mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Software

Reply
 
Thread Tools
Old 2021-02-05, 03:43   #518
4EvrYng
 
Jun 2020

816 Posts
Default

How many FFTs are in this version, please? In other words if I ran stress test at 15 min per FFT how long it would take to complete one full cycle of all FFTs?
4EvrYng is offline   Reply With Quote
Old 2021-02-05, 04:40   #519
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

11·673 Posts
Default

Quote:
Originally Posted by 4EvrYng View Post
How many FFTs are in this version, please? In other words if I ran stress test at 15 min per FFT how long it would take to complete one full cycle of all FFTs?
A lot. The exact answer depends on whether you have AVX or AVX-512 instructions available. Note, there's no real benefit (that I know of) to testing every FFT length.
Prime95 is online now   Reply With Quote
Old 2021-02-05, 08:00   #520
Viliam Furik
 
"Viliam Furík"
Jul 2018
Martin, Slovakia

2·13·17 Posts
Default

Quote:
Originally Posted by 4EvrYng View Post
How many FFTs are in this version, please? In other words if I ran stress test at 15 min per FFT how long it would take to complete one full cycle of all FFTs?
I didn't bother to count them either, but it is at least one full day, possibly a week.
Viliam Furik is offline   Reply With Quote
Old 2021-02-05, 18:08   #521
4EvrYng
 
Jun 2020

810 Posts
Default

Quote:
Originally Posted by Viliam Furik View Post
I didn't bother to count them either, but it is at least one full day, possibly a week.
It is not a week. V 29.8B6 had 96 and it would take me around 24hrs to pass through one whole set at 15min each. I'm trying to find out did number of sets in this version increase and, if yes, by how much.

Does anybody know exact answer, please?
4EvrYng is offline   Reply With Quote
Old 2021-02-05, 19:00   #522
Viliam Furik
 
"Viliam Furík"
Jul 2018
Martin, Slovakia

44210 Posts
Default

Quote:
Originally Posted by 4EvrYng View Post
It is not a week. V 29.8B6 had 96 and it would take me around 24hrs to pass through one whole set at 15min each. I'm trying to find out did number of sets in this version increase and, if yes, by how much.

Does anybody know exact answer, please?
I think George, as well as me, was talking about all different implementations of FFT, of which there are indeed a lot. I will try to count them, somehow.

BTW, George, I have probably found a bug in benchmarking code (v30.4b6).
I have set it to run from 1K to 65536K, all complex FFTs, not all FFT implementations, 12 cores, 1 worker, no hyperthreading, 5 seconds per FFT.

EDIT:
To answer the question, I don't think the number of different FFTs changed, AFAIK the only big change was the addition of PRP proofs.
Attached Thumbnails
Click image for larger version

Name:	bug.png
Views:	34
Size:	90.4 KB
ID:	24272  

Last fiddled with by Viliam Furik on 2021-02-05 at 19:51
Viliam Furik is offline   Reply With Quote
Old 2021-02-05, 19:46   #523
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

3·19·83 Posts
Default

Quote:
Originally Posted by 4EvrYng View Post
It is not a week. V 29.8B6 had 96 and it would take me around 24hrs to pass through one whole set at 15min each. I'm trying to find out did number of sets in this version increase and, if yes, by how much.

Does anybody know exact answer, please?
George already answered you that the exact answer depends on your CPU. Without that information, he can't answer you- but you can run the test and report here for future interested parties!
VBCurtis is offline   Reply With Quote
Old 2021-02-05, 20:26   #524
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

11·673 Posts
Default

Quote:
Originally Posted by Viliam Furik View Post
BTW, George, I have probably found a bug in benchmarking code (v30.4b6).
I have set it to run from 1K to 65536K, all complex FFTs, not all FFT implementations, 12 cores, 1 worker, no hyperthreading, 5 seconds per FFT.
AVX FFTs only go to 51200K. This is the ungraceful (disgraceful?) error message you get trying to run an FFT above 50M.
Prime95 is online now   Reply With Quote
Old 2021-02-06, 20:43   #525
4EvrYng
 
Jun 2020

23 Posts
Default

Quote:
Originally Posted by Viliam Furik View Post
EDIT:
To answer the question, I don't think the number of different FFTs changed, AFAIK the only big change was the addition of PRP proofs.
Thank you!
4EvrYng is offline   Reply With Quote
Old 2021-02-07, 23:57   #526
DrobinsonPE
 
Aug 2020

1308 Posts
Default

I have a CPU that slowed down significantly after doing a benchmark (see below). I do not remember this ever happening before. Is this something where it will speed back up after the next benchmark?

This is a new processor for me. It finished an LL-D on a 60M exponent and then started on a 109M exponent. the pasted text below starts right after the P-1 ended and the PRP starts.

