Divergent Posts from Benchmark Thread

[storm5510]

I have an RTX 2080. What is a proper number of threads to specify?

My GPU utilization never goes beyond 2%.


Ed's Note: This and the following posts have been copied/moved from the Msieve benchmark thread since they more directly involve getting a GPU running, rather than adjusting and providing benchmark values.

[frmky]

No threads. After compiling the msieve-lacuda-nfsathome branch with the option CUDA=75, run with
-nc2 -g 0 -v
where the number after -g specifies which gpu to use. Without the -g option, it uses only the cpu.

[storm5510]

Originally Posted by frmky

No threads. After compiling the msieve-lacuda-nfsathome branch with the option CUDA=75, run with
-nc2 -g 0 -v
where the number after -g specifies which gpu to use. Without the -g option, it uses only the cpu.

OK. Well, what i have is not Linux. It's Windows 10. Thanks for the reply!

[wombatman]

Originally Posted by storm5510

OK. Well, what i have is not Linux. It's Windows 10. Thanks for the reply!

If you're willing to do a little work, Windows Subsystem for Linux 2 works with CUDA now, so you can build and run it there.

[storm5510]

Originally Posted by wombatman

If you're willing to do a little work, Windows Subsystem for Linux 2 works with CUDA now, so you can build and run it there.

I found one Windows version which will run, to a point. It says it cannot find "sort_engine.dll" in a folder called "cub." It is right there in the proper place where the program looks for it.

I have the subsystem installed. It is Ubuntu 20.04 LTS. It's been a while since I've used it.

[frmky]

Originally Posted by storm5510

I found one Windows version which will run, to a point. It says it cannot find "sort_engine.dll" in a folder called "cub." It is right there in the proper place where the program looks for it.

That's an old version that doesn't contain my work on the GPU LA. I'm not aware of anyone that's compiled a Windows version with my changes.

There's a guide to using CUDA with WSL2 at https://docs.nvidia.com/cuda/wsl-user-guide/index.html.

[storm5510]

Originally Posted by frmky

That's an old version that doesn't contain my work on the GPU LA. I'm not aware of anyone that's compiled a Windows version with my changes.

There's a guide to using CUDA with WSL2 at https://docs.nvidia.com/cuda/wsl-user-guide/index.html.

What I have on this machine is WSL1. A better option, for me, is to install Ubuntu on an idle machine. I have v21.10 burned on a DVD-R. This would, at least, give me the GUI. I know enough console commands to get by. The system I am considering has a GTX 1080 in it. I have notes I made relative to that particular system, including CUDA driver installation. I'm not familiar with "bash" files so I would have to do a bit of reading.

Point me to a version I can try and I will give it a go.

[storm5510]

Disregard the above. I found several versions on James Heinrich's mirror page which will run. The one I am running now is "1018-vbits256-sandybridge."

[charybdis]

Originally Posted by storm5510

Disregard the above. I found several versions on James Heinrich's mirror page which will run. The one I am running now is "1018-vbits256-sandybridge."

None of the versions on mersenne.ca have the GPU linear algebra code.

[frmky]

Once you have installed the compiler and CUDA toolkit...
https://docs.nvidia.com/cuda/cuda-in...u-installation

sudo apt install git libgmp-dev
git clone https://github.com/gchilders/msieve_nfsathome.git -b msieve-lacuda-nfsathome
cd msieve_nfsathome
make all CUDA=75 VBITS=256

If you don't want to run from that directory, then copy msieve, *.ptx, and cub/*.so to the directory you are running in.

[storm5510]

Originally Posted by charybdis

None of the versions on mersenne.ca have the GPU linear algebra code.

Then where can they be found?