EVGA

Answered[ISSUE] New card from RMA - odd voltage curve

Author
chaosminionx
Superclocked Member
  • Total Posts : 169
  • Reward points : 0
  • Joined: 2016/07/04 19:45:05
  • Status: offline
  • Ribbons : 3
2016/10/13 18:26:06 (permalink)
I just got a new card via RMA, and it appears the new card has an issue with its voltages, it looks like it has a much funkier voltage setting than my previous card, is this a bug, or intended? 

Notes:

Full DDU cleanup, and reinstall of 372.90 was done.
Brand new Windows 10 install, with all drivers.
Issue is isolated to video cards, nothing else is having a problem.
SLI enabled, EVGA HB SLI (also tested with EVGA V2 bridge, and the normal bridge included with mobo)
Both cards on 130% power/92* BIOS

Up first is the idle settings in XOC:


GPU's put under load, in the screens you can see the one GPU has a PerfCap due to VRel - this is the "primary" card, meaning the DP cable is in this card to monitors:

 
GPU's under load, this is for the secondary card (slave card, no DP cable in it):


And now the oddest part, the voltage curves are HUGELY different --- the one card has a super short voltage curve, while the other has the typical curve length I have seen on 2 other 1080FTW (I have had 4 cards in total):

 
To recap:

With this new card it looks like it applies more voltage than the slave card, yet it does not clock up. My previous card that this one replaced never gave a VRel perfcap, and both cards would run the clocks applied to it. Previously I ran +130 power target, 92* temp, +130 gpu clock, +500 memory clock, and these settings would be applied to both GPUs resulting in a 2151MHz clock on both, and it was rock solid. With the introduction of this replacement card, that has the funky voltage curve I can apply the exact same settings, and I am getting a PerfCap reason of VRel, and it clocks to 2126MHz.

Is this just something I need to play with now given this super odd voltage curve? I have never seen one so short before, talking about the amount of steps in the voltage curve window.
post edited by chaosminionx - 2016/10/13 18:44:35
#1
Sajin
EVGA Forum Moderator
  • Total Posts : 49227
  • Reward points : 0
  • Joined: 2010/06/07 21:11:51
  • Location: Texas, USA.
  • Status: offline
  • Ribbons : 199
Re: [ISSUE] New card from RMA - PerfCap/Vrel issue 2016/10/13 18:36:48 (permalink)
Vrel by itself isn't a problem.
#2
chaosminionx
Superclocked Member
  • Total Posts : 169
  • Reward points : 0
  • Joined: 2016/07/04 19:45:05
  • Status: offline
  • Ribbons : 3
Re: [ISSUE] New card from RMA - PerfCap/Vrel issue 2016/10/13 18:41:29 (permalink)
Well, I guess I should have prefaced my post better. I know the VRel flag is probably because the card is going to 1093mV and that's probably peak voltage for that card. I guess my problem is how the voltage curves look; the one is more elongated and the other is short, which I think is causing the disparity between what I think should be happening, and what I see happening.

Any ideas on the short voltage curve, is that normal or is it bugged for some reason?
#3
Sajin
EVGA Forum Moderator
  • Total Posts : 49227
  • Reward points : 0
  • Joined: 2010/06/07 21:11:51
  • Location: Texas, USA.
  • Status: offline
  • Ribbons : 199
Re: [ISSUE] New card from RMA - PerfCap/Vrel issue 2016/10/13 18:44:21 (permalink)
I'd say it's abnormal for sure. 
#4
AHowes
CLASSIFIED ULTRA Member
  • Total Posts : 6502
  • Reward points : 0
  • Joined: 2005/09/20 15:38:10
  • Location: Macomb MI
  • Status: offline
  • Ribbons : 27
Re: [ISSUE] New card from RMA - PerfCap/Vrel issue 2016/10/13 18:47:37 (permalink)
Lost.. I didn't know precision X gave a custom voltage curve sorced from the chip.. did you run evga program whatever it's called to get that curve?

And I wouldn't doubt if the 2nd card on precision X is really the first card.. I feel the same about mine as the 2nd card runs 5-7c hotter then the first and I believe the first card works harder so it should be hotter then the 2 and there both under water.

