gpuOwL: an OpenCL program for Mersenne primality testing

[Xyzzy]

1070

2020-11-24 19:07:49 GeForce GTX 1070-0 OpenCL compilation in 1.61 s
2020-11-24 19:07:51 GeForce GTX 1070-0 77936867 OK        0 loaded: blockSize 400, 0000000000000003
2020-11-24 19:07:57 GeForce GTX 1070-0 77936867 OK      800   0.00%; 4528 us/it; ETA 4d 02:02; 1579c241dc63eca6 (check 1.87s)
2020-11-24 19:23:33 GeForce GTX 1070-0 77936867 OK   200000   0.26%; 4691 us/it; ETA 4d 05:17; f0b04b45b0855bd2 (check 1.96s)
2020-11-24 19:39:41 GeForce GTX 1070-0 77936867 OK   400000   0.51%; 4830 us/it; ETA 4d 08:01; c03f94396a5aa29e (check 2.12s)
2020-11-24 19:55:59 GeForce GTX 1070-0 77936867 OK   600000   0.77%; 4877 us/it; ETA 4d 08:46; b9decd65ca71b629 (check 2.04s)

[Dylan14]

For Arch I have created a working PKGBUILD for gpuowl, based on the stable tag version. You can find it here: https://aur.archlinux.org/packages/gpuowl/.

[preda]

Originally Posted by Dylan14

For Arch I have created a working PKGBUILD for gpuowl, based on the stable tag version. You can find it here: https://aur.archlinux.org/packages/gpuowl/.

Nice, thank you! I hope others will find the package useful.

[aheeffer]

This time I had one from the first batch delivered, a RTX 3060 ti from Gigabyte Aorus Ti Master

Factory clock setting is 1800 MHz and 1400 Mhz memory, CPU-Z 1.94.8 reports 2010 MHz clock and 6800 MHz memory. Mfaktc reports 1800 MHz clock and 7000 MHz memory.

Under Windows using gpuowl v6.11-380:

2020-12-04 15:06:46 GeForce RTX 3060 Ti-0 OpenCL compilation in 2.33 s
2020-12-04 15:06:48 GeForce RTX 3060 Ti-0 57885161 OK        0 loaded: blockSize 400, 0000000000000003
2020-12-04 15:06:48 GeForce RTX 3060 Ti-0 validating proof residues for power 8
2020-12-04 15:06:48 GeForce RTX 3060 Ti-0 Proof using power 8
2020-12-04 15:06:50 GeForce RTX 3060 Ti-0 57885161 OK      800   0.00%; 2379 us/it; ETA 1d 14:15; 5727fe6a7225c273 (check 1.00s)
2020-12-04 15:14:49 GeForce RTX 3060 Ti-0 57885161 OK   200000   0.35%; 2399 us/it; ETA 1d 14:26; de62d6db1ad5092d (check 1.01s)
2020-12-04 15:22:51 GeForce RTX 3060 Ti-0 57885161 OK   400000   0.69%; 2401 us/it; ETA 1d 14:20; 45e043b36f3556e1 (check 1.01s)

and for Moebius

2020-12-04 15:24:04 GeForce RTX 3060 Ti-0 OpenCL compilation in 2.05 s
2020-12-04 15:24:06 GeForce RTX 3060 Ti-0 77936867 OK        0 loaded: blockSize 400, 0000000000000003
2020-12-04 15:24:06 GeForce RTX 3060 Ti-0 validating proof residues for power 8
2020-12-04 15:24:06 GeForce RTX 3060 Ti-0 Proof using power 8
2020-12-04 15:24:10 GeForce RTX 3060 Ti-0 77936867 OK      800   0.00%; 3232 us/it; ETA 2d 21:59; 1579c241dc63eca6 (check 1.35s)
2020-12-04 15:35:04 GeForce RTX 3060 Ti-0 77936867 OK   200000   0.26%; 3276 us/it; ETA 2d 22:44; f0b04b45b0855bd2 (check 1.36s)
2020-12-04 15:45:56 GeForce RTX 3060 Ti-0 77936867 OK   400000   0.51%; 3256 us/it; ETA 2d 22:08; c03f94396a5aa29e (check 1.36s)

[Viliam Furik]

Quick question: How much PCIe bandwidth does gpuOwl need if no monitor is connected? I found out that GPUs when used for mining cryptocurrencies, can work with PCIe x1 Gen 1. Is that the case for gpuOwl, too?

[kriesel]

Originally Posted by Viliam Furik

Quick question: How much PCIe bandwidth does gpuOwl need if no monitor is connected? I found out that GPUs when used for mining cryptocurrencies, can work with PCIe x1 Gen 1. Is that the case for gpuOwl, too?

