Ryzen and Threadripper

Message boards : Number crunching : Ryzen and Threadripper
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 53 · 54 · 55 · 56 · 57 · 58 · 59 . . . 69 · Next

AuthorMessage
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2032744 - Posted: 17 Feb 2020, 1:07:46 UTC - in response to Message 2032738.  

I have been following this on my news feed. AMD swears there is no difference when using Win/Pro vs. Win/Pro for Enterprise and is now trying to "find" the bug.....

Toms Hardware (no relation) didn't find a difference. But the highly respected phoronix.com and anandtech.com did.

Tom

Probably the difference is due the choice of motherboards and memory used.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2032744 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5126
Credit: 276,046,078
RAC: 462
Message 2032750 - Posted: 17 Feb 2020, 1:53:41 UTC

If you want high thread counts but don't care if the CPU tasks crunch at very high speed consider a 16c/32t CPU like a 1950x on eBay for under $400 (Buy it now) or a 2950x for $600+

What you get is a previous generation CPU (and price) with a competitive total of Pcie lanes so you may be able to have lots of CPU threads/cores (aka Core Envy) and run as many gpu's as you can afford.

I have not tested the very high gpu count theory so don't say I promised you it would run 10+ video cards... :)

Tom
A proud member of the OFA (Old Farts Association).
ID: 2032750 · Report as offensive     Reply Quote
jsm

Send message
Joined: 1 Oct 16
Posts: 124
Credit: 51,135,572
RAC: 298
Isle of Man
Message 2034181 - Posted: 27 Feb 2020, 11:22:48 UTC

I have been holding back on a purchase of an 3990X as my experience with the 2990WX suggests that the W option is to be preferred. I cannot find any up to date mention of this flavour and I wonder if anybody has? I really want to see whether the thread block with the 2990 has been resolved so that I can crunch a full 128 threads on Seti (:
jsm
ID: 2034181 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2034191 - Posted: 27 Feb 2020, 14:22:07 UTC

Don't think there will be any version of the 3990 called "W". It is just the 3990. Plenty of reviews of the 3990 already to read.

Project has not changed its default recognition of 100 cpus maximum as far as I know.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2034191 · Report as offensive     Reply Quote
jsm

Send message
Joined: 1 Oct 16
Posts: 124
Credit: 51,135,572
RAC: 298
Isle of Man
Message 2034278 - Posted: 27 Feb 2020, 21:11:42 UTC - in response to Message 2034191.  
Last modified: 27 Feb 2020, 21:16:59 UTC

Even 100 would be interesting!
There was lots of references last year to both an X and a WX version but perhaps that was just hot air.
jsm
ID: 2034278 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2034315 - Posted: 28 Feb 2020, 0:25:28 UTC

Actually discovered I was wrong about how many cpus can be recognized by BOINC. Here is an Epyc 7742 with 64C/128T and BOINC reports it having 128 processors.
https://setiathome.berkeley.edu/show_host_detail.php?hostid=8887418
So you can safely go ahead with a Threadripper 3990X cpu and be able to run all threads on cpu tasks if you want to.

Plus with the cpu task limit raised to 150 now, you can even get all of them tasked with a job with the first download. (Granted the project needs to be able to supply work . . . which it is currently unable to do)
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2034315 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5126
Credit: 276,046,078
RAC: 462
Message 2034323 - Posted: 28 Feb 2020, 1:02:46 UTC - in response to Message 2034181.  

I have been holding back on a purchase of an 3990X as my experience with the 2990WX suggests that the W option is to be preferred. I cannot find any up to date mention of this flavour and I wonder if anybody has? I really want to see whether the thread block with the 2990 has been resolved so that I can crunch a full 128 threads on Seti (:
jsm


The reviews seem to agree that the memory model problem has gone away for this update. And depending on the flavor of the week sometimes we can download/crunch upto 150 on the cpus, sometimes we can't.

From an efficiency point of view usually a late model gpu gives you more output per unit. But as one of the myriad people with "core envy" I can't really fault you for wanting to crunch near 128 threads.

I upgraded to a 3950x for that very reason.

The only real question is how to maximize the "through-put" while bumping up against the 100/150 cpu task limit.

Tom
A proud member of the OFA (Old Farts Association).
ID: 2034323 · Report as offensive     Reply Quote
juan BFP Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 9786
Credit: 572,710,851
RAC: 3,799
Panama
Message 2034325 - Posted: 28 Feb 2020, 1:04:36 UTC

The question is: Boinc see the 128 threads, but it will allow the user crunch in all the threads?

Just for the fun to know.
ID: 2034325 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2034328 - Posted: 28 Feb 2020, 1:24:08 UTC - in response to Message 2034325.  

The question is: Boinc see the 128 threads, but it will allow the user crunch in all the threads?

Just for the fun to know.

I don't know why not? Unless BOINC code enumerating the processor count is completely isolated from the BOINC task execution code.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2034328 · Report as offensive     Reply Quote
juan BFP Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 9786
Credit: 572,710,851
RAC: 3,799
Panama
Message 2034333 - Posted: 28 Feb 2020, 1:43:36 UTC - in response to Message 2034328.  