Intel i9 9900K @ 5.2Ghz Single HUGE Custom Water Loop.
Asus Z390 ROG Extreme XI MB
G.Skill Trident Z 32GB (4x8GB) 4266MHz DDR4 
EVGA 2080ti K|NGP|N w/ Hydro Copper block.  
34" Dell Alienware AW3418DW 1440 Ultra Wide GSync Monitor
Thermaltake Core P7 Modded w/ 2x EK Dual D5 pump top,2 x EK XE 480 2X 360 rads.1 Corsair 520 Rad.
#5
ty_ger07
Insert Custom Title Here
  • Total Posts : 16602
  • Reward points : 0
  • Joined: 2008/04/10 23:48:15
  • Location: traveler
  • Status: offline
  • Ribbons : 271
Re: [ISSUE] New card from RMA - PerfCap/Vrel issue 2016/10/13 18:47:54 (permalink)
Looks normal to me.  It looks like the new card you got just needs about 30mV more for stability and therefore it provides you a slightly lower boosted overclock.

ASRock Z77 • Intel Core i7 3770K • EVGA GTX 1080 • Samsung 850 Pro • Seasonic PRIME 600W Titanium

#6
chaosminionx
Superclocked Member
  • Total Posts : 169
  • Reward points : 0
  • Joined: 2016/07/04 19:45:05
  • Status: offline
  • Ribbons : 3
Re: [ISSUE] New card from RMA - PerfCap/Vrel issue 2016/10/13 18:50:19 (permalink)
So I guess what I need to do is work out clocking then manually, guess I need to read more on this voltage curve stuff, it's not quite working for me haha
#7
AHowes
CLASSIFIED ULTRA Member
  • Total Posts : 6502
  • Reward points : 0
  • Joined: 2005/09/20 15:38:10
  • Location: Macomb MI
  • Status: offline
  • Ribbons : 27
Re: [ISSUE] New card from RMA - PerfCap/Vrel issue 2016/10/13 18:53:37 (permalink)
You can run them both at different voltages.

Intel i9 9900K @ 5.2Ghz Single HUGE Custom Water Loop.
Asus Z390 ROG Extreme XI MB
G.Skill Trident Z 32GB (4x8GB) 4266MHz DDR4 
EVGA 2080ti K|NGP|N w/ Hydro Copper block.  
34" Dell Alienware AW3418DW 1440 Ultra Wide GSync Monitor
Thermaltake Core P7 Modded w/ 2x EK Dual D5 pump top,2 x EK XE 480 2X 360 rads.1 Corsair 520 Rad.
#8
chaosminionx
Superclocked Member
  • Total Posts : 169
  • Reward points : 0
  • Joined: 2016/07/04 19:45:05
  • Status: offline
  • Ribbons : 3
Re: [ISSUE] New card from RMA - PerfCap/Vrel issue 2016/10/13 18:55:24 (permalink)
Yeah, I just was hoping just setting generic offset would work like it did :)
#9
Sajin
EVGA Forum Moderator
  • Total Posts : 49227
  • Reward points : 0
  • Joined: 2010/06/07 21:11:51
  • Location: Texas, USA.
  • Status: offline
  • Ribbons : 199
Re: [ISSUE] New card from RMA - PerfCap/Vrel issue 2016/10/13 19:16:00 (permalink)
ty_ger07
Looks normal to me.  It looks like the new card you got just needs about 30mV more for stability and therefore it provides you a slightly lower boosted overclock.


Too bad you can't add more mV to the curve.
#10
chaosminionx
Superclocked Member
  • Total Posts : 169
  • Reward points : 0
  • Joined: 2016/07/04 19:45:05
  • Status: offline
  • Ribbons : 3
Re: [ISSUE] New card from RMA - PerfCap/Vrel issue 2016/10/13 19:30:18 (permalink)
I guess I don't understand this voltage curve screen fully... Where are you seeing that it needed 30mV more for stability? If I can clear up the confusion on how the offset works, I might be okay... I for some reason am.completely lost when it comes to that curve haha
#11
ty_ger07
Insert Custom Title Here
  • Total Posts : 16602
  • Reward points : 0
  • Joined: 2008/04/10 23:48:15
  • Location: traveler
  • Status: offline
  • Ribbons : 271
Re: [ISSUE] New card from RMA - PerfCap/Vrel issue 2016/10/13 20:17:43 (permalink) ☼ Best Answerby chaosminionx 2016/10/14 12:21:22
I don't know for sure either since I still own GTX 580s which aren't even GPU Boost 1.0 cards much less GPU Boost 3.0 cards.  I do own a GPU Boosted (hacked) GTX 680M in my laptop, so I sort of have an idea of how it works.
 
