gpuOwL: an OpenCL program for Mersenne primality testing

[Viliam Furik]

Rerun is expected to be completed about a week from now, FYI.

[kriesel]

Originally Posted by Prime95

The second proof failure is here: 108979853

I do not know if Bruno Victal is a forum member and can comment on what may have happened.

"Result" field for a Cert needs something more informative than n/a in all cases; match or mismatch, pass or fail, something. Maybe also proof power in case of a success.

[UBR47K]

Originally Posted by Prime95

The second proof failure is here: 108979853

I do not know if Bruno Victal is a forum member and can comment on what may have happened.

Hi, that is the first tested exponent from a recently assembled computer (the GPU is a Fury X from a previous machine).

No significant changes have been made (I believe):

1. gpuowl path to openCL was modified (Makefile)
2. rocm-smi --setsclk 2
3. config.txt: -proof 9 -nospin

[Prime95]

Do you have a gpuowl.log file that we can look at for error messages?

[petrw1]

Originally Posted by petrw1

Ok thanks for trying

GPUOwl doing P-1 in Colab that is.
We'll see how long my free GPU lasts for this and how far into a (7GhzDay) P-1 that gets me.

[UBR47K]

Originally Posted by Prime95

Do you have a gpuowl.log file that we can look at for error messages?

sent via PM

[DrobinsonPE]

I have been testing my Ryzen 3 3200G computer with both mfakto and gpuowl. I posted my mfakto results in the mfakto thread. The last few days I have been testing gpuowl with some interesting results. I think I computed all of the GHz-day/day correctly.

Computer - Asrock DeskMini A300, AMD Ryzen 3 3200G, 16GB RAM, 256GB SSD 
Programs - Windows 10, Prime95 30.3 Build 4, GPUOWL v6.11-364

Prime95 times running alone - (power usage 81W)(29.9GHz-day/day)
[Sep 13 15:43] Iteration: 49270000 / 104908800 [46.96%], ms/iter: 11.411, ETA: 7d 08:22
[Sep 13 15:45] Iteration: 49280000 / 104908800 [46.97%], ms/iter: 11.547, ETA: 7d 10:25

Prime95 times with GPUOWL PM-1 running - (power usage 91W)(25.5GHz-day/day)(14.7% slower)
[Sep 13 16:34] Iteration: 49500000 / 104908800 [47.18%], ms/iter: 13.921, ETA: 8d 22:15
[Sep 13 16:36] Iteration: 49510000 / 104908800 [47.19%], ms/iter: 14.036, ETA: 8d 23:59

Prime95 times with GPUOWL PRP running - (power usage 94W)(25.5GHz-day/day)(14.7% slower)
[Sep 14 22:13] Iteration: 56730000 / 104908800 [54.07%], ms/iter: 13.899, ETA: 7d 18:00
[Sep 14 22:16] Iteration: 56740000 / 104908800 [54.08%], ms/iter: 14.005, ETA: 7d 19:23

GPUOWL PM-1 times running alone - (power usage 52W)(26.6GHz-day/day)
2020-09-16 22:47:03 gfx902-0 105159643 P1   520000  36.06%; 15870 us/it; ETA 0d 04:04; 212733bbd10c822d
2020-09-16 22:49:42 gfx902-0 105159643 P1   530000  36.75%; 15859 us/it; ETA 0d 04:01; 865315d9f49abf1f
2020-09-17 05:19:27 gfx902-0 105159643 P2  966/2880: 70687 primes; setup 14.65 s,  17.759 ms/prime
2020-09-17 05:40:42 gfx902-0 105159643 P2 1104/2880: 70891 primes; setup 14.37 s,  17.777 ms/prime
results - 14.9848 GHz-Day

GPUOWL PM-1 times with Prime95 running - (power usage 91W)(13.5GHz-day/day)(P-1 is 51% slower when Prime95 is running)
2020-09-13 16:56:02 gfx902-0 105008249 P1   100000   6.93%; 30389 us/it; ETA 0d 11:20; 45968c2f6d885a93
2020-09-13 17:01:08 gfx902-0 105008249 P1   110000   7.63%; 30631 us/it; ETA 0d 11:20; c23f0f38db0b80a3
2020-09-14 15:23:00 gfx902-0 105008249 P2 2208/2880: 70658 primes; setup 24.73 s,  34.802 ms/prime
2020-09-14 16:04:30 gfx902-0 105008249 P2 2346/2880: 70408 primes; setup 28.15 s,  34.968 ms/prime
Results - 14.9848 GHz-Day

GPUOWL PRP times running alone - (power usage 54W)(21.1GHz-day/day)(PRP is 20.7% slower than P-1)
2020-09-16 11:16:39 gfx902-0 83926723 OK  6800000   8.10%; 12752 us/it; ETA 11d 09:12; 703c56f18bd2e959 (check 5.17s)
2020-09-16 11:59:15 gfx902-0 83926723 OK  7000000   8.34%; 12753 us/it; ETA 11d 08:30; 6e98ee9be1a9552f (check 5.30s)

GPUOWL PRP times with Prime95 running - (power usage 94W)(11.1GHz-day/day)(PRP is 43.8% slower when Prime95 is running)
2020-09-15 09:29:42 gfx902-0 83926723 OK  2000000   2.38%; 24253 us/it; ETA 22d 23:56; d14c26a629f08e54 (check 10.06s)
2020-09-15 10:51:19 gfx902-0 83926723 OK  2200000   2.62%; 24437 us/it; ETA 23d 02:47; 00fa1ec580617f52 (check 9.77s)

[storm5510]

I did a major hardware upgrade today which i wrote about in my blog. I was hoping for some significant improvement in P-1 run times. That didn't happen. The Stage 1 time was shorter, but Stage 2 was longer. Basically, no gain. I haven't tried a DC or PRP yet. I have a feeling they would not do any better either. I shifted my P-1 work to Prime95.

I want to use this where it would do the most good. Since it sails with mfaktc-2047, I am considering TF's to a bit level where they would be P-1 ready. 76 or 77 bits. By no means am I disappointed. Just not the result I was hoping for.

[petrw1]

Originally Posted by petrw1

GPUOwl doing P-1 in Colab that is.
We'll see how long my free GPU lasts for this and how far into a (7GhzDay) P-1 that gets me.

Getting over 5 hours a day last few days.
P100 is doing about 14GhzDays per hour..
T4 is about 5 times SLOWER.
Haven't gotten a K80 or P4 yet.

[Prime95]

Originally Posted by Prime95

The second proof failure is here: 108979853

I do not know if Bruno Victal is a forum member and can comment on what may have happened.

And another proof failure http://mersenne.org/M108979987

[ATH]

Originally Posted by Prime95

And another proof failure http://mersenne.org/M108979987

Another shift 0 PRP tests from gpuowl probably. Maybe another FFT issue with an old version like I had.