Linux Mint 20.1, mprime V30.3b6, i7-9700K, B365M DS3H, 16GB DDR4-2666

Code:
[Work thread Feb 8 07:37] Setting affinity to run helper thread 6 on CPU core #7
[Work thread Feb 8 07:37] Setting affinity to run helper thread 5 on CPU core #6
[Work thread Feb 8 07:37] Setting affinity to run helper thread 4 on CPU core #5
[Work thread Feb 8 07:37] Setting affinity to run helper thread 7 on CPU core #8
[Work thread Feb 8 07:37] Setting affinity to run helper thread 3 on CPU core #4
[Work thread Feb 8 07:37] Setting affinity to run helper thread 2 on CPU core #3
[Work thread Feb 8 07:37] Setting affinity to run helper thread 1 on CPU core #2
[Work thread Feb 8 07:37] Starting Gerbicz error-checking PRP test of M109831223 using FMA3 FFT length 6M, Pass1=1536, Pass2=4K, clm=1, 8 threads
[Work thread Feb 8 07:37] Preallocating disk space for the proof interim residues file p109831223.residues
[Work thread Feb 8 07:38] PRP proof using power=9 and 64-bit hash size.
[Work thread Feb 8 07:38] Proof requires 7.0GB of temporary disk space and uploading a 137MB proof file.
[Work thread Feb 8 07:39] Iteration: 10000 / 109831223 [0.00%], ms/iter:  7.121, ETA: 9d 01:13
[Work thread Feb 8 07:40] Iteration: 20000 / 109831223 [0.01%], ms/iter:  6.850, ETA: 8d 16:57
[Work thread Feb 8 07:41] Iteration: 30000 / 109831223 [0.02%], ms/iter:  6.851, ETA: 8d 16:56
[Work thread Feb 8 07:42] Iteration: 40000 / 109831223 [0.03%], ms/iter:  6.850, ETA: 8d 16:54
[Work thread Feb 8 07:43] Iteration: 50000 / 109831223 [0.04%], ms/iter:  6.853, ETA: 8d 16:58
..........
[Work thread Feb 8 11:05] Iteration: 1830000 / 109831223 [1.66%], ms/iter:  6.712, ETA: 8d 09:22
[Work thread Feb 8 11:06] Iteration: 1840000 / 109831223 [1.67%], ms/iter:  6.767, ETA: 8d 11:00
[Work thread Feb 8 11:07] Iteration: 1850000 / 109831223 [1.68%], ms/iter:  6.735, ETA: 8d 10:00
[Main thread Feb 8 11:08] Benchmarking multiple workers to tune FFT selection.
[Work thread Feb 8 11:08] Stopping PRP test of M109831223 at iteration 1859123 [1.69%]
[Work thread Feb 8 11:08] Worker stopped while running needed benchmarks.
[Main thread Feb 8 11:08] Timing 6144K FFT, 8 cores, 1 worker.  Average times:  7.85 ms.  Total throughput: 127.42 iter/sec.
[Main thread Feb 8 11:09] Timing 6144K FFT, 8 cores, 1 worker.  Average times:  7.58 ms.  Total throughput: 131.99 iter/sec.
[Main thread Feb 8 11:09] Timing 6144K FFT, 8 cores, 1 worker.  Average times:  7.60 ms.  Total throughput: 131.63 iter/sec.
[Main thread Feb 8 11:09] Timing 6144K FFT, 8 cores, 1 worker.  Average times:  7.52 ms.  Total throughput: 133.01 iter/sec.
[Main thread Feb 8 11:09] Timing 6144K FFT, 8 cores, 1 worker.  Average times:  7.40 ms.  Total throughput: 135.08 iter/sec.
[Main thread Feb 8 11:09] Timing 6144K FFT, 8 cores, 1 worker.  Average times:  7.70 ms.  Total throughput: 129.93 iter/sec.
[Main thread Feb 8 11:10] Timing 6144K FFT, 8 cores, 1 worker.  Average times:  7.31 ms.  Total throughput: 136.72 iter/sec.
[Main thread Feb 8 11:10] Timing 6144K FFT, 8 cores, 1 worker.  Average times:  7.32 ms.  Total throughput: 136.59 iter/sec.
[Main thread Feb 8 11:10] Timing 6144K FFT, 8 cores, 1 worker.  Average times:  7.15 ms.  Total throughput: 139.81 iter/sec.
[Main thread Feb 8 11:10] Timing 6144K FFT, 8 cores, 1 worker.  Average times:  7.43 ms.  Total throughput: 134.54 iter/sec.
[Main thread Feb 8 11:11] Timing 6144K FFT, 8 cores, 1 worker.  Average times:  7.11 ms.  Total throughput: 140.66 iter/sec.
[Main thread Feb 8 11:11] Timing 6144K FFT, 8 cores, 1 worker.  Average times:  7.07 ms.  Total throughput: 141.45 iter/sec.
