Installation and running of mkfactc

[kruoli]

Originally Posted by daxmick

I've been running mfaktc for about 2 hours now. Here is a snapshot of the work as it progresses:

Date    Time | class   Pct |   time     ETA | GHz-d/day    Sieve     Wait
Jan 05 16:44 | 3852  83.4% |  8.257  21m53s |    217.72    82485    n.a.%
Jan 05 16:44 | 3853  83.5% |  8.337  21m57s |    215.63    82485    n.a.%
Jan 05 16:44 | 3856  83.6% |  8.164  21m22s |    220.20    82485    n.a.%
Jan 05 16:44 | 3873  83.8% |  8.211  21m21s |    218.94    82485    n.a.%
Jan 05 16:44 | 3877  83.9% |  8.585  22m11s |    209.40    82485    n.a.%
Jan 05 16:44 | 3880  84.0% |  8.234  21m08s |    218.33    82485    n.a.%
Jan 05 16:44 | 3885  84.1% |  8.152  20m47s |    220.53    82485    n.a.%
Jan 05 16:45 | 3888  84.2% |  8.248  20m54s |    217.96    82485    n.a.%
Jan 05 16:45 | 3892  84.3% |  8.263  20m48s |    217.56    82485    n.a.%
Jan 05 16:45 | 3897  84.4% |  8.238  20m36s |    218.22    82485    n.a.%
Jan 05 16:45 | 3900  84.5% |  8.245  20m29s |    218.04    82485    n.a.%
Jan 05 16:45 | 3901  84.6% |  8.133  20m04s |    221.04    82485    n.a.%
Jan 05 16:45 | 3912  84.7% |  8.187  20m03s |    219.58    82485    n.a.%
Jan 05 16:45 | 3913  84.8% |  8.140  19m48s |    220.85    82485    n.a.%
Jan 05 16:46 | 3916  84.9% |  8.165  19m44s |    220.17    82485    n.a.%
Jan 05 16:46 | 3921  85.0% |  8.198  19m41s |    219.29    82485    n.a.%
Jan 05 16:46 | 3925  85.1% |  8.277  19m44s |    217.20    82485    n.a.%
Jan 05 16:46 | 3928  85.2% |  8.144  19m16s |    220.74    82485    n.a.%
Jan 05 16:46 | 3933  85.3% |  8.153  19m10s |    220.50    82485    n.a.%
Jan 05 16:46 | 3936  85.4% |  8.150  19m01s |    220.58    82485    n.a.%

It appears to be working but I don't know if it is running efficiently. My GPU is at 99% and it appears I'm getting approx 220Ghz-d/day. Does that sound about right for a GeForce GTX 950?

If you look at this chart and search for "950", you'll see that 220 GHz*d/d is pretty good.

[daxmick]

Originally Posted by kruoli

If you look at this chart and search for "950", you'll see that 220 GHz/d is pretty good.

Nice graph. So... anyone got a couple Tesla 100v cards they could spare for me? LOL

[chalsall]

Originally Posted by daxmick

Nice graph. So... anyone got a couple Tesla 100v cards they could spare for me? LOL

Sure... Just sign up for Amazon, Google, and/or M$'s cloud computing.

Seriously, they all offer them. But they aren't cheap.

[TheJudger]

Originally Posted by daxmick

Nice graph. So... anyone got a couple Tesla 100v cards they could spare for me? LOL

Tesla V100 for what? The can do exactly 0.0000000000GHz equivalent in TF today!

Ofc. they are more fun for LL

Oliver

[chalsall]

Originally Posted by TheJudger

The can do exactly 0.0000000000GHz equivalent in TF today! Oliver

Any understanding as to why?

Your code was some of the best CUDA. And then it suddenly went to zero.

Weird.

[Mark Rose]

Originally Posted by daxmick

I've been running mfaktc for about 2 hours now. Here is a snapshot of the work as it progresses:

Date    Time | class   Pct |   time     ETA | GHz-d/day    Sieve     Wait
Jan 05 16:44 | 3852  83.4% |  8.257  21m53s |    217.72    82485    n.a.%
Jan 05 16:44 | 3853  83.5% |  8.337  21m57s |    215.63    82485    n.a.%
Jan 05 16:44 | 3856  83.6% |  8.164  21m22s |    220.20    82485    n.a.%
Jan 05 16:44 | 3873  83.8% |  8.211  21m21s |    218.94    82485    n.a.%
Jan 05 16:44 | 3877  83.9% |  8.585  22m11s |    209.40    82485    n.a.%
Jan 05 16:44 | 3880  84.0% |  8.234  21m08s |    218.33    82485    n.a.%
Jan 05 16:44 | 3885  84.1% |  8.152  20m47s |    220.53    82485    n.a.%
Jan 05 16:45 | 3888  84.2% |  8.248  20m54s |    217.96    82485    n.a.%
Jan 05 16:45 | 3892  84.3% |  8.263  20m48s |    217.56    82485    n.a.%
Jan 05 16:45 | 3897  84.4% |  8.238  20m36s |    218.22    82485    n.a.%
Jan 05 16:45 | 3900  84.5% |  8.245  20m29s |    218.04    82485    n.a.%
Jan 05 16:45 | 3901  84.6% |  8.133  20m04s |    221.04    82485    n.a.%
Jan 05 16:45 | 3912  84.7% |  8.187  20m03s |    219.58    82485    n.a.%
Jan 05 16:45 | 3913  84.8% |  8.140  19m48s |    220.85    82485    n.a.%
Jan 05 16:46 | 3916  84.9% |  8.165  19m44s |    220.17    82485    n.a.%
Jan 05 16:46 | 3921  85.0% |  8.198  19m41s |    219.29    82485    n.a.%
Jan 05 16:46 | 3925  85.1% |  8.277  19m44s |    217.20    82485    n.a.%
Jan 05 16:46 | 3928  85.2% |  8.144  19m16s |    220.74    82485    n.a.%
Jan 05 16:46 | 3933  85.3% |  8.153  19m10s |    220.50    82485    n.a.%
Jan 05 16:46 | 3936  85.4% |  8.150  19m01s |    220.58    82485    n.a.%

