Why is CPU use much lower on one of two RTX 2060 cards?

[markdotgooley]

First card is an RTX 2060KO. Second is a regular RTX 2060.

First: GPU utilization usually around 95%, uses usually around 30% or less of 8.0 GT/s bandwidth on PCI link, runs around 70-75C.

Second: GPU utilization usually around 75% though occasionally over 90%, uses typically around 75% or more of 5.0 GT/s bandwidth, runs cooler at around 65C but it’s positioned lower in the case too.

Why is utilization usually lower on card 2? Is the lower bandwidth slowing it down? Information is from NVIDIA X Server Settings display on my Ubuntu box, which seems about the only source I can find.

[Rel25917]

May just be getting smaller units on the second card(low atom count) smaller unit cant use up a whole high end card. Could be bandwidth but linux is much better than windows on that.

[PantherX]

I am not sure if you mean CPU usage (topic title) or GPU usage (post contents).

As you have Nvidia GPUs, each GPU will use 1 CPU so you should see 2 CPUs pegged at around 100% while the GPU is folding. This is true on Windows and Linux. Any CPU cycles "stolen" from it may negatively impact GPU performance.

Some GPU Projects are different than others. Are you comparing the same Project across both? Also,is it PCIe x2, PCIe x3, PCIe x4 and what speeds are they operating at?

[markdotgooley]

GPU Utilization is what the "NVIDIA X Server Settings" application on my Ubuntu machine calls it. I have a 6-core Ryzen 5 1600 AF that I keep running 10 threads of F@H, with 2 threads to tend to the GPUs. Maybe I should stop folding on the CPU entirely? The application says that the first (near-fully utilized) GPU is on "Bus Type: PCI Express x16 Gen3" and the other is on "Bus Type: PCI Express x16 Gen2". I should figure out how to snapshot what the application displays.

[PantherX]

As a test, I would suggest that you set the CPU Slot to Finish and then see what the GPU utilization is.

I am guessing that the configuration is as follows:
PCIe 3 x16 -> RTX 2060 KO
PCIe 2 x16 -> RTX 2060

From what I have read about Linux, both should perform at their full potential without issues as overhead in Linux is measurably less when compared to Windows.

[markdotgooley]

As a test, I would suggest that you set the CPU Slot to Finish and then see what the GPU utilization is.

I am guessing that the configuration is as follows:
PCIe 3 x16 -> RTX 2060 KO
PCIe 2 x16 -> RTX 2060

From what I have read about Linux, both should perform at their full potential without issues as overhead in Linux is measurably less when compared to Windows.

Will give it a try. My guess is that they should both work alike and that the difference in bandwidth isn't big enough to matter. Maybe I should ask elsewhere as well...

[PantherX]

There is this massive topic (viewtopic.php?f=38&t=28847) and a short topic (viewtopic.php?f=38&t=34932) that provides lots of data points regarding folding and PCIe performance. Feel free to skim read them to see if they match your experience or not.

[MeeLee]

As a test, I would suggest that you set the CPU Slot to Finish and then see what the GPU utilization is.

I am guessing that the configuration is as follows:
PCIe 3 x16 -> RTX 2060 KO
PCIe 2 x16 -> RTX 2060

From what I have read about Linux, both should perform at their full potential without issues as overhead in Linux is measurably less when compared to Windows.

most motherboards have an x16 slot, but only offer x16 speeds to 1 slot.
Adding a second slot, most surely halves the speed of the first slot, and the second slot can be anywhere from x8, x4, or x1 speeds, even on an x16 slot on the motherboard.

The 2060KO is a higher end model with some disabled cores.
By default it's about 13% faster than the regular 2060.
Some models are MUCH faster (based on failed 2080 / 2080 Super cores), some are only marginally faster (based on 2070 / 2070 Super cores).
But either way, the KO is faster than a regular 2060.

[markdotgooley]

Will have a look. Thanks!