[Main thread Feb 8 11:11] Timing 6144K FFT, 8 cores, 1 worker.  Average times:  7.59 ms.  Total throughput: 131.72 iter/sec.
[Main thread Feb 8 11:11] Timing 6144K FFT, 8 cores, 1 worker.  Average times:  7.14 ms.  Total throughput: 140.00 iter/sec.
[Main thread Feb 8 11:11] Timing 6144K FFT, 8 cores, 1 worker.  Average times:  7.28 ms.  Total throughput: 137.27 iter/sec.
[Main thread Feb 8 11:12] Timing 6144K FFT, 8 cores, 1 worker.  Average times:  8.05 ms.  Total throughput: 124.15 iter/sec.
[Main thread Feb 8 11:12] Timing 6144K FFT, 8 cores, 1 worker.  Average times:  7.50 ms.  Total throughput: 133.42 iter/sec.
[Main thread Feb 8 11:12] Timing 6144K FFT, 8 cores, 1 worker.  Average times:  7.05 ms.  Total throughput: 141.75 iter/sec.
[Main thread Feb 8 11:12] 
[Main thread Feb 8 11:12] Throughput benchmark complete.
[Work thread Feb 8 11:12] Benchmarks complete, restarting worker.
[Work thread Feb 8 11:12] Setting affinity to run helper thread 7 on CPU core #8
[Work thread Feb 8 11:12] Setting affinity to run helper thread 6 on CPU core #7
[Work thread Feb 8 11:12] Setting affinity to run helper thread 5 on CPU core #6
[Work thread Feb 8 11:12] Setting affinity to run helper thread 4 on CPU core #5
[Work thread Feb 8 11:12] Setting affinity to run helper thread 3 on CPU core #4
[Work thread Feb 8 11:12] Setting affinity to run helper thread 2 on CPU core #3
[Work thread Feb 8 11:12] Setting affinity to run helper thread 1 on CPU core #2
[Work thread Feb 8 11:12] Resuming Gerbicz error-checking PRP test of M109831223 using FMA3 FFT length 6M, Pass1=1K, Pass2=6K, clm=1, 8 threads
[Work thread Feb 8 11:12] PRP proof using power=9 and 64-bit hash size.
[Work thread Feb 8 11:12] Proof requires 7.0GB of temporary disk space and uploading a 137MB proof file.
[Work thread Feb 8 11:12] Iteration: 1859124 / 109831223 [1.69%].
[Work thread Feb 8 11:12] Iteration: 1860000 / 109831223 [1.69%], ms/iter:  6.919, ETA: 8d 15:31
[Work thread Feb 8 11:14] Iteration: 1870000 / 109831223 [1.70%], ms/iter:  7.067, ETA: 8d 19:56
[Work thread Feb 8 11:15] Iteration: 1880000 / 109831223 [1.71%], ms/iter:  7.015, ETA: 8d 18:21
[Work thread Feb 8 11:16] Iteration: 1890000 / 109831223 [1.72%], ms/iter:  7.014, ETA: 8d 18:18
[Work thread Feb 8 11:17] Iteration: 1900000 / 109831223 [1.72%], ms/iter:  7.034, ETA: 8d 18:53
[Work thread Feb 8 11:18] Iteration: 1910000 / 109831223 [1.73%], ms/iter:  7.017, ETA: 8d 18:20
[Work thread Feb 8 11:20] Iteration: 1920000 / 109831223 [1.74%], ms/iter:  7.286, ETA: 9d 02:24
[Work thread Feb 8 11:21] Iteration: 1930000 / 109831223 [1.75%], ms/iter:  7.038, ETA: 8d 18:57
[Work thread Feb 8 11:22] Iteration: 1940000 / 109831223 [1.76%], ms/iter:  7.023, ETA: 8d 18:28

Last fiddled with by DrobinsonPE on 2021-02-08 at 00:02 Reason: adding information
DrobinsonPE is offline   Reply With Quote
Old 2021-02-08, 07:08   #527
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
Jun 2011
Thailand

100100100101112 Posts
Default

Quote:
Originally Posted by 4EvrYng View Post
How many FFTs are in this version, please? In other words if I ran stress test at 15 min per FFT how long it would take to complete one full cycle of all FFTs?
An "all possible ways and combinations" benchmark (not stress test) with 5 seconds per test took about 46 hours in my computer (i7-6950x).
LaurV is offline   Reply With Quote
Reply

Thread Tools


All times are UTC. The time now is 05:01.

Mon Apr 12 05:01:14 UTC 2021 up 3 days, 23:42, 1 user, load averages: 2.31, 1.83, 1.77

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.