Unusually low 3d benchmark: EVGA GTX 980 4GB

[LG-28]

- This is a regular 980, not 980ti

I've used PerformanceTest for years on countless machines, i've never encountered this particular issue.

It seems no matter what the #'s aren't coming back as they should, even though everything outside of PerformanceTest indicates things are fine.

When i'm watching the tests they seem to be running as expected (no unexpected lag, stutters)

This video card will not benchmark above ~8.5k at best, but most often comes back in the high 7's or very early 8's, nowhere near the current 11260 average for this card on the site

I'm testing the card using a Dell T3500, 12GB@1333, 5675 Xeon CPU - I've wiped and clean installed Windows (21H1) 2x, testing in between, with and without newest drivers installed directly from NVIDIA.

Notes about screen+resolution: I'm stuck on a 1280@1024 (recommended res) 15" monitor atm, i tested @1024x768 same #'s

Because of the unusual results i decided to also test using:

3dmark:

Time Spy #'s seem spot on:

Time Spy score: 4365-4379, where the average is 4215
Time Spy graphics score: 4516-4522, where the average is 4202

Fire strike #'s seem spot on:

Fire strike score: 11649-11758, where the average is 11652-11846
Fire strike graphics score: 14518-14543, where the average is 14150
Fire strike physics score: 11244-11265, where the average is 11441

Superposition score: 9352-9371 / where the range for 4GB 980 is 8289-10728, and most falling into the 9k's

Relying on these ^^ benchmark #'s is new to me - but the #'s appear to indicate there is no problem with the card. Nothing is over or underclocked here (GPU/MEM/SYS etc) - all within stock specs.

I'm watching temps, clock speeds (mem+gpu) and power utilization internally and externally when tests are running, not seeing anything unusual. The card appears to be drawing its in-spec power of up to 165W properly.

I know the T3500/5675 is a dated combo, and i plan on testing this card in a brand new machine soon, but i've never seen PerformanceTest 3d benchmark #'s *THIS* far off before.

There's quite a bit more and different t-shooting steps i've tried to figure this out, so i'm not listing everything, however i've tried things such as adjusting to high performance power management, performance oriented 3d settings in nvidia control panel, latest system bios already OK, etc etc

I keep being pushed back into the same unusual conclusion, that only according to PerformanceTest is this video card being reported as under performing.

I could entertain this being a 21H1, Windows Update, or Nvidia-pushed (via MS) problem or something which is why i wiped the machine again today and tested in PerformanceTest directly after finishing 21H1 install, no difference in #'s

The only thing that maybe seemed off is a couple tools reporting the max bus utilization during benchmarks as barely 50% - i pulled and re-confirmed the card is in fact in an x16 slot. This particular bus utilization situation (if it's actually a problem at all) is not something i've dug deep into before. (it's possible that's entirely what it should be, given the bw ceiling is very high)

[David (PassMark)]

The Xeon X5675 is a 10 year old CPU. Where as the GTX 980 is more around 6 years old. So there is a mis-match in performance. Meaning the CPU will be somewhat of a bottleneck. Also the newer DirectX10, 11 & 12 tests were designed to run at higher resolutions than 1280x1024. The result is penalised if the monitor can't run at the expected resolution for the test (1080p and 4K for DX12).

Finally, low monitor resolution reduces load on the video card and shifts load to the CPU instead. So a weak CPU has more significant impact at very low resolutions.

[LG-28]

So is this penalization related? I thought the penalty was calculated in a way to where the final # would still be comparable?

Yeah the CPU is old, but it does benchmark above avg performance, CPU cores aren't pegging (been keeping track)

Supposedly i'll be able to test this card in a brand new machine as soon as tomorrow, I'll update as i go along, thanks for the response.

[David (PassMark)]

So is this penalization related?

I suspect it is a combination of the items I detailed above.

CPU cores aren't pegging

The 3D tests don't use 12 CPU cores. So you won't see them all pegged in the 3D test.
The tests are more single / dual thread limited. So even 10 year old i3 or Pentium CPUs are better than this Xeon for single threaded performance.
Most Xeons aren't great gaming CPUs.

[LG-28]

Update: Dropped card into a new AMD build and the 980 is now consistently scoring 11.9k+

[David (PassMark)]

Slightly above average now.

Thanks for the follow up.