High GPU (100%) usage across a range of cards but not on all systems
-
Wondering if any reports have come in on 4.37 of 99-100% GPU usage consistently on a range of Nvidia Cards…
We have two servers running with GPU’s stuck at 100% utilisation and what’s even more interesting is although the GPU is being slaughtered, the temps are excessively low on these cards for the output… 45-55C.
This isn’t happening on my computer (3070ti) but we do have the issue on our servers (1050ti, 980ti) and a couple of our clients who run a 1080 and another client running a 4090RTX. All having GPU @ 100 but very cold cards…?
I have made sure that all involved have the latest Nvidia drivers installed and we also did a comparison utilising the TE Benchmark in 4.36 and 4.37. In 4.36 All servers and clients did not experience 100% utilisation and ran as normal. Then in 4.37 were back to 100% utilisation. Although running cooler and with the performance increase as advertised with the new version.
-
@unleashedcode what tool is reporting the % utilization?
I notice Windows 10 taskmgr doesn’t seem to realistically depict the gpu headroom… I have a 3060 Ti and unless foul weather or busy airport, running 4k @60hz isn’t enough for the fan to kick on. It stays in its lowest frequency…
TaskMgr reports ~99% utilization… but other tools like Nvidia FrameView report ~30-40% utilization, as the frequency fluctuates based on load.
-
@airtex2019 Just used the standard windows task manager. Although, as mentioned above in 4.36 tests was reporting normal readouts… only in 4.37 is it displaying 100%, (the TE Benchmark is in horrible weather and is designed as a stress test) but yes agreed I wouldn’t bank too much on windows task manager… although the varying results do have us scratching our heads
-
@unleashedcode I just double-checked and … it’s complicated? At least on my system.
Test: dropping into to the ILS Foul Weather Landing TR…
fps cap of 30 => FrameView and TaskMgr reported same utilization, about 65%. GPU frequency was 780 MHz.
fps cap of 60 => the numbers diverged … FrameView=80% at 1305 MHz; TaskMgr=99%
On a “work per frame” basis the NVidia FrameView numbers balance (80x1305/60 =~ 65x780/30) so that metric makes sense to me.
But I can’t imagine how or why TaskMgr sometimes aligns to that but doesn’t when the gpu clock frequency increases.
-
A lot of things changed between 4.36 and 4.37 in terms of multi-threading and GPU optimization…
Not sure what to do with your report, it is creating issues in BMS? OS becomes unstable?
Just be aware that in the future we will push the CPU / GPU optim even further…
-
@MaxWaldorf Not really any problem and not caused stability issues as I know of yet… but something the dev team would show interest in. Again, not happening on all systems and I am fully aware of the increased requirements of 4.37 and that 4.38 will be more again. I am pretty sure that although BMS requires a little extra… running a 4090 at max gpu usage is a bit of an over-reach, also against the fact that a 3070ti is running at 30-40% usage in BMS. The problem could lie in a windows scheduling or other service or even Nvidia with their drivers or even elsewhere from that… just reporting that it’s not a one off situation.
Just bringing it to the attention of the devs… it may prove useful or not… I was asked kindly by a dev to post the issue here regardless, so I did…
-
I think the resolution here is trust FrameView, don’t trust TaskMgr. And in either case the “% utilization” number has to be interpreted in context of how fast the gpu clock is running relative to its maximum.
(And obviously things like fps-cap, v-sync/triple-buffer and display resolutions, windowed vs fullscreen, all the in-game and cfg graphics options all need to be the same on these various PCs, or it’s a big bag of apples and oranges. I think you know that – just stating the obvious for the record.)
-
I don’t trust windows task manager…
Anydesk also puts the GPU to 100% and is not gpu intensive…
There is not much we are going to do anyway (not worth spending time on it if no real issue behind)…
Cheers
-
-
Guys, the most reliable tool for tracking GPU utilization is Nvidia SMI and is engaged through command line. Just open CMD and launch “nvidia-smi -l” You should get a read of utilization and VRAM, Core temp and fan utilization constantly.