|
2016-08-05, 03:06
|
#2498 |
Mar 2010
1100110112 Posts |
Code:
4.2 | Date Time | Test Num Iter Residue | FFT Error ms/It Time | ETA Done | | Aug 05 06:15:11 | M57885161 100000 0x0000000000000000 | 3136K 0.00000 0.0422 4.22s | 40:42 0.17% | | Aug 05 06:15:16 | M57885161 200000 0x0000000000000000 | 3136K 0.00000 0.0438 4.38s | 41:23 0.34% | | Aug 05 06:15:20 | M57885161 300000 0x0000000000000000 | 3136K 0.00000 0.0436 4.36s | 41:30 0.51% | | Aug 05 06:15:24 | M57885161 400000 0x0000000000000000 | 3136K 0.00000 0.0369 3.69s | 39:55 0.69% | | Aug 05 06:15:27 | M57885161 500000 0x0000000000000000 | 3136K 0.00000 0.0307 3.06s | 37:45 0.86% | | Aug 05 06:15:30 | M57885161 600000 0x0000000000000000 | 3136K 0.00000 0.0307 3.07s | 36:17 1.03% | | Aug 05 06:15:33 | M57885161 700000 0x0000000000000000 | 3136K 0.00000 0.0310 3.10s | 35:17 1.20% | | Aug 05 06:15:36 | M57885161 800000 0x0000000000000000 | 3136K 0.00000 0.0307 3.07s | 34:28 1.38% | | Aug 05 06:15:39 | M57885161 900000 0x0000000000000000 | 3136K 0.00000 0.0308 3.08s | 33:50 1.55% | | Aug 05 06:15:42 | M57885161 1000000 0x0000000000000000 | 3136K 0.00000 0.0310 3.10s | 33:20 1.72% | | Aug 05 06:15:46 | M57885161 1100000 0x0000000000000000 | 3136K 0.00000 0.0348 3.48s | 33:15 1.90% | 5.0 | Date Time | Test Num Iter Residue | FFT Error ms/It Time | ETA Done | | Aug 05 06:16:19 | M57885161 100000 0x0000000000000000 | 3136K 0.00000 0.0242 2.42s | 23:19 0.17% | | Aug 05 06:16:21 | M57885161 200000 0x0000000000000000 | 3136K 0.00000 0.0244 2.44s | 23:23 0.34% | | Aug 05 06:16:24 | M57885161 300000 0x0000000000000000 | 3136K 0.00000 0.0253 2.53s | 23:39 0.51% | | Aug 05 06:16:27 | M57885161 400000 0x0000000000000000 | 3136K 0.00000 0.0252 2.52s | 23:45 0.69% | | Aug 05 06:16:29 | M57885161 500000 0x0000000000000000 | 3136K 0.00000 0.0246 2.46s | 23:41 0.86% | | Aug 05 06:16:31 | M57885161 600000 0x0000000000000000 | 3136K 0.00000 0.0246 2.46s | 23:37 1.03% | | Aug 05 06:16:34 | M57885161 700000 0x0000000000000000 | 3136K 0.00000 0.0245 2.45s | 23:33 1.20% | | Aug 05 06:16:36 | M57885161 800000 0x0000000000000000 | 3136K 0.00000 0.0250 2.50s | 23:33 1.38% | 5.5 | Aug 05 06:27:53 | M57885161 10000 0x0000000000000000 | 3136K 0.00000 2.1930 21.93s | 1:11:15:20 0.01% | | Aug 05 06:28:15 | M57885161 20000 0x0000000000000000 | 3136K 0.00000 2.1837 21.83s | 1:11:10:29 0.03% | 6.0 | Date Time | Test Num Iter Residue | FFT Error ms/It Time | ETA Done | | Aug 05 06:28:43 | M57885161 30000 0x0000000000000000 | 3136K 0.00000 2.2186 19.96s | 1:11:18:16 0.05% | | Aug 05 06:29:05 | M57885161 40000 0x0000000000000000 | 3136K 0.00000 2.2149 22.14s | 1:11:22:16 0.06% | Does anyone have a CuLu binary compiled with the latest toolkit? Last fiddled with by Karl M Johnson on 2016-08-05 at 03:29 |
|
|
|
2016-08-05, 04:37
|
#2499 |
|
Mar 2010
3×137 Posts |
Hmm, despite the residue being wrong, it still works:
Code:
Starting M756839 fft length = 42K | Date Time | Test Num Iter Residue | FFT Error ms/It Time | ETA Done | | Aug 05 07:36:06 | M756839 100000 0x0000000000000000 | 42K 0.00000 0.0246 2.46s | 0:16 13.21% | | Aug 05 07:36:09 | M756839 200000 0x0000000000000000 | 42K 0.00000 0.0252 2.52s | 0:13 26.42% | | Aug 05 07:36:11 | M756839 300000 0x0000000000000000 | 42K 0.00000 0.0250 2.50s | 0:11 39.63% | | Aug 05 07:36:14 | M756839 400000 0x0000000000000000 | 42K 0.00000 0.0246 2.46s | 0:08 52.85% | | Aug 05 07:36:16 | M756839 500000 0x0000000000000000 | 42K 0.00000 0.0245 2.45s | 0:06 66.06% | | Aug 05 07:36:19 | M756839 600000 0x0000000000000000 | 42K 0.00000 0.0251 2.51s | 0:03 79.27% | | Aug 05 07:36:21 | M756839 700000 0x0000000000000000 | 42K 0.00000 0.0249 2.49s | 0:01 92.49% | M( 756839 )P, n = 42K, CUDALucas v2.05.1, estimated total time = 0:18 |
|
|
|
|
2016-08-05, 05:45
|
#2500 |
|
Mar 2010
3·137 Posts |
CuFFT seems broken, though, hence the wrong FFT size selected in my first test.