Below is my interpretation.  It appears that my guess of "about 30 mV" wasn't completely accurate.  It looks like the primary card needs about 43 mV more at base clock frequency.
 
This is an approximation based on an approximation.  The graph is just approximated points and therefore the numbers below aren't 100% correct but hopefully illustrate the difference between the two.  I don't even know if I am reading all the points of the graph properly (blue bars versus yellow line), but at least it should illustrate the difference between the two.  If nothing else, it should show the correct proportion of difference or ratio between the two even if the actual numbers are not trustworthy.
 
My conclusion is that the primary card needs approximately 31 to 43 mV more than the secondary card to maintain just plain base clock frequency, boosts up to max of about 27 MHz less than the secondary card, and reaches max reliability voltage about 6 mV sooner than the secondary card.  Assuming that these estimates are true or at least in line with reality, I would conclude that the new card you received just happens to not be as high up on the silicon lottery winner's list.
 

http://img.techpowerup.org/161014/6th91qt.png
 
 
These are the calculations I used based on approximations shown on the graphs (assuming that the graphs are accurate and assuming that I am reading the graphs correctly):
 
 
Min base voltage of primary card:
491 pixel postion - 247 pixel postion = 244 pixels
1162 mV - 881 mV = 281 mV spread
281 mV / 244 pixels = 1.15 mV per pixel
257 pixel postion - 247 pixel position = 10 pixels
10 pixels * 1.15 mV per pixel = 11.5 mV more than 881 mV = 893 mV

Min base voltage of secondary card:
No further calculation needed because it lines up with the graph.
850 mV

Min base voltage difference:
Primary card needs 43 mV more than secondary card at minimum boost.

Max boost voltage of primary card:
499 pixel postion - 491 pixel position = 8 pixels
8 pixels * 1.15 mV per pixel = 9.2 mV more than 1162 mV = 1171 mV

Max boost voltage of secondary card:
1243 pixel position - 978 pixel position = 265 pixels
1162 mV - 850 mV = 312 mV spread
312 mV / 265 pixels = 1.18 mV per pixel
1256 pixel position - 1243 pixel position = 13 pixels
13 pixels * 1.18 mV per pixel = 15.3 mV more than 1162 mV = 1177 mV

Max boost voltage difference:
Secondary card may be allowed to use 6 mV more than primary card at max boost.

Primary card min base frequency:
286 pixel position - 151 pixel position = 135 pixels
2.5 GHz - 1.6 GHz = 900 MHz
900 MHz / 135 pixels = 6.7 MHz per pixel
286 pixel postion - 283 pixel position = 3 pixels
5 pixels * 6.7 MHz per pixel = 20.1 MHz more than 1.6 GHz = 1.62 GHz

Secondard card min base frequency:
Same as primary card.
1.62 GHz

Min base frequency difference:
No difference.

Primary card boost frequency at 1171 mV:
286 pixel position - 239 pixel position = 47 pixels
47 pixels * 6.7 MHz per pixel = 314.9 MHz more than 1.6 GHz = 1.91 GHz

Secondary card boost frequency at 1177 mV:
286 pixel position - 235 pixel position = 51 pixels
51 pixels * 6.7 MHz per pixel = 341.7 MHz more than 1.6 GHz = 1.94 GHz

Boost frequency difference:
Primary card boosts 27 MHz less than secondary card at roughtly the same max default core voltage.
post edited by ty_ger07 - 2016/10/14 05:13:33

ASRock Z77 • Intel Core i7 3770K • EVGA GTX 1080 • Samsung 850 Pro • Seasonic PRIME 600W Titanium

#12
chaosminionx
Superclocked Member
  • Total Posts : 169
  • Reward points : 0
  • Joined: 2016/07/04 19:45:05
  • Status: offline
  • Ribbons : 3
Re: [ISSUE] New card from RMA - PerfCap/Vrel issue 2016/10/14 09:15:25 (permalink)
ty_ger07
.....

 
Fantastic post, I am going to try my hand at better manual settings, now that the explanation you gave makes sense to me as to what is going on - very detailed, I will report back with what I get once I get some bench time with it. I appreciate you taking the time to do all that...
#13
Jump to:
  • Back to Mobile