Overclock.net › Forums › Graphics Cards › NVIDIA › NVIDIA GTX 1080 Ti Owner's Club
New Posts  All Forums:Forum Nav:

NVIDIA GTX 1080 Ti Owner's Club - Page 702

post #7011 of 10809
Quote:
Originally Posted by viz0id View Post

Good to see that it is not just me. But you say it is a great power eater, and it shows "Pwr" on perf caps. But the TDP hardly goes over 103% so i just cant seem to wrap my head around why it does that.

I can see power hovering about 112-115% when the volts and frequency start getting unstable. At 103% they are still stable.
BTW, as far as I gathered from testing my FE it's almost impossible to get stable 2000 MHz and over on a FE in most of modern 4k games, may be except BF1 - this is the only demanding thing that allows me to see the magic stable 2GHz.
post #7012 of 10809
Quote:
Originally Posted by MRCOCOset View Post

Palit 350W Bios on ASUS GTX 1080TI FE. Work thumb.gif My video >>>

While I agree you are peaking 330w+, you are power limiting like crazy. Just look at how much its clocking down frown.gif

Show me the same test with more stable clks and GPU-Z with maxs on power. Barely limiting I pulling closer to 300w averages.
MSI
(16 items)
 
  
CPUMotherboardGraphicsRAM
i7 6700k MSI XPower Gaming Titanium Edition MSI 980Ti Lightning Gskill RipJaw V 3200 C16 
Hard DriveCoolingCoolingCooling
Samsung 950 Pro Raid0 Water cooled with 750w chiller EK mono block - MSI XPGTi Bitspower - MSI Lightning 
OSMonitorKeyboardPower
Win 10 Home ACER X34 - 3440x1440  Corsair K65 Seasonic 1050w - Silent Snow 
CaseMouseMouse PadAudio
Silverstone RV05 - White Razer Epic Naga Chroma Artizan - Hayete Otsu - Soft Corsair Void Wireless - Carbon 
  hide details  
Reply
MSI
(16 items)
 
  
CPUMotherboardGraphicsRAM
i7 6700k MSI XPower Gaming Titanium Edition MSI 980Ti Lightning Gskill RipJaw V 3200 C16 
Hard DriveCoolingCoolingCooling
Samsung 950 Pro Raid0 Water cooled with 750w chiller EK mono block - MSI XPGTi Bitspower - MSI Lightning 
OSMonitorKeyboardPower
Win 10 Home ACER X34 - 3440x1440  Corsair K65 Seasonic 1050w - Silent Snow 
CaseMouseMouse PadAudio
Silverstone RV05 - White Razer Epic Naga Chroma Artizan - Hayete Otsu - Soft Corsair Void Wireless - Carbon 
  hide details  
Reply
post #7013 of 10809
Quote:
Originally Posted by ALSTER868 View Post

I can see power hovering about 112-115% when the volts and frequency start getting unstable. At 103% they are still stable.
BTW, as far as I gathered from testing my FE it's almost impossible to get stable 2000 MHz and over on a FE in most of modern 4k games, may be except BF1 - this is the only demanding thing that allows me to see the magic stable 2GHz.

If i put on PerfCap on "MAX" the highest max TDP% i get is 104,2%, while it is bouncing around.

So i can choose to have 2063 on core and +100 mem. Or i can have 1999core and +200 on mem. I even tried to underclock and use like 1900 on core. Still can't go over +200 memory. Feel like something is really really wrong.

Might end up using my 45 day open return on the card. It has got some pretty hefty coil whine as well.
post #7014 of 10809
Quote:
Originally Posted by BrainSplatter View Post

Hitting the 'voltage limit' is normal GPU boost behaviour. Your new card just hits it more often than your old one because it has a higher VID. My assumption was, that with an identical voltage curve, the behaviour would be the same for your 2 cards. But it seems to be more complex than that. I already noticed with my 2 FE cards (their VIDs are also different) that the voltage points in the editor seems to jump around a little different for each card when I press 'apply', even when I try to use the exact same curve.

In the end, the one thing which matters is which maximum OC u get from your new card. Since that depends on silicon lottery and your new one might not get as high as your old one. But if it runs stable with something between 2000-2050 Mhz then that would be a perfectly fine result. The new card will probably run a little hotter/louder than the old one due to the higher VID.

Should get another new card, it might have an even higher VID, overclock less or maybe hit 2100Mhz with 1v wink.gif Just a matter of luck.


Your right the second card does pull a higher voltage than my first card 1.063v vs 1.050v, I haven't spent much time overclocking but it seems to be capable of 2050Mhz on Valley benchmark on the stock FE cooler. The card will be on water if I decide to keep it so temps aren't an issue.

I was really only concerned as I have had 3 different 1080Ti FE cards and this was the only one which has had constant voltage Limit warning popping up at load and no matter what I do I cant get rid of it. Temp and Power Limits are easy to negate and my thinking is so should the Voltage Limit but this doesnt seem to be the case.

Anyway I will bench this card some more and if its stable I will keep it, although I must say I do not like seeing that Voltage Limit constantly being on especially at stock.
post #7015 of 10809
Here's a similar video that shows the difference between GPU-Z and Nvidia-smi. I think even with monitoring set at 0.1sec, GPU-Z just isnt sampling the card enough, compared to the 119 samples nvsmi takes over a ~2 second period.

Of course that also means GPU-Z wont be seeing all of the perfcaps at 0.1second intervals.

