Question Issues with testing EVGA GTX 1080 SC ACX 3.0

JWMiddleton

Diamond Member
Aug 10, 2000
5,582
100
106
I recently (ASSUMED) I bought an EVGA GTX 1080 SC ACX 3.0 from a TeAm member. I was previously running a GTX 1660 Super. I simply took out 1660 and put in the 1080. It ran fine and games worked. When I would wake up the monitor after it went to sleep, the screen would come on, then flash. No issues until the next sleep session.

I wanted to see the improvement with the 1080, so I ran a few benchmarks. Heaven and Valley did great, but Userbenchmark and 3DMark Time Spy (DX 12) gave me errors as neither of them would give me a score. Time Spy would not complete a run and said "an error has occurred." Userbenchmark indicated an issue with the GPU, but didn't say what it was. I tried 3D Mark Fire Strike (DX 11) and it ran fine.

In talking with the seller I used DDU to uninstall the driver and reinstalled latest one. The screen no longer flashed, but same issue with Userbenchmark and Time Spy. It is not a big deal that they don't work, but I would like to see if I can get it resolved.

I wondered if what I'm seeing in GPU-Z has anything to do with it. The info displayed looks correct, as you can see in the attachment, and when I click Lookup, it brings up a later version of the card with Pascal a 104-410-A1 chip vs the older 104-400-A1 version. So, Lookup shows an EVGA GTX 1080 SC2 w/ iCX Cooler 11Gbps. Turns out that I ASSUMED incorrectly. :(

System in Sig is up to date. I do not have any OC utilities installed. So, any ideas?

John

GPU-Z GTX1080.jpg
 
Last edited:

maluckey1

Senior member
Mar 15, 2018
331
144
86
Perhaps the previous user re-flashed the BIOS? I sometimes do that for certain features or different voltage limits. You could flash the card to the most current version for your card and see if that fixes it.
 
  • Like
Reactions: JWMiddleton

JWMiddleton

Diamond Member
Aug 10, 2000
5,582
100
106
Perhaps the previous user re-flashed the BIOS?
As I was typing the post the same thing came to mind. So, I searched for BIOS updates for the card and found links in EVGA forums. I was instructed to input the serial number, which I got off the box. I got a msg saying that the number was not found, so I checked it again and took out the card to make sure it matched what was on the box. It did, BUT on the front of the card it said iCX. So, GPU-Z was correct. Eh, not so fast....

This is driving my 75 year old brain crazy! When I first looked up "Evga GTX 1080 SC", which is what was in the post for sale. I got the info that it was the ACX 3.0 version. The iCX version shows up as SC2 in MOST posts. When I look up the part number the EVGA website has it as EVGA GeForce GTX 1080 SC GAMING 11Gbps, which is what is on the box. Well, the box does not have SC anywhere on it. The box does have a 2017 copyright date on it, which is when the SC2 was released (Apr. 27, 2017 vs May 27, 2016.) On the very top of the card it does have SC in lights, funny that the box doesn't mention it.

So, the mystery continues and I'm still looking to see if there is a BIOS update available.

John

iCX.jpg
 
Last edited:

solidsnake1298

Senior member
Aug 7, 2009
206
115
116
3DMark tracks GPU temps and clocks after a run completes and displays it in a graph at the bottom of the page with your score. I would be interested in seeing what that temp/clock speed graph looks like. I don't remember if it also displays the temp/clock graph after a failed run.
 

JWMiddleton

Diamond Member
Aug 10, 2000
5,582
100
106
3DMark tracks GPU temps and clocks after a run completes and displays it in a graph at the bottom of the page with your score. I would be interested in seeing what that temp/clock speed graph looks like. I don't remember if it also displays the temp/clock graph after a failed run.
Yep, I did capture that. I hope this screen shot is enough. I've tried to run it again through Stream and it won't load, sits at 0% downloading.

3DMark Error Occurred.jpg
 
Last edited:

solidsnake1298

Senior member
Aug 7, 2009
206
115
116
Hmmmm. The lines for GPU temp and GPU load aren't there. The GPU clocks look fine, but they appear to be base and not boost. My EVGA 1070 ACX 3 boosts to over 2000Mhz. Does Time Spy even run? Or just immediately fail? Based on the GPU not boosting and the short run time, I am assuming it immediately fails.

To me, this feels like a not quite stable OC. Perhaps, as has been suggested, someone flashed it with higher boost tables? Maybe download EVGA Precision X1 and underclock the mem and GPU just a little?
 

JWMiddleton

Diamond Member
Aug 10, 2000
5,582
100
106
The lines for GPU temp and GPU load aren't there.
I originally saved the results as a PDF file, but can't attach that on this forum. So, I did a screen shot of what I could see. I then deleted the PDF file and emptied the trash.

I can load and run Fire Strike, but Time Spy will not install again. Here is what I got from Fire Strike. I am using the Demo version. Seems that it also had a problem as it says it load all SystemInfo data.

FireStrike_GTX1080.png
 

solidsnake1298

Senior member
Aug 7, 2009
206
115
116
Yeaaah. Something weird is going on. I have a full copy of 3DMark and can search their database of user scores. A 1080 should have a graphics score in the 25-26k range. Between Time Spy crashing and the low-ish score in Fire Strike, it sounds like memory stability issues.
 

solidsnake1298

Senior member
Aug 7, 2009
206
115
116
So I searched for your part number, as indicated by the "6283" at the end of the GPU-Z device ID. This came up. It is an SC card equipped with EVGA's iCX cooler. Hence both the SC and iCX branding.


I searched EVGA's forum for BIOS updates and what the latest version is and found that the version you have installed, 86.04.66.00.80, is the latest and only version. In late 2016 EVGA issued an update to various 1060, 1070, and 1080 models to adjust fan curves only. But there was no update issued for "6283". My guess is that your 1080 variant was released after that BIOS update was issued.

