GPU LL Testing FAQ

[bayanne]

Is this there way of getting everyone to start using Metal, if so had anyone got any Mersenne stuff working with Metal?

[M344587487]

Something along those lines. In my biased opinion it's just another anti-consumer effort in a long line of anti-consumer efforts putting unnecessary barriers up to create unnecessary walled gardens to solely benefit Apple.



If you're lucky there might be some form of OpenCL 2.0 on Metal translation layer and if you're even luckier it might not be too costly in overhead. MoltenVK does it for Vulkan, MoltenGL does it for OpenGL. Can't seem to find one for OpenCL 2.0.

[M344587487]

Originally Posted by bayanne

I thought I would ask again if anyone has got Mersenne in any form running on a AMD GPU on a Mac

https://www.mersenneforum.org/showpo...postcount=1508

[Blackadder]

Which is faster: CPU or GPU LL testing? (I have an Intel i7-3820 CPU @ 3.60GHz and a 2x SLI GeForce 670 GTX (compute capability: 3.0))

[mrh]

Originally Posted by Blackadder

Which is faster: CPU or GPU LL testing? (I have an Intel i7-3820 CPU @ 3.60GHz and a 2x SLI GeForce 670 GTX (compute capability: 3.0))

Since the i7-3820 has four memory channels, I would think it would be faster, even though the memory isn't terribly fast. The 670 GTX would be better suited to TF, IMO.

[Blackadder]

Originally Posted by mrh

Since the i7-3820 has four memory channels, I would think it would be faster, even though the memory isn't terribly fast. The 670 GTX would be better suited to TF, IMO.

To be honest I'm asking that to figure out if the hassle to get CUDALucas to work on Linux (Xubuntu 18.04) is worth the effort. Had to format my OS because of broken package issues—probably because of my own ignorance, but fortunately I like a clean install once in a while—and It's not particularly clear which versions of CUDA, Ubuntu, and CUDALucas are compatible with one another beforehand.

It would be handy to LL test on a GPU though, then I can keep using my PC at full speed while running LL-testing all day. But it's a luxury, I mostly keep running mprime regardless.

[mrh]

Originally Posted by Blackadder

To be honest I'm asking that to figure out if the hassle to get CUDALucas to work on Linux (Xubuntu 18.04) is worth the effort. Had to format my OS because of broken package issues—probably because of my own ignorance, but fortunately I like a clean install once in a while—and It's not particularly clear which versions of CUDA, Ubuntu, and CUDALucas are compatible with one another beforehand.

It would be handy to LL test on a GPU though, then I can keep using my PC at full speed while running LL-testing all day. But it's a luxury, I mostly keep running mprime regardless.

On my system running mprime on the cpu has much less impact on the "feel" of the system than running something on the gpu. I only use the gpu for computation when I am away from the desktop. Also, unless you have a water cooled gpu, it will be noisy.

[kriesel]

Originally Posted by Blackadder

Which is faster: CPU or GPU LL testing? (I have an Intel i7-3820 CPU @ 3.60GHz and a 2x SLI GeForce 670 GTX (compute capability: 3.0))

Use the benchmarks pages at mersenne.ca to estimate throughput. Also note the error checking is much better in PRP than in LL. GpuOwL has the excellent detection rate Gerbicz error check, while CUDALucas does not have even the 50% detection rate Jacobi check. Some NVIDIA gpus can run GpuOwL now, some can't. Since the GTX670 specs say OpenCL 1.2, it may be able to. (Some gpus indicate a mix of 1.1 and 1.2, and fail to run it.) It's a 7 year old design. More throughput per watt hour is available in newer gpu models.
The backlog of double-checking LL tests is now several years long. Double checking PRP is much less of an issue because of the very effective Gerbicz check.

[Blackadder]

Originally Posted by mrh

On my system running mprime on the cpu has much less impact on the "feel" of the system than running something on the gpu. I only use the gpu for computation when I am away from the desktop. Also, unless you have a water cooled gpu, it will be noisy.

Thanks for the warning, I'll stick to my CPU because noise is a big no.

[Viliam Furik]

Hello, I've got RTX 2080Ti and when I run CUDALucas, it shows only 2% power consumption. Is there a way (probably some settings) to make it run faster and take more power (not necessarily the latter). Currently, it runs about 3 ms/iteration.

[ewmayer]

In looking at the GPU subforum through a n00b user's eyes, it strikes me what a mess it is. I want to be able to get to the best practices for my target GPU/OS in post 1 of a "GPU how-to" thread. This thread has 2 problems in that regard:

1. Title, "GPU LL Testing FAQ", reflects the old days when nVidia was the only game in town - it should be retitled to reflect that it's specific to CudaLucas on nVidia gfx cards, and "users of ___ cards should visit the ____ thread(s)".

2. Whatever was initially posted in Post #1, as a new user I expect to see either a list of, or link to, a Best Practices guide right there, and to have same updated on a running basis to reflect changes in Best Practices and/or new editions of hardware and software of the particular family covered by the thread. The PDF in Post #1 of this thread was last updated in 2012, based on the post edit history.