All the timings are 0 
|
|
|
|
|
2016-08-05, 05:54
|
#2501 | |
|
Romulan Interpreter
Jun 2011
Thailand
7×1,373 Posts |
Quote:
 Of course, everything is prime with those residues  . Try a non-prime. |
|
|
|
|
|
2016-08-05, 06:11
|
#2502 |
|
Mar 2010
3×137 Posts |
Nailed it!
Everything is prime. 1080 isn't meant for fp64 anyways, it should excel at TF when mfaktc/CUDA gets a fix. As for the benchmarks, I wouldn't even trust the slow benchmarks, not to mention the 23 minutes for M57. Until this issue gets fixed, the data is unreliable I'm afraid. Last fiddled with by Karl M Johnson on 2016-08-05 at 06:14 |
|
|
|
|
2016-08-24, 00:43
|
#2503 | |
Sep 2010
So Cal
5010 Posts |
Quote:
P.S. This thing is slightly faster than a Titan Black, which was a pleasant surprise 
|
|
|
|
|
|
2016-08-24, 05:01
|
#2504 | |
"David"
Jul 2015
Ohio
11·47 Posts |
Quote:
|
|
|
|
|
|
2016-08-24, 06:15
|
#2505 |
|
Sep 2010
So Cal
2×52 Posts |
I think there is a good chance that NVIDIA will address the issues in the final, production release. People that are willing to shell out big bucks on Pascal architectures, pretty much deserve what they paid for.
P.S. I just noticed in another thread that you are running about 90 GPU'S?  Gee, no wonder you're climbing the LL ladder like a bat out of hell  Currently, you're one ahead of me. 
|
|
|
|
|
2016-08-25, 10:33
|
#2506 | |
|
Mar 2010
3·137 Posts |
Quote:
Are you sure that the Titan X(P) is producing correct results? As far as I've read, Nvidia didn't want us to have another non-Tesla GPU with Teslas' fp64 performance, so even new Pascal Titan has poor dp fp performance. Of course, all this applies to CuLu. |
|
|
|
|
|
2016-08-26, 00:06
|
#2507 | |
|
Sep 2010
So Cal
628 Posts |
Quote:
I'm fairly confident the residuals I'm getting are accurate, especially since the memory is downclocked to 9000 on BOTH Linux/Windows. For example, I have ran a DC on M102543227 and the results have been verified. Also currently running another DC on a similar exponent. Hopefully, the residues will match up as well. Even with the piss poor FP64 performance, The card appears to be functioning quite well. If only this card would have the SAME FP64 ratio as the Titan Black, ( 1/3 ), it would be a buzzsaw  and of course priced up to a K80 Tesla, if not more. 
|
|
|
|
|
|
2016-08-27, 03:52
|
#2508 |
|
Mar 2010
3×137 Posts |
Cheers!
Some benchmark data with the new version below. It looks consistent with airsquirrels' timings (3.65ms/iter on same GPU and Mersenne candidate), but is a bit faster due to factory overclocking. Not quite sure why, but the drivers do not select the max clock profile for the memory, so I had to manually overclock it to default value (5.006GHz) Robert_JD, would you kindly benchmark your Titan X on M76042667 with 4096K FFT size, for comparison? Here are the DLLs, if anyone else wants to give this new binary a go: http://www108.zippyshare.com/v/Wg0AFopm/file.html Code:
| Aug 27 06:44:21 | M76042667 34000 0x166818e8f1bcad6d | 4096K 0.21875 3.3098 6.61s | 2:23:46:54 0.04% | | Aug 27 06:44:28 | M76042667 36000 0x41398e04d2474533 | 4096K 0.21875 3.3072 6.61s | 2:23:40:16 0.04% | | Date Time | Test Num Iter Residue | FFT Error ms/It Time | ETA Done | | Aug 27 06:44:34 | M76042667 38000 0x6c6d2d1ec54c8443 | 4096K 0.21875 3.3109 6.62s | 2:23:34:35 0.04% | | Aug 27 06:44:41 | M76042667 40000 0xa1b6980c147b1945 | 4096K 0.21875 3.3098 6.61s | 2:23:29:23 0.05% | | Aug 27 06:44:47 | M76042667 42000 0x3952b7b74e553555 | 4096K 0.21875 3.3079 6.61s | 2:23:24:33 0.05% | Last fiddled with by Karl M Johnson on 2016-08-27 at 03:57 |
|
|
|
 |
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 |