As far as I can tell, the only difference between the GP104-400 and 410 is that the 410 uses a 11Gbps GDDR5X memory controller whereas the 400 uses a 10Gbps GDDR5X memory controller.

I will run Unigen when I get home and eat dinner (sometime after 7pm EDT).

Again, I am curious if underclocking your GPU and memory a little bit would help.
 

JWMiddleton

Diamond Member
Aug 10, 2000
5,582
100
106
OK, who says you can't teach an old dog new tricks! :) Well, I ran Precision X1 utility, but don't know what to do with the scan results of +80.

Also, DL'ed OCCT and ran both 3D and VRAM tests. They seem fine. Although I only ran them for 2 minutes each.

Precision X1 VF Curve Tuner.jpgOCCT v9.0 3D Test.jpgOCCT v9.0 VRAM Test.jpg
 

solidsnake1298

Senior member
Aug 7, 2009
206
115
116
The OCCT tests look fine. GPU temps look fine.

Set the GPU and Mem clocks to -50 as shown below. Set the power target to its highest. Set the voltage to its highest (though it doesn't technically do anything for Pascal). Apply.

Precision X1 VF Curve Tuner.jpg
 

JWMiddleton

Diamond Member
Aug 10, 2000
5,582
100
106
Set the GPU and Mem clocks to -50 as shown below. Set the power target to its highest. Set the voltage to its highest (though it doesn't technically do anything for Pascal). Apply.
I did all that and re-ran Valley. Frame rate dropped by 4.8 fps. I pulled up Precission X1 and stated noticing screen flickering and lines running through it. I reset GPU to default and rebooted. Screen is back to normal. I realize that this is a 4 years old card. Should it be doing this? Should I return it?

John
 

Furious_Styles

Senior member
Jan 17, 2019
447
151
86
I did all that and re-ran Valley. Frame rate dropped by 4.8 fps. I pulled up Precission X1 and stated noticing screen flickering and lines running through it. I reset GPU to default and rebooted. Screen is back to normal. I realize that this is a 4 years old card. Should it be doing this? Should I return it?

John
Try taking it out and cleaning the contacts and seeing if that helps.
 

solidsnake1298

Senior member
Aug 7, 2009
206
115
116
My 1070 is older than yours (purchased late 2016) and works fine while your 1080 variant was first manufactured in early 2017. Your card should NOT be doing this.

Ask the seller if they've ever re-pasted or changed thermal pads, etc. There is no consistency in this card's behavior. Why would reducing the boost clocks cause green lines on the screen? Perhaps removing the heat sink and back plate is in order. Give it a physical inspection? Post pictures of the bare board, front and back?
 

maluckey1

Senior member
Mar 15, 2018
331
144
86
Hmmm, have you looked in OCCT to see what the temperature of the GPU hottest core is during a bench run? This sounds similar to what my old card did when it was overheating.

-Have you checked to see if the previous owner maybe had a different cooler, and put this one back on with minimal/crappy thermal paste?

-Lastly, for the 8 pin (2 x 4) power-in, have you tried 2 cables instead of 1 daisy-chained cable? Not sure of the amperage draw, but on a marginal PSU (or older PSU), it could certainly make a difference

M
 

JWMiddleton

Diamond Member
Aug 10, 2000
5,582
100
106
Hmmm, have you looked in OCCT to see what the temperature of the GPU hottest core is during a bench run? This sounds similar to what my old card did when it was overheating.

-Have you checked to see if the previous owner maybe had a different cooler, and put this one back on with minimal/crappy thermal paste?

-Lastly, for the 8 pin (2 x 4) power-in, have you tried 2 cables instead of 1 daisy-chained cable? Not sure of the amperage draw, but on a marginal PSU (or older PSU), it could certainly make a difference

M
Mine is a 6+2 for the 8 pin. I have a very new SeaSonic FOCUS PX Series 750W 80 Plus Platinum Modular ATX PSU. So, that shouldn't be an issue.

I posted some OCCT results earlier. SS said that they looked fine.

I sent the seller a link to this thread to get his take on it.

John
 

maluckey1

Senior member
Mar 15, 2018
331
144
86
Mine is a 6+2 for the 8 pin. I have a very new SeaSonic FOCUS PX Series 750W 80 Plus Platinum Modular ATX PSU. So, that shouldn't be an issue.

I posted some OCCT results earlier. SS said that they looked fine.

I sent the seller a link to this thread to get his take on it.

John
Yup, I agree, the PSU is fine. Daisy chaining is only bad on a marginal PSU.

The 1080 is fairly throttle happy with temperatures. For what you were running, the temps seem a little high, especially for such a short run. The temps would certainly be higher in games and benches than that. You can try lower the power target by 10 percent, and raise the temperature target to maximum and see if that helps. If it does, it's cooling/power related (with boost 3.0 they're inter-reliant to a degree).

There is no consistency in this card's behavior. Why would reducing the boost clocks cause green lines on the screen? Perhaps removing the heat sink and back plate is in order. Give it a physical inspection? Post pictures of the bare board, front and back?
You beat me to the question by seconds! I agree, somethings not right. If it was my rig, I'd focus next on cooling.
 

solidsnake1298

Senior member
Aug 7, 2009
206
115
116
Here is my result. I don't know of a database of Valley results to compare yours to. But your 1080 has ~25% more CUDA cores than my 1070, significantly faster GDDR5X memory, on top of you having a significantly faster CPU. So you beating me by 10-15% does not line up with my expectations of a 1080 based on its spec sheet.

1626476060926.png
 

ASK THE COMMUNITY