CUDALucas (a.k.a. MaclucasFFTW/CUDA 2.3/CUFFTW) - Page 258

TheJudger · 2020-11-01, 01:02

CUDALucas 2.06, CUDA 11.1.1, Quick&Dirty run, did a "./CUDALucas -cufftbench 2048 32768 20" before the following runs on each GPU.

A100 PCIe, actual clock rate 1200-1230 MHz and power 250 Watt during the run

Code:

Using threads: square 256, splice 128.
Starting M57885161 fft length = 3136K
|   Date     Time    |   Test Num     Iter        Residue        |    FFT   Error     ms/It     Time  |       ETA      Done   |
|  Nov 01  01:38:49  |  M57885161     10000  0x76c27556683cd84d  |  3136K  0.16016   0.2899    2.89s  |      4:39:40   0.01%  |
|  Nov 01  01:38:52  |  M57885161     20000  0xfd8e311d20ffe6ab  |  3136K  0.17969   0.2854    2.85s  |      4:37:26   0.03%  |
|  Nov 01  01:38:55  |  M57885161     30000  0xce0d85ab0065a232  |  3136K  0.16406   0.2861    2.86s  |      4:36:53   0.05%  |
[...]
|  Nov 01  01:39:59  |  M57885161    250000  0x6d8473f6fd9a63c1  |  3136K  0.17188   0.2890    2.89s  |      4:38:07   0.43%  |
|  Nov 01  01:40:01  |  M57885161    260000  0x214c6beab9d34d77  |  3136K  0.17969   0.2891    2.89s  |      4:38:03   0.44%  |
|  Nov 01  01:40:04  |  M57885161    270000  0x1e4eb1f3c280c344  |  3136K  0.16406   0.2890    2.89s  |      4:37:59   0.46%  |
|  Nov 01  01:40:07  |  M57885161    280000  0x678b32b35f88d73d  |  3136K  0.17188   0.2891    2.89s  |      4:37:56   0.48%  |
|  Nov 01  01:40:10  |  M57885161    290000  0x4f51138c1f7f6301  |  3136K  0.17188   0.2889    2.88s  |      4:37:52   0.50%  |
|  Nov 01  01:40:13  |  M57885161    300000  0x4332497b4eefcf79  |  3136K  0.17188   0.2892    2.89s  |      4:37:48   0.51%  |

Code:

Using threads: square 256, splice 128.
Starting M100709233 fft length = 5488K
|   Date     Time    |   Test Num     Iter        Residue        |    FFT   Error     ms/It     Time  |       ETA      Done   |
|  Nov 01  01:50:52  | M100709233     10000  0x65cca90366e670ef  |  5488K  0.23047   0.6465    6.46s  |     18:05:07   0.00%  |
|  Nov 01  01:50:59  | M100709233     20000  0xb557cc85bc57b0bc  |  5488K  0.25000   0.6450    6.45s  |     18:03:44   0.01%  |
|  Nov 01  01:51:05  | M100709233     30000  0xd758016f2a75aebd  |  5488K  0.21875   0.6463    6.46s  |     18:03:55   0.02%  |
|  Nov 01  01:51:12  | M100709233     40000  0x769ac25053698341  |  5488K  0.22656   0.6475    6.47s  |     18:04:28   0.03%  |
|  Nov 01  01:51:18  | M100709233     50000  0xc9a8e17405097ed2  |  5488K  0.23438   0.6477    6.47s  |     18:04:49   0.04%  |
|  Nov 01  01:51:25  | M100709233     60000  0xba8b8115d1b0e392  |  5488K  0.21875   0.6487    6.48s  |     18:05:19   0.05%  |
|  Nov 01  01:51:31  | M100709233     70000  0x6f32a109753b2af0  |  5488K  0.21680   0.6490    6.49s  |     18:05:42   0.06%  |
|  Nov 01  01:51:38  | M100709233     80000  0x390b740965708e6d  |  5488K  0.23438   0.6495    6.49s  |     18:06:04   0.07%  |
|  Nov 01  01:51:44  | M100709233     90000  0x67d98743ef47ec14  |  5488K  0.22656   0.6497    6.49s  |     18:06:22   0.08%  |
|  Nov 01  01:51:51  | M100709233    100000  0x84a6b8fc954d383b  |  5488K  0.21875   0.6500    6.50s  |     18:06:38   0.09%  |

Quadro RTX 8000, actual clock rate 1890 MHz and power 210-215 Watt during the run

Code:

Using threads: square 256, splice 128.
Starting M57885161 fft length = 3456K
|   Date     Time    |   Test Num     Iter        Residue        |    FFT   Error     ms/It     Time  |       ETA      Done   |
|  Nov 01  01:37:11  |  M57885161     10000  0x76c27556683cd84d  |  3456K  0.01685   1.7461   17.46s  |   1:04:04:19   0.01%  |
|  Nov 01  01:37:28  |  M57885161     20000  0xfd8e311d20ffe6ab  |  3456K  0.01611   1.7516   17.51s  |   1:04:06:40   0.03%  |
|  Nov 01  01:37:46  |  M57885161     30000  0xce0d85ab0065a232  |  3456K  0.01758   1.7596   17.59s  |   1:04:09:50   0.05%  |
[...]
|  Nov 01  01:44:18  |  M57885161    250000  0x6d8473f6fd9a63c1  |  3456K  0.01660   1.7835   17.83s  |   1:04:28:52   0.43%  |
|  Nov 01  01:44:36  |  M57885161    260000  0x214c6beab9d34d77  |  3456K  0.02051   1.7834   17.83s  |   1:04:28:44   0.44%  |
|  Nov 01  01:44:54  |  M57885161    270000  0x1e4eb1f3c280c344  |  3456K  0.01743   1.7834   17.83s  |   1:04:28:35   0.46%  |
|  Nov 01  01:45:11  |  M57885161    280000  0x678b32b35f88d73d  |  3456K  0.01758   1.7834   17.83s  |   1:04:28:26   0.48%  |
|  Nov 01  01:45:29  |  M57885161    290000  0x4f51138c1f7f6301  |  3456K  0.01758   1.7834   17.83s  |   1:04:28:16   0.50%  |
|  Nov 01  01:45:47  |  M57885161    300000  0x4332497b4eefcf79  |  3456K  0.01855   1.7834   17.83s  |   1:04:28:06   0.51%  |

Code:

Using threads: square 256, splice 128.
Starting M100709233 fft length = 5760K
|   Date     Time    |   Test Num     Iter        Residue        |    FFT   Error     ms/It     Time  |       ETA      Done   |
|  Nov 01  01:51:48  | M100709233     10000  0x65cca90366e670ef  |  5760K  0.06580   3.2809   32.80s  |   3:19:46:24   0.00%  |
|  Nov 01  01:52:21  | M100709233     20000  0xb557cc85bc57b0bc  |  5760K  0.06250   3.3074   33.07s  |   3:20:08:07   0.01%  |
|  Nov 01  01:52:55  | M100709233     30000  0xd758016f2a75aebd  |  5760K  0.06641   3.3288   33.28s  |   3:20:26:57   0.02%  |
|  Nov 01  01:53:28  | M100709233     40000  0x769ac25053698341  |  5760K  0.06744   3.3346   33.34s  |   3:20:38:32   0.03%  |
|  Nov 01  01:54:01  | M100709233     50000  0xc9a8e17405097ed2  |  5760K  0.06250   3.3346   33.34s  |   3:20:45:15   0.04%  |
|  Nov 01  01:54:35  | M100709233     60000  0xba8b8115d1b0e392  |  5760K  0.06836   3.3346   33.34s  |   3:20:49:33   0.05%  |
|  Nov 01  01:55:08  | M100709233     70000  0x6f32a109753b2af0  |  5760K  0.06348   3.3346   33.34s  |   3:20:52:29   0.06%  |
|  Nov 01  01:55:41  | M100709233     80000  0x390b740965708e6d  |  5760K  0.06250   3.3347   33.34s  |   3:20:54:32   0.07%  |
|  Nov 01  01:56:15  | M100709233     90000  0x67d98743ef47ec14  |  5760K  0.06641   3.3346   33.34s  |   3:20:56:00   0.08%  |
|  Nov 01  01:56:48  | M100709233    100000  0x84a6b8fc954d383b  |  5760K  0.06641   3.3346   33.34s  |   3:20:57:04   0.09%  |

Geforce RTX 3090, actual clock rate 1905-1935 MHz and power 330-335 Watt during the run

Code:

Using threads: square 256, splice 128.
Starting M57885161 fft length = 3200K
|   Date     Time    |   Test Num     Iter        Residue        |    FFT   Error     ms/It     Time  |       ETA      Done   |
|  Nov 01  01:37:16  |  M57885161     10000  0x76c27556683cd84d  |  3200K  0.10156   1.4418   14.41s  |     23:10:45   0.01%  |
|  Nov 01  01:37:30  |  M57885161     20000  0xfd8e311d20ffe6ab  |  3200K  0.09766   1.4469   14.46s  |     23:13:00   0.03%  |
|  Nov 01  01:37:45  |  M57885161     30000  0xce0d85ab0065a232  |  3200K  0.10938   1.4567   14.56s  |     23:16:44   0.05%  |
[...]
|  Nov 01  01:43:12  |  M57885161    250000  0x6d8473f6fd9a63c1  |  3200K  0.10156   1.4978   14.97s  |     23:45:11   0.43%  |
|  Nov 01  01:43:27  |  M57885161    260000  0x214c6beab9d34d77  |  3200K  0.09946   1.4982   14.98s  |     23:45:28   0.44%  |
|  Nov 01  01:43:42  |  M57885161    270000  0x1e4eb1f3c280c344  |  3200K  0.10156   1.4985   14.98s  |     23:45:44   0.46%  |
|  Nov 01  01:43:57  |  M57885161    280000  0x678b32b35f88d73d  |  3200K  0.10156   1.4989   14.98s  |     23:45:58   0.48%  |
|  Nov 01  01:44:12  |  M57885161    290000  0x4f51138c1f7f6301  |  3200K  0.10156   1.4985   14.98s  |     23:46:09   0.50%  |
|  Nov 01  01:44:28  |  M57885161    300000  0x4332497b4eefcf79  |  3200K  0.09766   1.4990   14.99s  |     23:46:20   0.51%  |