My mining rig style systems say x1 to x16 extenders are fine, for gpuowl or mfaktc etc. There may be some performance loss during gpu-host transfers, but it's offset by better cooling and higher gpu count.
If you find a utility to easily look at PCIe traffic stats on Windows, please post a link.
https://community.intel.com/t5/Softw...re/td-p/765471
Alas, Radeon Vii, on PCIe V2, x1 to x16 extender, and GPU-Z combo, does not display Bus interface load, although RTX2080Super does (note the 0% usage there, on PCIe x16 V1.1). On a GTX1080 in PCIe v2x16 direct, it's running 48% in 183M PRP. I don't think I have any PCIe v1.0 to test. Even the Core2 Duos and Quadro 4000 or 2000s are V1.1 or higher. I'm running gpuowl 6.x and 7.x on Radeons with a PCIe Gen1 setting in the motherboard BIOS, and getting within ~10% on Windows 10 of what others report getting for throughput on Linux.

[kriesel]

Some interesting instrumentation here: https://github.com/opcm/pcm/blob/master/README.md

[DrobinsonPE]

After switching to Windows 10, I got gpuowl V6.11-380 to run on my ASRock Deskmini A300W, AMD A8-9600, 16GB DDR-4, SSD.

I tried multiple times to get ROCm to work in linux but it seems like I still have some learning to do before I can make that happen. After seeing the mprime performance of the AMD A8-9600, (max output 13 GHz-D/day while using 70W power) I shut down the computer and started looking for a Zen 2 series cpu to replace it. I got bored waiting for the cpu prices to come down and decided to try again with Windows 10. The iGPU works better than I thought it would. Mfakto results will be posted in the proper thread as well.

Moebius, here is your timing:

\gpuowl\611380>gpuowl-win -iters 200000 -prp 77936867
2020-12-05 18:14:05 gpuowl v6.11-380-g79ea0cc
2020-12-05 18:14:05 config: -iters 200000 -prp 77936867
2020-12-05 18:14:05 Bristol Ridge-0 77936867 FFT: 4M 1K:8:256 (18.58 bpw)
2020-12-05 18:14:11 Bristol Ridge-0 OpenCL compilation in 4.82 s
2020-12-05 18:14:17 Bristol Ridge-0 77936867 OK      800 loaded: blockSize 400, 1579c241dc63eca6
2020-12-05 18:14:17 Bristol Ridge-0 validating proof residues for power 8
2020-12-05 18:14:17 Bristol Ridge-0 Proof using power 8
2020-12-05 18:14:34 Bristol Ridge-0 77936867 OK     1600   0.00%; 13289 us/it; ETA 11d 23:41; 0f62a1fcc1c78fe9 (check 5.49s)
2020-12-05 18:58:18 Bristol Ridge-0 77936867 OK   200000   0.26%; 13201 us/it; ETA 11d 21:04; f0b04b45b0855bd2 (check 5.19s)
2020-12-05 18:58:23 Bristol Ridge-0 Stopping, please wait..
2020-12-05 18:58:33 Bristol Ridge-0 77936867 OK   200800   0.26%; 12658 us/it; ETA 11d 09:20; 895b034c5473a608 (check 5.21s)
2020-12-05 18:58:33 Bristol Ridge-0 Exiting because "stop requested"

[preda]

Originally Posted by Viliam Furik

Quick question: How much PCIe bandwidth does gpuOwl need if no monitor is connected? I found out that GPUs when used for mining cryptocurrencies, can work with PCIe x1 Gen 1. Is that the case for gpuOwl, too?

GpuOwl uses very very little PCIe bandwidth. I regularly run it over PCIe x1 Gen1 without significant slowdown.

[moebius]

Thx to aheeffer and DrobinsonPE. The values ​​are now in the list.

[Xyzzy]

Radeon Pro WX 2100

2020-12-10 18:40:54 gfx804-0 OpenCL compilation in 1.72 s
2020-12-10 18:41:18 gfx804-0 77936867 OK      800   0.00%; 14048 us/it; ETA 12d 16:08; 1579c241dc63eca6 (check 5.68s)
2020-12-10 18:43:33 gfx804-0 77936867 OK    10000   0.01%; 14107 us/it; ETA 12d 17:22; fc4f135f7cf4ad29 (check 5.68s)
2020-12-10 18:46:00 gfx804-0 77936867 OK    20000   0.03%; 14084 us/it; ETA 12d 16:50; 3cd1bd9d5e09cbc5 (check 5.68s)
2020-12-10 18:48:26 gfx804-0 77936867 OK    30000   0.04%; 14085 us/it; ETA 12d 16:49; c4e0ff35e3290d98 (check 5.68s)
2020-12-10 18:50:53 gfx804-0 77936867 OK    40000   0.05%; 14086 us/it; ETA 12d 16:48; dffe1b1b0d748128 (check 5.69s)
2020-12-10 18:53:19 gfx804-0 77936867 OK    50000   0.06%; 14087 us/it; ETA 12d 16:46; 52e286945371ed29 (check 5.68s)
2020-12-10 18:55:47 gfx804-0 77936867 OK    60000   0.08%; 14174 us/it; ETA 12d 18:37; 0945da4dc08bdd95 (check 5.69s)