One of my two GPUs is not budging during benchmark test

[brad3378]

I have a very old computer with dual GTX 770M GPUs.
I'm using 3DFlow Aerial, so both should be active.

While running the benchmark, I noticed that the Windows 10 Task Manager is showing GPU-0 with only about 25% activity and it is not showing any activity on GPU-1.
Does this indicate a problem with my computer? Is GPU-1 dead?

Here are my benchmark results:

https://benchmark.3dflow.net/226a680bcb794b5826c3ed3940dc5b25

I'm also curious about the poor results for the Mesh & Texture scores. Is this a weakness of my SSD, GPUs or CPU?

I am already shopping for a new desktop replacement style laptop computer with a budget of about $5000, but before I make my decision, I am trying to determine my current strengths and weaknesses because even a relatively small jump from a mobile 2070 to a mobile 2080 GPU is about a $500 upgrade.

[Andrea Alessi]

Hi Brad!

the windows task manager isn't the best to be honest, so i wouldn't rely on it too much. I would use a different third party software or the in-zephyr tool. However, it may simply be that it has been de-activated in zephyr. Make sure both cards are actually shown in the application CUDA settings and enabled (if you click on the icon, it will toggle the use of any cards). It should look like this:



If indeed it looks like this, if you want to be 100% sure the card is used, you should check the log. It should look like this (in this example, a 3 card setup is shown)

[12:01:10] [CPU 0] [DJI_0002.JPG] Loading image...
[12:01:10] [CPU 0] [DJI_0003.JPG] Loading image...
[12:01:10] [CPU 0] [DJI_0004.JPG] Loading image...
[12:01:10] [CPU 0] [DJI_0001.JPG] Loading image...
[12:01:14] [GPU 0] [DJI_0003.JPG] Detecting keypoints...
[12:01:14] [GPU 2] [DJI_0001.JPG] Detecting keypoints...
[12:01:14] [GPU 1] [DJI_0002.JPG] Detecting keypoints...
[12:01:14] [GPU 1] [DJI_0002.JPG] Computing descriptors...
[12:01:14] [GPU 2] [DJI_0001.JPG] Computing descriptors...
[12:01:14] [GPU 0] [DJI_0003.JPG] Computing descriptors...

If indeed you only see GPU0 and no GPU1 rows, then your card is likely not used (either the driver is not installed, or the card is disabled in windows, or has been disabled in zephyr, or you should see a lot of errors and retries on a different device)

I think in this case the poor results are due to the CPU clock being the bottleneck, but i may be wrong. Can you share the full benchmark log ? thanks!

On a side note, I think the current benchmark version hasn't been calibrated by us very well for the texturing phase (each new release we test and get timings with a reference machine, so we can retarget/calibrate scores for the changes of the new version), so i would take the texturing results with a grain of salt.

[brad3378]

Thank you Andrea.
You were right. I looked at the log and my computer is using both GPUs.
I ran the benchmark over again today and picked up 30 points on my final score for a 10% increase.

https://benchmark.3dflow.net/94618adcc5094857d0d72a4604edb66a

There are only two differences I can think of between today's benchmark and the one I ran before (on same software version). The first difference is that I followed the instructions on this video to force my GPUs into "High Performance" mode in Windows 10.

https://www.youtube.com/watch?v=EZm6VtNN0g8

I should mention that the first time I ran the benchmark with the new High Performance setting, my final benchmark score increased from 300 to 320. Today with the same settings I hit 330. The second difference I'm aware of is that I started the new benchmark today on a cold machine so maybe that explains the difference between the 320 and 330 score. I am speculating that I may be experiencing throttling issues due to temperatures. This Alienware 18 computer was purchased in 2013 and has never had the factory thermal paste removed, so I should consider fixing that even if this becomes a backup computer.

A log showing my 330 score is attached to this message.

----
UPDATE: I just reran the benchmark again and with zero changes my final results leaped to 392! I have no explanation for the difference.
https://benchmark.3dflow.net/67b3034...09a092bebeae4c

[Andrea Alessi]

Happy to help!

i wouldn't worry too much about a few points - the score is calculated by comparing the computaton times of your computer to a test-run we did on our workstation, and slightly adjusted. Since we cannot expect the users to wait hours, we had to settle with a very small dataset, which means that random slowdowns (e.g. windows being windows and deciding to do stuff because reasons) of a few seconds can reflect to a slightly lower score. My guess is that your first run maybe you had something in background (windows, antivirus, who knows) so i wouldn't worry too much about it and keep that in mind when comparing with other results

If you are struggling in choosing the hardware of your new workstation, feel free to post the configuration here and I'll be more than happy to comment on it