Code:

Using threads: square 256, splice 128.
Starting M100709233 fft length = 5760K
|   Date     Time    |   Test Num     Iter        Residue        |    FFT   Error     ms/It     Time  |       ETA      Done   |
|  Nov 01  01:51:30  | M100709233     10000  0x65cca90366e670ef  |  5760K  0.06250   2.8093   28.09s  |   3:06:35:01   0.00%  |
|  Nov 01  01:51:59  | M100709233     20000  0xb557cc85bc57b0bc  |  5760K  0.06348   2.8430   28.43s  |   3:07:02:50   0.01%  |
|  Nov 01  01:52:27  | M100709233     30000  0xd758016f2a75aebd  |  5760K  0.06641   2.8559   28.55s  |   3:07:18:58   0.02%  |
|  Nov 01  01:52:56  | M100709233     40000  0x769ac25053698341  |  5760K  0.06445   2.8562   28.56s  |   3:07:26:55   0.03%  |
|  Nov 01  01:53:25  | M100709233     50000  0xc9a8e17405097ed2  |  5760K  0.06250   2.8767   28.76s  |   3:07:38:24   0.04%  |
|  Nov 01  01:53:53  | M100709233     60000  0xba8b8115d1b0e392  |  5760K  0.06250   2.8821   28.82s  |   3:07:47:23   0.05%  |
|  Nov 01  01:54:22  | M100709233     70000  0x6f32a109753b2af0  |  5760K  0.06250   2.8868   28.86s  |   3:07:54:48   0.06%  |
|  Nov 01  01:54:51  | M100709233     80000  0x390b740965708e6d  |  5760K  0.06396   2.8909   28.90s  |   3:08:01:06   0.07%  |
|  Nov 01  01:55:20  | M100709233     90000  0x67d98743ef47ec14  |  5760K  0.06250   2.8922   28.92s  |   3:08:06:08   0.08%  |
|  Nov 01  01:55:49  | M100709233    100000  0x84a6b8fc954d383b  |  5760K  0.06250   2.8934   28.93s  |   3:08:10:17   0.09%  |

Oliver

axn · 2020-11-01, 03:13

Quote:

Originally Posted by TheJudger

CUDALucas 2.06, CUDA 11.1.1, Quick&Dirty run, did a "./CUDALucas -cufftbench 2048 32768 20" before the following runs on each GPU.

Those are impressive numbers from A100. I suspect these would be even more impressive if gpuOWL was used - is that something you can test out?

moebius · 2020-11-01, 03:18

Quote:

Originally Posted by axn

Those are impressive numbers from A100. I suspect these would be even more impressive if gpuOWL was used - is that something you can test out?

I have requested such values, they should be very similar.

axn · 2020-11-01, 03:24

Quote:

Originally Posted by moebius

I have requested such values, they should be very similar.

I suspect gpuOwl will complete these in 3hrs/12hrs respectively compared to Cudalucas 4.5hrs/18hrs.

Of course, at USD 12k, these aren't exactly "affordable".

moebius · 2020-11-01, 03:41

Quote:

Originally Posted by axn

I suspect gpuOwl will complete these in 3hrs/12hrs respectively compared to Cudalucas 4.5hrs/18hrs.

Of course, at USD 12k, these aren't exactly "affordable".

but CudaLucas without proofs.....

kriesel · 2020-11-01, 14:16

Quote:

Originally Posted by moebius

but CudaLucas without proofs.....

To my knowledge no one has implemented LL proof yet.
Also, CUDALucas does not include the Jacobi symbol check, although mlucas, prime95 and some versions of gpuowl have.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Don't DC/LL them with CudaLucas	LaurV	Data	131	2017-05-02 18:41
CUDALucas / cuFFT Performance on CUDA 7 / 7.5 / 8	Brain	GPU Computing	13	2016-02-19 15:53
CUDALucas: which binary to use?	Karl M Johnson	GPU Computing	15	2015-10-13 04:44
settings for cudaLucas	fairsky	GPU Computing	11	2013-11-03 02:08
Trying to run CUDALucas on Windows 8 CP	Rodrigo	GPU Computing	12	2012-03-07 23:20