maintenance, and consumes more gas, I mean look at race cars. And when we look throughout the world, high performance is usually corralated with cost and consumption. Could you imagine expecting to dry your hair in a minute with a 100 watt hair dryer??
So I am curious as to why most reviewers and critics of PC parts constantly bash the power consumption of GPU's, but yet expect top of the line performance??
reply
I have a curious question. Gamers for the most part want performance Now if we used cars for example, if you want performance, not only does it cost more, requires more maintenance, and consumes more gas, I mean look at race cars. And when we look throughout the world, high performance is usually corralated with cost and consumption. Could you imagine expecting to dry your hair in a minute with a 100 watt hair dryer??
So I am curious as to why most reviewers and critics of PC parts constantly bash the power consumption of GPU's, but yet expect top of the line performance??
reply
Spark
I'll have to admit, I cannot blame videocards for this kind of problem. I had an 450w ATX power supply powering a P8H61M-LX3 R.20 with a n i3 2120 with 2 4GB DDR3 sticks and a GT610. Although I've used all the USB slots for peripherals. Still though, it should not have been enough to overload a 450w but it showed the same intermittent issue. This suggests a board power issue but I can never be too sure.
EDIT: I double checked. It seemed that it was a 450w since I recall upgrading to a 600w and noted an increase of 150w in my log notes.
reply
I'll have to admit, I cannot blame videocards for this kind of problem. I had an 450w ATX power supply powering a P8H61M-LX3 R.20 with a n i3 2120 with 2 4GB DDR3 sticks and a GT610. Although I've used all the USB slots for peripherals. Still though, it should not have been enough to overload a 450w but it showed the same intermittent issue. This suggests a board power issue but I can never be too sure.
EDIT: I double checked. It seemed that it was a 450w since I recall upgrading to a 600w and noted an increase of 150w in my log notes.
reply
wakkatidus1234
My 5700 xt taiichi spiked thenhell out of mine all the damn time. Apex legends about to win a big match. A bit yoo much going on with het another 3rd party gank. Pewwwwwwwww. Bye bye victory. Piwered back on 10 minute ban. When i sold it nd went bavk to my old card no issues. Wasnt even of the oc switch iption as that one wad even worse lol. These are 100 % gpu fault. Rven have a good power supply with a mighty single rail yo handle oc. Id hate to think how often it would be with an average vheaper rail.
reply
My 5700 xt taiichi spiked thenhell out of mine all the damn time. Apex legends about to win a big match. A bit yoo much going on with het another 3rd party gank. Pewwwwwwwww. Bye bye victory. Piwered back on 10 minute ban. When i sold it nd went bavk to my old card no issues. Wasnt even of the oc switch iption as that one wad even worse lol. These are 100 % gpu fault. Rven have a good power supply with a mighty single rail yo handle oc. Id hate to think how often it would be with an average vheaper rail.
reply
Add a review, comment
Other channel videos
