gpuOwL: an OpenCL program for Mersenne primality testing

[kriesel]

Originally Posted by Xyzzy

Somehow we got gpuowl to work on our integrated graphics. (It runs so slow that it isn't worth doing, but it is neat that it works at all!)

2020-09-25 13:26:07 Intel(R) UHD Graphics 630-1 OpenCL compilation in 6.71 s
2020-09-25 13:26:21 Intel(R) UHD Graphics 630-1 77936867 OK        0 loaded: blockSize 400, 0000000000000003
2020-09-25 13:26:56 Intel(R) UHD Graphics 630-1 77936867 OK      800   0.00%; 29530 us/it; ETA 26d 15:17; 1579c241dc63eca6 (check 11.88s)
2020-09-25 15:07:23 Intel(R) UHD Graphics 630-1 77936867 OK   200000   0.26%; 30198 us/it; ETA 27d 04:05; f0b04b45b0855bd2 (check 11.97s)

Yes. HD630<UHD630. Your posted timings extrapolate to 6-7 weeks for one PRP of a wavefront exponent. And it would contend for the TDP and memory bandwidth available for mprime etc on the cpu. Which version gpuowl was it?

https://www.techpowerup.com/gpu-spec...hics-630.c2962
https://gpu.userbenchmark.com/Compar...56797vsm178724

[moebius]

Originally Posted by Xyzzy

Somehow we got gpuowl to work on our integrated graphics. (It runs so slow 2020-09-25 15:07:23 Intel(R) UHD Graphics 630-1 77936867 OK 200000 0.26%; 30198 us/it; ETA 27d 04:05; f0b04b45b0855bd2 (check 11.97s)

Without wanting to offend you, don't do it, it's an excessive waste of electricity.

[Xyzzy]

Originally Posted by kriesel

Yes. HD630<UHD630. Your posted timings extrapolate to 6-7 weeks for one PRP of a wavefront exponent. And it would contend for the TDP and memory bandwidth available for mprime etc on the cpu.

We were not using the integrated graphics for the display so maybe that contributed to it working.

Originally Posted by kriesel

Which version gpuowl was it?

6.11

Originally Posted by moebius

Without wanting to offend you, don't do it, it's an excessive waste of electricity.

[preda]

Originally Posted by Neutron3529

Is it possible to test SP in gpuowl now?
I will soon bought nvidia RTX 3090 and want to test whether 3090 generate good results
I want to test dp,sp and int32(maybe using M31), is it possible using gpuowl to perform such test?


anyway, thanks for your great program!

For a developer wanting to explore, those past experiments are available in git history; but they're not active now.

I'm planning to implement a double-SP in gpuowl next (1 month timeframe).

[Neutron3529]

Originally Posted by preda

For a developer wanting to explore, those past experiments are available in git history; but they're not active now.

I'm planning to implement a double-SP in gpuowl next (1 month timeframe).

Thank you for your reply!


...


I'll wait both your implementation and the water-cooling version of RTX 3090.
(A sad thing... maybe you'll be faster than RTX3090 )

[kriesel]

Originally Posted by Neutron3529

Is it possible to test SP in gpuowl now?
I will soon bought nvidia RTX 3090 and want to test whether 3090 generate good results
I want to test dp,sp and int32(maybe using M31), is it possible using gpuowl to perform such test?

V1.9 & 1.10 implemented DP, SP, M31, and M61. It's unlikely to run on NVIDIA though. FFT lengths supported then were 2M, 4M, 8M only IIRC.
https://github.com/preda/gpuowl/blob/NTT/README.md
https://www.mersenneforum.org/showpo...39&postcount=4

[storm5510]

Originally Posted by kriesel

V1.9 & 1.10 implemented DP, SP, M31, and M61. It's unlikely to run on NVIDIA though. FFT lengths supported then were 2M, 4M, 8M only IIRC.

I want to get this straight. I have looked at the last several postings above. They read like features will be added which will not function on Nvidia GPU's, making it unusable. Is this correct?

[VBCurtis]

Originally Posted by storm5510

I want to get this straight. I have looked at the last several postings above. They read like features will be added which will not function on Nvidia GPU's, making it unusable. Is this correct?

Do you understand how version numbers work? If you do, you can figure out whether 1.9 and 1.10 are in the future or the past. I say this since the context of the posts escapes you entirely.

[kriesel]

Originally Posted by kriesel

V1.9 & 1.10 implemented DP, SP, M31, and M61. It's unlikely to run on NVIDIA though. FFT lengths supported then were 2M, 4M, 8M only IIRC.
https://github.com/preda/gpuowl/blob/NTT/README.md
https://www.mersenneforum.org/showpo...39&postcount=4

Second link includes this text:

Nov 2017 V1.9 binaries for Windows http://www.mersenneforum.org/showpos...&postcount=226
Jan 2018 V1.9 binaries updated for Windows http://www.mersenneforum.org/showpos...&postcount=272

Preda develops gpuowl on AMD gpus, linux, and its rocm driver. He has been generous enough to support Windows also from the very beginning, and with a little help from others has also since V6.5 (May 2019) supported NVIDIA gpus somewhat. Any support in gpuowl of Intel IGPs or AMD IGPUs etc is probably a lucky coincidence. V6.11 has had an amazingly long run, with at least 380 commits apparently. (I wonder which of those XYZZY referred to above.) Preda's now planning what gpuowl V7.x will or won't offer. Hopefully it will continue the NVIDIA support.

[preda]

Originally Posted by kriesel

Hopefully it will continue the NVIDIA support.

Yes, I'm as excited as anybody about the new shiny hardware coming out.

I also like AMD's open philosophy (open source), and unfortunately Nvidia doesn't share in that. That's the historical reason why I don't own any Nvidia GPU ATM.

OTOH AMD has been underwhelming on the GPU front since the great Radeon VII. Simply no worthy AMD GPU available now. (and that's without saying anything about the software side, the ROCm development..)

So, if Nvidia builds good hardware, we should put it to good use if possible. I plan to explore a bit SP, which if successful would benefit Nvidia HW most.

[storm5510]

Originally Posted by VBCurtis

Do you understand how version numbers work? If you do, you can figure out whether 1.9 and 1.10 are in the future or the past. I say this since the context of the posts escapes you entirely.

I should know how version numbers work. I've at this since 1988. The gpuOwl version I am running now is 6.11.380. Before was 6.11.340. I do not understand how 1.9 and 1.10 fit into this scenario, unless they go way back in time and were early releases.