Question 1080ti low utilization/scores on 5950x

[Justinus]

Since I sold my 6900xt for having awful coil whine with the waterblock installed, I migrated my old 1080ti into the loop with the 5950x.

I figured it should be about as fast or maybe even quicker due to the 5950x's superior memory speed, clocks, and IPC vs. the old 5960x.

3dmark shows a different story, though - The GPU only sits around 94-97% utilization (always 99-100% on the old 5960x box) and scores 5-6% lower at the same GPU clocks.

For reference, with the 6900XT installed with the same CPU and memory config on the 5950x, it would hit 99-100% utilization in 3dmark.

I checked the pci-e connection and it's getting a full 16x 3.0 connection. Is there any configuration change or troubleshooting steps I can try to get the card to actually hit 99-100% utilization? It seems so strange since the system was clearly capable of driving a card ~2x as fast at 100%, that it cannot drive the 1080ti at 100%.

On a side note, the 980ti I had in this same rig before the 1080ti would hit 99-100% utiliation as well. Just the 1080ti acting up for some reason I cannot figure out!

5950x stock boost (dynamic all core overclock at 4.6 ghz when heavily loaded)
Crosshair VIII Dark Hero X570
2x16gb dual rank b-die at 3800CL14 1:1
Latest driver 461.40, clean install done since 980ti swap.

 

[gk1951]

The obvious answer is the GTX1080TI is an older architecture vs the 6900XT. I don't mean to be smug but I would have put up with the coil whine until another 6900XT became available. In addition, how much actual difference is there between the 94-97% utilization vs 99-100%?

BTW you have an incredible rig!

 

[Stuka87]

Did you fully remove the AMD drivers before installing the nVidia drivers again?

It is a bit odd that the CPU would feed a much faster GPU without issue, but the older slower card isn't. So there must be something else going on. Maybe a power limitation, or a driver issue.

 

[CP5670]

It may be working correctly. The drivers don't work the same way and may run into CPU bottlenecks in different ways, or different parts of the GPU may be bottlenecked. I sometimes see GPU utilization of only 70-80% even when the framerate is not that high, especially in VR.

 

[Justinus]

Did you fully remove the AMD drivers before installing the nVidia drivers again?

It is a bit odd that the CPU would feed a much faster GPU without issue, but the older slower card isn't. So there must be something else going on. Maybe a power limitation, or a driver issue.

This is a fresh windows install that has only had the 980ti and 1080ti installed on it.

The obvious answer is the GTX1080TI is an older architecture vs the 6900XT. I don't mean to be smug but I would have put up with the coil whine until another 6900XT became available. In addition, how much actual difference is there between the 94-97% utilization vs 99-100%?

BTW you have an incredible rig!

I considered keeping the 6900xt running on the reference cooler with less whine, but instead sold it as soon as I had a buyer. I'm not sure I'm going to buy another one given the new tariff prices - if I wanted to spend $1500 on a GPU I could have bought a 3090 several times before I finally nabbed the 6900xt for $1060.

The difference appears to be around 5-6%. Just enough to definitely not be margin of error, but not enough to really be noticeable without watching framerate.

It may be working correctly. The drivers don't work the same way and may run into CPU bottlenecks in different ways, or different parts of the GPU may be bottlenecked. I sometimes see GPU utilization of only 70-80% even when the framerate is not that high, especially in VR.

I had considered that Pascal is old enough it probably doesn't get driver optimizations for newer platforms or CPUs, so perhaps it is doing something or other to hit a bottleneck that could otherwise be avoided.

 

[Mopetar]

I had considered that Pascal is old enough it probably doesn't get driver optimizations for newer platforms or CPUs, so perhaps it is doing something or other to hit a bottleneck that could otherwise be avoided.

Odd that the 980 Ti wouldn't exhibit similar behavior in that case. Do you have another rig you could put the card into just to run some tests so see if it's isolated to just your main machine?

 

[Justinus]

Odd that the 980 Ti wouldn't exhibit similar behavior in that case. Do you have another rig you could put the card into just to run some tests so see if it's isolated to just your main machine?

The card came out of the old 5960x box over the weekend where I have saved 3dmark benchmarks online from, which is what I'm comparing to.

If the 1080ti is only slightly bottlenecked, the 980ti being 40% slower wouldn't be able to exhibit the same behavior.


For the lulz I ran the aida64 GPGPU benchmark to compare to how it performed in the old box, and every single GPGPU test is 5-10%+ faster on the new system that has the 3dmark bottleneck.



You can see here on the firestrike the difference in graphics score despite the same GPU clocks.

Result (3dmark.com)

And here the timespy difference in graphics score at same GPU clocks.

Result (3dmark.com)

I have yet to have time to collect more data for comparison.

 

[alcoholbob]

You should've kept the 6900XT and sold your PSU instead. Coil whine while gaming is due to the interaction between your power supply and the graphics card. It would be much easier to change out the cheaper part and find a PSU that doesn't cause coil whine on your graphics card than the other way around.

 

[Justinus]

You should've kept the 6900XT and sold your PSU instead. Coil whine while gaming is due to the interaction between your power supply and the graphics card. It would be much easier to change out the cheaper part and find a PSU that doesn't cause coil whine on your graphics card than the other way around.

The coil whine was not PSU related. I tested three different OEM/design units with no effect.

The whole thread is a moot point now since I nabbed a 6900xt Strix last night.

 

[Shmee]

Nice, congrats on the find!