Question Zotac 1080ti Amp! Extreme worse performance after paste change

Jun 6, 2019
6
0
10
Hey there! I recently changed the stock paste (tg kryonaut) of the 1080 ti and switched the rubber pad with a thermal pad (it's quite a bad one have to admit) like gamers nexus recommended in his review.
The thing is that now it runs very cool at 70 °c max at 2000 mhz but for some reason the passmark benchmark got worse. Before the mod I had usually 14500 to 15000 but now it's about 1000 points lower at better core clocks and temps.
Furthermore it seems to me that the gpu is still getting very hot when I touch it.
What could be the issue here?
 

Eximo

Titan
Ambassador
Did you increase the core clock because it is running cooler now?

That can actually be a mistake with Pascal unless you can keep it cool through the entire frequency range. You might get more consistent output if you reduce the clock speed.

If it was left at stock, it should have stayed boosted longer with the same settings and possibly increase the score. Adding more offset will let it run faster, but then it might hit a thermal or power ceiling and have to back down.
 
  • Like
Reactions: ezπz
Jun 6, 2019
6
0
10
Did you increase the core clock because it is running cooler now?

That can actually be a mistake with Pascal unless you can keep it cool through the entire frequency range. You might get more consistent output if you reduce the clock speed.

If it was left at stock, it should have stayed boosted longer with the same settings and possibly increase the score. Adding more offset will let it run faster, but then it might hit a thermal or power ceiling and have to back down.

Yeah I left the settings at stock and it boosted better. When I ocd the core clock by about 50 mhz I was at my old benchmark results but I had to also oc my r7 1700x to 4Ghz. I will try to test this because in games the performance didn't change much I think.
So you would suggest downclocking the chip a bit and playing with power control? But why is the card performing worse with better thermals?
 

Eximo

Titan
Ambassador
I don't have the numbers in front of me to say, but generally, clock speed increase equals an increase in power consumption. That will lead to more heat. By making the chip cooler, it can run harder, and no matter how good the heat dissipation that means the chip has to handle that power draw if only for milliseconds.

GPU Boost is an algorithm, and it does its best to get maximum performance out of a set of known values. You changed one of its expectations and the effect may be an overall reduction in performance as it overshoots the silicons capabilities.

Benchmarks are also hard hitting tests. So if it can boost to 2000Mhz during the light parts, and then overheats even slightly during the harder parts it will down clock briefly, perhaps below a certain figure. If reduce the offset you may get a more consistent result. Best to track it during a bench and see what the clock speed does. Then reduce it and test again to see if the overall average is higher.

I didn't see a mention of memory overclocking. Did you do any? You can get quite a decent improvement from overclocking the memory a little. I have one of the early 1080s, so the memory could only go a little above stock, I think I have 189Mhz offset or something like that. 10388Mhz comes to mind for effective frequency.
 
  • Like
Reactions: ezπz
Jun 6, 2019
6
0
10
I don't have the numbers in front of me to say, but generally, clock speed increase equals an increase in power consumption. That will lead to more heat. By making the chip cooler, it can run harder, and no matter how good the heat dissipation that means the chip has to handle that power draw if only for milliseconds.

GPU Boost is an algorithm, and it does its best to get maximum performance out of a set of known values. You changed one of its expectations and the effect may be an overall reduction in performance as it overshoots the silicons capabilities.

Benchmarks are also hard hitting tests. So if it can boost to 2000Mhz during the light parts, and then overheats even slightly during the harder parts it will down clock briefly, perhaps below a certain figure. If reduce the offset you may get a more consistent result. Best to track it during a bench and see what the clock speed does. Then reduce it and test again to see if the overall average is higher.

I didn't see a mention of memory overclocking. Did you do any? You can get quite a decent improvement from overclocking the memory a little. I have one of the early 1080s, so the memory could only go a little above stock, I think I have 189Mhz offset or something like that. 10388Mhz comes to mind for effective frequency.
Thank you very much. It's strange because the frequency in msi afterburner is displayed being constantly on 2025 mhz. The memory is clocked at 12k mhz so this shouldn't be the issue. The temps never hit 70 c. Also interesting is that power draw reaches just 75%.
 
Jun 6, 2019
6
0
10
I think I found the problem... Msi after burner is showing me voltage limit 1 during the tests. It says temp limit 0, power limit 0 but voltage limit 1. So it's not getting enaugh voltage? How can it be fixed?
 

Eximo

Titan
Ambassador
With BIOS modding or shunt modding. Not something I have ever attempted.

Nvidia has put a hard lock on their GPUs, I think max you can do with most cards is 115%, some will do 125%.

2025Mhz constant isn't really all that bad for 1080Ti.
 
  • Like
Reactions: ezπz
Jun 6, 2019
6
0
10
With BIOS modding or shunt modding. Not something I have ever attempted.

Nvidia has put a hard lock on their GPUs, I think max you can do with most cards is 115%, some will do 125%.

2025Mhz constant isn't really all that bad for 1080Ti.
OK didn't know that but 2025 is quite good. But the thing is before that the card got much hotter with only 1950 total 1970 in boost and the results were better. Funny thing is my card won't hit 60 c during the benchmark tests something is definitely off. It's 70 c in games.
 

Eximo

Titan
Ambassador
Hmm, not sure. It could be a coincidence with some windows updates that slowed things down. They've been pushing out microcode updates to mitigate some of the side channel attack vulnerabilities. And there was that big one related to remote desktop last month, that even went out to Windows XP.

You could try a clean install of the drivers maybe, or test out an older driver revision and see if that helps.

As an experiment, try setting the offset back down and seeing how it performs. I know I played with mine a bit and found it did tend to perform better by having the offset set well below instability. I recall getting my GPU up to 2145Mhz or something like that, but it was only stable about 95% of the time, settled on 2100Mhz. The smaller GPUs tended to do a little better, I would suspect due to local heating. Could be recoverable errors that it is dealing with and having to reprocess before moving on, so not dealing with that might make it perform better.
 
  • Like
Reactions: ezπz