mersenneforum.org  

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

Reply
 
Thread Tools
Old 2021-04-05, 15:15   #133
axn
 
axn's Avatar
 
Jun 2003

2×5×7×71 Posts
Default

I suspect that on a local system, you wouldn't be able to tell the difference. But anyways, George will have to confirm if this is even possible or not.
axn is offline   Reply With Quote
Old 2021-04-05, 20:20   #134
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

22·1,873 Posts
Default

Quote:
Originally Posted by axn View Post
George, Would it be possible to reduce the size of P-1 stage 2 checkpoint files? Using this with colab / google drive, it takes a very long time to stop/restart during stage 2 - it writes 100-200 MB of save file. I am guessing it is somehow saving the stage 2 prime bitmap or something? I think it would be faster to recompute the state, rather than load it from disk (with google drive).
Theoretically possible but I'm very afraid I'd make a mistake. Even the most subtle change in initial state can affect how the primes are paired.

You might try MaximumBitArraySize=n in prime.txt. This limits the maximum bit array size to n MB (default is 250). This will have some negative impact on performance, make a few runs yourself to see if it is significant.
Prime95 is offline   Reply With Quote
Old 2021-04-09, 06:31   #135
axn
 
axn's Avatar
 
Jun 2003

136A16 Posts
Default

Quote:
Originally Posted by Prime95 View Post
You might try MaximumBitArraySize=n in prime.txt. This limits the maximum bit array size to n MB (default is 250). This will have some negative impact on performance, make a few runs yourself to see if it is significant.
Is it safe to change this in the middle of a Stage 2 run?
axn is offline   Reply With Quote
Old 2021-04-09, 16:06   #136
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

22·1,873 Posts
Default

Quote:
Originally Posted by axn View Post
Is it safe to change this in the middle of a Stage 2 run?
Probably. Never tested.
Prime95 is offline   Reply With Quote
Old 2021-04-13, 08:40   #137
Happy5214
 
Happy5214's Avatar
 
"Alexander"
Nov 2008
The Alamo City

3×7×29 Posts
Default

P95/mprime really needs a way to manually force a proof upload. I periodically have Wi-Fi issues on my laptop (my primary GIMPS computer), and would like a way to upload proofs (I do PRP-CF, so there are quite a few of them) that have backlogged while the Wi-Fi has been out (especially if it will go out again soon). Perhaps add it to the manual communication menu?
Happy5214 is offline   Reply With Quote
Old 2021-04-15, 05:39   #138
Happy5214
 
Happy5214's Avatar
 
"Alexander"
Nov 2008
The Alamo City

60910 Posts
Default

An unrelated gripe. I have 5 workers on 8 cores on my laptop. They're never all running at the same time (I run 3 max at a time, 2 cores each). But the benchmarks still use all 5 workers on all 8 cores, a usecase that never happens and causes inaccurate timing due to the core overlap and throttling. It also causes issues on my desktop, as it runs benchmarks with 3 workers on 3 cores, an unused-in-practice scenario which causes it to overheat (it's an old Core 2 box). Can you add a setting to tune the automated benchmarking to a particular worker/core combination other than the full max?
Happy5214 is offline   Reply With Quote
Old 2021-04-15, 05:48   #139
axn
 
axn's Avatar
 
Jun 2003

2×5×7×71 Posts
Default

Quote:
Originally Posted by Happy5214 View Post
An unrelated gripe. I have 5 workers on 8 cores on my laptop. They're never all running at the same time (I run 3 max at a time, 2 cores each).
Then why not have just 3 workers (or 2 workers)? You're putting the program in an impossible situation. It'd be best to just turn off the benchmark altogether.
axn is offline   Reply With Quote
Old 2021-04-15, 06:49   #140
Happy5214
 
Happy5214's Avatar
 
"Alexander"
Nov 2008
The Alamo City

3×7×29 Posts
Default

Quote:
Originally Posted by axn View Post
Then why not have just 3 workers (or 2 workers)? You're putting the program in an impossible situation. It'd be best to just turn off the benchmark altogether.
Because I run 3 different work types in rotation (3 PRP-CF, 1 ECM, 1 P-1).
Happy5214 is offline   Reply With Quote
Old 2021-04-15, 08:02   #141
axn
 
axn's Avatar
 
Jun 2003

115528 Posts
Default

Quote:
Originally Posted by Happy5214 View Post
Because I run 3 different work types in rotation (3 PRP-CF, 1 ECM, 1 P-1).
Why not have just three workers and change their worktype as and when you want to switch over? Should be just as easy as bringing workers up/down.

Or ... Have two instances of P95 and keep PRP workers in one, and ECM/P-1 in another.

Anyway, with the current setup, your best bet is to just turn off automated benchmarking.
axn is offline   Reply With Quote
Old 2021-04-17, 06:56   #142
Falkentyne
 
Mar 2011

24 Posts
Default

Quote:
Originally Posted by Prime95 View Post
Probably. Never tested.
I may have found a "bug" with the FFT speed or size or iterations.

I don't know if it's the same for 10900k (it probably is), but on 11900k, if you disable FMA3, AVX512F and AVX in local.txt and run a small FFT stress test, each loop (multiple loops per FFT size though) finishes in 1 minute. Example, small FFT -->48k, finishes each loop in 1 minute. After several loops it goes to the next FFT.

If you disable AVX512, AVX2 and AVX in the stress test options instead, each loop lasts I think 2 or 3 minutes.

So the loop speed is different if you disable all the AVX settings in local.txt (from undoc.txt) versus disabling them in the stress test options.

Is this intended?
Falkentyne is offline   Reply With Quote
Old 2021-04-17, 08:17   #143
Happy5214
 
Happy5214's Avatar
 
"Alexander"
Nov 2008
The Alamo City

3×7×29 Posts
Default

I'm running a P-1 in stage 2, and it's saying it's 100% complete, but it's still running and printing "100% complete" reports at the normal rate. Is there a data issue or is this a problem with the printout that can be safely ignored without compromising the validity of the data until a fix is posted?
Happy5214 is offline   Reply With Quote
Reply

Thread Tools


All times are UTC. The time now is 00:34.

Wed May 19 00:34:07 UTC 2021 up 40 days, 19:14, 0 users, load averages: 2.23, 2.14, 2.02

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.