It appears to be working but I don't know if it is running efficiently. My GPU is at 99% and it appears I'm getting approx 220Ghz-d/day. Does that sound about right for a GeForce GTX 950?

There are a couple of settings at the bottom of the mfaktc.ini file that are worth playing with.

[daxmick]

Originally Posted by Mark Rose

There are a couple of settings at the bottom of the mfaktc.ini file that are worth playing with.

Are you just going to tease me or tell me what you are thinking...

[ET_]

Originally Posted by daxmick

Are you just going to tease me or tell me what you are thinking...

You can play with the sieve size and see if enhancing it you get any speedups...

[TheJudger]

Hey Chris,

Originally Posted by chalsall

Any understanding as to why?

Your code was some of the best CUDA. And then it suddenly went to zero.

Weird.

Yep, follow the link and read a little bit. Seems for each new HW family they f*ckup something in their compilers. With CUDA 9.0 and 9.1 mfaktc just fails every single testcase in the builtin selftest and refuses further work thus 0GHz equivalent today just because it doesn't work. They (nvidia) told me they found the bug and next CUDA release should fix this. For Volta I don't really care because those chips should do LL not TF! I'm more concerned about the nextgen consumer/gamer generation GPUs.

Oliver

[moebius]

Originally Posted by kruoli

If you look at this chart and search for "950", you'll see that 220 GHz*d/d is pretty good.

GTX560TI OC values for comparison

Date Time | class Pct | time ETA | GHz-d/day Sieve Wait
Jan 06 18:54 | 3080 66.8% | 2.478 13m10s | 219.72 82485 n.a.%
Jan 06 18:54 | 3083 66.9% | 2.399 12m43s | 226.96 82485 n.a.%
Jan 06 18:54 | 3087 67.0% | 2.350 12m25s | 231.69 82485 n.a.%
Jan 06 18:54 | 3092 67.1% | 2.379 12m32s | 228.86 82485 n.a.%
Jan 06 18:54 | 3095 67.2% | 2.348 12m20s | 231.89 82485 n.a.%
Jan 06 18:54 | 3096 67.3% | 2.315 12m07s | 235.19 82485 n.a.%
Jan 06 18:54 | 3108 67.4% | 2.322 12m07s | 234.48 82485 n.a.%
Jan 06 18:55 | 3111 67.5% | 2.320 12m04s | 234.68 82485 n.a.%
Jan 06 18:55 | 3116 67.6% | 2.321 12m02s | 234.58 82485 n.a.%
Jan 06 18:55 | 3120 67.7% | 2.317 11m58s | 234.99 82485 n.a.%
Jan 06 18:55 | 3131 67.8% | 2.313 11m55s | 235.39 82485 n.a.%
Jan 06 18:55 | 3132 67.9% | 2.331 11m58s | 233.58 82485 n.a.%
Jan 06 18:55 | 3140 68.0% | 2.322 11m53s | 234.48 82485 n.a.%
Jan 06 18:55 | 3143 68.1% | 2.313 11m48s | 235.39 82485 n.a.%
Jan 06 18:55 | 3147 68.2% | 2.337 11m53s | 232.98 82485 n.a.%
Jan 06 18:55 | 3152 68.3% | 2.312 11m43s | 235.50 82485 n.a.%
Jan 06 18:55 | 3155 68.4% | 2.314 11m41s | 235.29 82485 n.a.%
Jan 06 18:55 | 3168 68.5% | 2.324 11m42s | 234.28 82485 n.a.%
Jan 06 18:55 | 3171 68.6% | 2.324 11m40s | 234.28 82485 n.a.%

[masser]

Originally Posted by daxmick

Are you just going to tease me or tell me what you are thinking...

As mentioned, there are a few settings in the mfaktc.ini file that can be adjusted to improve performance; I've seen throughput improve by over 10% with my GT1030.
Congrats on getting the code compiled and enjoy the TF fun!
http://mersenneforum.org/showthread.php?t=21718