The run is FS Extreme. TDP limits like crazy in the first scene, which isnt surprising given nvsmi is reading > 330W. The second scene starts around the 50sec mark. 2075Mhz @ 1.075V and plenty of peak readings > 300W from nvsmi. GPU-Z doesnt show any perfcaps but, as I suggested, that doesnt mean they are not there, it might just be missing them.
Edited by mshagg - 4/25/17 at 6:11am
Main rig
(9 items)
 
  
CPUMotherboardGraphicsRAM
i7 5930k Asus X99 Deluve Nvidia GTX 1080Ti  Corsair Dominator 3000Mhz 
Hard DriveCoolingOSPower
Intel 600p 1x420 rad, 1x480 rad, D5 Pump Windows 10 Pro x64 EVGA 1000w 
Case
Enthoo Primo 
  hide details  
Reply
Main rig
(9 items)
 
  
CPUMotherboardGraphicsRAM
i7 5930k Asus X99 Deluve Nvidia GTX 1080Ti  Corsair Dominator 3000Mhz 
Hard DriveCoolingOSPower
Intel 600p 1x420 rad, 1x480 rad, D5 Pump Windows 10 Pro x64 EVGA 1000w 
Case
Enthoo Primo 
  hide details  
Reply
post #7016 of 10809
Quote:
Originally Posted by mshagg View Post

OK so found a spot on the curve, with some mem OC, that pushes the Palit BIOS to the power limit. 2050Mhz at 1.061V, with some power limiting in those first few scenes of SuPo.

Peak TDP was 109%, which equates to 327W.

I think it's a stretch to say we're going to get 350W from it. Presumably we are running up against the sub-limit for the motherboard's power supply, given Palit would be drawing that extra power from the 2x8-pin. That said, these are far better results than I've had with any of the other BIOSes.

Thanks for looking into this
post #7017 of 10809
Quote:
Originally Posted by viz0id View Post

If i put on PerfCap on "MAX" the highest max TDP% i get is 104,2%, while it is bouncing around.

So i can choose to have 2063 on core and +100 mem. Or i can have 1999core and +200 on mem. I even tried to underclock and use like 1900 on core. Still can't go over +200 memory. Feel like something is really really wrong..

Really seems weird this kind of behaviour. I'm also hitting the pwr limits severely but not the way you are. I think it's worth of considering to return it.
post #7018 of 10809
Quote:
Originally Posted by MRCOCOset View Post

Palit 350W Bios on ASUS GTX 1080TI FE. Work thumb.gif My video >>>

Hey this is great, thank you for doing a thorough test on this. Looks like your test is in line with mshagg. The bios gives an extra 30 watts. But as other people experience, it either crashes, works but doesn't get as good of a score as other bios.

You're definitely hitting power limits though, probably holding higher clocks than stock but 30 watts isn't enough to eliminate power limiting.

So far this is the highest tdp of all the bios.
Edited by SlimJ87D - 4/25/17 at 6:32am
post #7019 of 10809
Quote:
Originally Posted by viz0id View Post

If i put on PerfCap on "MAX" the highest max TDP% i get is 104,2%, while it is bouncing around.

So i can choose to have 2063 on core and +100 mem. Or i can have 1999core and +200 on mem. I even tried to underclock and use like 1900 on core. Still can't go over +200 memory. Feel like something is really really wrong.

Might end up using my 45 day open return on the card. It has got some pretty hefty coil whine as well.

Just testing my hypothesis against these observations - in that GPU-Z, at its lowest setting, will only poll the GPU every 0.1 seconds. I believe GPU Boost is working at a far higher frequency than this - nvsmi for example, can take 119 observations over a ~2second period, about 600% more frequent than GPU-Z.

So, who really knows what is actually happening, when looking at GPU-Z's max values. It may have spiked well above 104% TDP between the 0.1 seconds it was looking.
Main rig
(9 items)
 
  
CPUMotherboardGraphicsRAM
i7 5930k Asus X99 Deluve Nvidia GTX 1080Ti  Corsair Dominator 3000Mhz 
Hard DriveCoolingOSPower
Intel 600p 1x420 rad, 1x480 rad, D5 Pump Windows 10 Pro x64 EVGA 1000w 
Case
Enthoo Primo 
  hide details  
Reply
Main rig
(9 items)
 
  
CPUMotherboardGraphicsRAM
i7 5930k Asus X99 Deluve Nvidia GTX 1080Ti  Corsair Dominator 3000Mhz 
Hard DriveCoolingOSPower
Intel 600p 1x420 rad, 1x480 rad, D5 Pump Windows 10 Pro x64 EVGA 1000w 
Case
Enthoo Primo 
  hide details  
Reply
post #7020 of 10809
Quote:
Originally Posted by mshagg View Post

Just testing my hypothesis against these observations - in that GPU-Z, at its lowest setting, will only poll the GPU every 0.1 seconds. I believe GPU Boost is working at a far higher frequency than this - nvsmi for example, can take 119 observations over a ~2second period, about 600% more frequent than GPU-Z.

So, who really knows what is actually happening, when looking at GPU-Z's max values. It may have spiked well above 104% TDP between the 0.1 seconds it was looking.

That is true. But when run long enough, would it not catch some higher values over a longer period? Eventually one would think that it would be polling the information at a higher TDP?

And if it is true like you said, and it is in fact power throttling, i find it weird that just an increase of +10 on the memory would make it go totally banans. It's like no perfcaps what so ever. Increase by +10 on the mem clock and it goes mental.

Really annoying that i cant get an understand of it.

EDIT: I can even run the card at 2088 without pwr throttle, but once the temp goes above whatever in the 30-40C range it goes down to 2075, then again down to 2062 on somewhere in the 50's. (soft temp throttle)
Edited by viz0id - 4/25/17 at 6:32am
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: NVIDIA
Overclock.net › Forums › Graphics Cards › NVIDIA › NVIDIA GTX 1080 Ti Owner's Club