mersenneforum.org  

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

Reply
 
Thread Tools
Old 2017-03-26, 14:45   #23
Ducho
 
Nov 2016
Toronto, ON, CA

5 Posts
Default 29.1 b14 from TF to LL keeps using HT cores

Hi

I think that I found an issue with the new "Use HT" options.
Using 29.1 beta 14, Windows 10
Xeon 1650v4, DDR4-2400 ECC

So far it has been working fine with 28.10.
Then I moved to 29.1 beta 14.

Using default conf (TF with HT and LL without HT) as suggested.
I observed that:
1- if I have running TF workers, after the TF job finishes
2- and a LL is next in line,
3- LL worker will use HT, even if configuration is to not use HT.
Ducho is offline   Reply With Quote
Old 2017-03-26, 20:33   #24
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

748510 Posts
Default

Quote:
Originally Posted by Ducho View Post

Using default conf (TF with HT and LL without HT) as suggested.
I observed that:
1- if I have running TF workers, after the TF job finishes
2- and a LL is next in line,
3- LL worker will use HT, even if configuration is to not use HT.
I looked at the code and dont see anything obvious. I'll try to "trick" the code into thinking I have a hyperthreaded CPU and do some debugging. Anyway, I won't be able to post a fix for about a week.
Prime95 is offline   Reply With Quote
Old 2017-03-30, 04:43   #25
S485122
 
S485122's Avatar
 
Sep 2006
Brussels, Belgium

167010 Posts
Default Rolling average

It seems to me that the rolling average in this version differs a lot from the previous versions and it does not seem to correct itself quickly.
(For instance a CPU that used to have a rolling average of 2200 in version 28, completing a double check in 60 hours will stay with that value and estimate the work units will take only 24 hours off by 2,5 approximatively.)
While in the status and in the reports to PrimeNet the ETA is wrong, in the screen output the ETA is correct (of course. )

Jacob
S485122 is offline   Reply With Quote
Old 2017-04-01, 00:10   #26
tului
 
Jan 2013

22×17 Posts
Default

Ryzen 1800X arrives tomorrow. 29.1b14 is the recommended binary?
tului is offline   Reply With Quote
Old 2017-04-01, 19:25   #27
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

3·5·499 Posts
Default

Quote:
Originally Posted by tului View Post
Ryzen 1800X arrives tomorrow. 29.1b14 is the recommended binary?
Yes
Prime95 is offline   Reply With Quote
Old 2017-04-02, 04:33   #28
Yoshi24517
 
Yoshi24517's Avatar
 
Mar 2017

17 Posts
Default

While upgrading, I moved my current LL test files over (I'm LL testing a number in the 80 millions right now), and server files, and settings from the v28.10 folder to the v29.1 folder, and went over to my account, and had an unnamed CPU, so I had to merge the 2 CPUs (the one that said v28.10 with v29.1) and rename the unnamed one to the correct name. Just saying that might happen to you guys.
Yoshi24517 is offline   Reply With Quote
Old 2017-04-02, 09:38   #29
GP2
 
GP2's Avatar
 
Sep 2003

3×863 Posts
Default

I'm running 29.1 with what I presume to be build 14, on a c4.2xlarge instance on Amazon AWS with 4 cores.

When I start mprime from scratch and specify two workers, the default values for the number of cores are logically inconsistent and give a bad result:

Code:
Number of workers to run (1): 2

Code:
Options for worker #1

Type of work to get (0): 101
CPU cores to use (multithreading) (3):

Options for worker #2

Type of work to get (0): 101
CPU cores to use (multithreading) (3):

You have allocated more cores than are available.  This is likely to
GREATLY REDUCE performance.  Do you want to correct this? (Y):
I'm not sure why it proposes 3 cores per worker instead of 2. Even if it makes sense to propose 3 cores for the first worker, if that value is accepted then it should propose only 1 core for the second worker. Accepting both defaults leads to poor performance, as the scary error message implies. This will be pretty confusing to newbies. Maybe the program should keep track of a running total of how many cores the user has already allocated, and ensure it never exceeds the total cores?

I'm not sure if this is specific to 29.1 or if it occurred in previous versions, because in the past I simply edited the local.txt files directly. I don't know if it is specific to this particular type of machine, or if it happens with any 4-core machine.
GP2 is offline   Reply With Quote
Old 2017-04-02, 09:50   #30
GP2
 
GP2's Avatar
 
Sep 2003

3×863 Posts
Default

One other very minor issue:

Suppose you start mprime 29.1 in a directory that has no prime.txt or local.txt files, but does have a pre-existing worktodo.txt file.

Suppose the worktodo.txt file has entries for two workers, and when you run through the first-time menu initialization you also specify two workers.

Then for some reason, mprime will first convert the existing worktodo.txt file to single-worker format, and then fetch a new assignment for the second worker, rather than simply using the existing worktodo.txt file as-is.

Before:

Code:
[Worker #1]
DoubleCheck=...old assignment...

[Worker #2]
DoubleCheck=...old assignment...
After:

Code:
[Worker #1]
DoubleCheck=...old assignment...

;;MOVED;;[Worker #2]
DoubleCheck=...old assignment...

[Worker #2]
DoubleCheck=...new assignment!!!...
Again, I don't know if this happened in prior versions before 29.1, because I usually just edited local.txt and prime.txt directly to create two-worker versions, instead of running through the first-time initialization menus.
GP2 is offline   Reply With Quote
Old 2017-04-03, 04:01   #31
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

3·5·499 Posts
Default

Quote:
Originally Posted by Ducho View Post
Hi, I think that I found an "use HT" issue.

After finishing TF low limits task, LL started using hyperthreading cores.
HT Configured for TF but not for LL (as suggested).
Fixed in build 15.
Prime95 is offline   Reply With Quote
Old 2017-04-03, 04:19   #32
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

748510 Posts
Default

Quote:
Originally Posted by rudi_m View Post

29.1 build 8 was still slow enough for this CPU. Are there any options to run build 14 a bit slower? I have tried already HyperthreadTF=0.
You can try smaller values of UseMaxSieverAllocs. See undoc.txt. And be prepared to change it again when build 15 comes out.
Prime95 is offline   Reply With Quote
Old 2017-04-03, 04:50   #33
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

3·5·499 Posts
Default

Quote:
Originally Posted by Gimarel View Post
I get a SIGFPE when I try to use Advanced/Time in 29.1 build 14 for Linux 64-bit.
Fixed in build 15.
Prime95 is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Prime95 version 27.3 Prime95 Software 148 2012-03-18 19:24
Prime95 version 26.3 Prime95 Software 76 2010-12-11 00:11
Prime95 version 25.5 Prime95 PrimeNet 369 2008-02-26 05:21
Prime95 version 25.4 Prime95 PrimeNet 143 2007-09-24 21:01
When the next prime95 version ? pacionet Software 74 2006-12-07 20:30

All times are UTC. The time now is 14:48.

Sun May 9 14:48:30 UTC 2021 up 31 days, 9:29, 0 users, load averages: 1.75, 1.69, 1.67

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.