I don't know why not?

Because the old 64 device limit. Does it applies to any device or only to the GPU?
ID: 2034333 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2034339 - Posted: 28 Feb 2020, 2:37:51 UTC - in response to Message 2034333.  

I don't know why not?

Because the old 64 device limit. Does it applies to any device or only to the GPU?

I thought it applied only to gpus.

Would have to walk the code to see if it applies to all devices.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2034339 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2034341 - Posted: 28 Feb 2020, 2:49:04 UTC - in response to Message 2034339.  

<max_ncpus>N</max_ncpus>
An upper bound on NCPUS (default: 64)


This is part of ProjectOptions. So this is set at each project via one of the configuration files.

https://boinc.berkeley.edu/trac/wiki/ProjectOptions

I thought I saw a piece of code in sched_config.cpp located at boinc-client_release-7-7.16/sched that said they were aware of the new multi-core devices.

const int MAX_NCPUS = 64;
    // max multiplier for daily_result_quota.
    // need to change as multicore processors expand


So how do we prove that you can get more than 64 cpu tasks running at a time on these new cpus?
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2034341 · Report as offensive     Reply Quote
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13949
Credit: 208,696,464
RAC: 304
Australia
Message 2034356 - Posted: 28 Feb 2020, 4:18:41 UTC - in response to Message 2034341.  

So how do we prove that you can get more than 64 cpu tasks running at a time on these new cpus?
Sounds like you need to buy one & see how it goes...
Grant
Darwin NT
ID: 2034356 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2034359 - Posted: 28 Feb 2020, 4:26:53 UTC - in response to Message 2034356.  

Ha ha ha LOL . . . . sure soon as I win the lottery.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2034359 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5126
Credit: 276,046,078
RAC: 462
Message 2034400 - Posted: 28 Feb 2020, 13:32:38 UTC - in response to Message 2034341.  


So how do we prove that you can get more than 64 cpu tasks running at a time on these new cpus?


I agree it would be nice if we all could go out and buy the 64c/128t systems (core envy). But I would note that the "Heavy Metal" https://setiathome.berkeley.edu/forum_thread.php?id=83455 thread has a sampling of systems with very high cpu thread counts that could be asked/queried and examined on this subject.

Plus this guy: https://setiathome.berkeley.edu/hosts_user.php?sort=expavg_credit&rev=0&show_all=0&userid=4735

Tom
A proud member of the OFA (Old Farts Association).
ID: 2034400 · Report as offensive     Reply Quote
juan BFP Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 9786
Credit: 572,710,851
RAC: 3,799
Panama
Message 2034401 - Posted: 28 Feb 2020, 13:37:09 UTC - in response to Message 2034359.  

Ha ha ha LOL . . . . sure soon as I win the lottery.

+1
ID: 2034401 · Report as offensive     Reply Quote
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22803
Credit: 416,307,556
RAC: 380
United Kingdom
Message 2034414 - Posted: 28 Feb 2020, 14:54:36 UTC - in response to Message 2034341.  

You need to look at the source code for the server, not the client, and be prepared for the same variable name to appear in several different context with subtly different meanings in each :-(
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 2034414 · Report as offensive     Reply Quote
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14690
Credit: 200,643,578
RAC: 874
United Kingdom
Message 2034416 - Posted: 28 Feb 2020, 15:24:33 UTC - in response to Message 2034414.  

You need to look at the source code for the server, not the client, and be prepared for the same variable name to appear in several different context with subtly different meanings in each :-(
The BOINC code jungle is not safe for exploration without a lot of cunning and a good GPS. Keith's snippet is best defined as

https://github.com/BOINC/boinc/blob/master/sched/sched_config.cpp#L43

- showing that it's in the /sched/ folder. That's the clearest indication that this is, in fact, server code.

It's also present - unchanged - in the breakout server branch:

https://github.com/BOINC/boinc/blob/server_release/1/1.2/sched/sched_config.cpp#L43

That's the supposedly verified and safe branch used to create what we know as server version 715 - and we all know where that went :-(

FWIW, the figure was increased from 16 to 64 ten years ago, in the completely unrelated (another bad habit of David's) https://github.com/BOINC/boinc/commit/392f02f8b8085b925b53fbf731a221c4f2b7a0d6 (scroll down to the very bottom).
ID: 2034416 · Report as offensive     Reply Quote
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22803
Credit: 416,307,556
RAC: 380
United Kingdom
Message 2034418 - Posted: 28 Feb 2020, 16:00:27 UTC

Thanks Richard.
And that just about nails it - we are stuck at 64 cores until someone climbs into the server code (with all the appropriate PPE as described by Richard - I think I'd add a very long bit of string secured to one's entry gate so if the worst happens you can always follow the string back to where you started, with luck).
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 2034418 · Report as offensive     Reply Quote
jsm

Send message
Joined: 1 Oct 16
Posts: 124
Credit: 51,135,572
RAC: 298
Isle of Man
Message 2034542 - Posted: 29 Feb 2020, 6:52:57 UTC - in response to Message 2034418.  

Noting this limitation appears to be specifically for Windows versions then what about Linux version code?
jsm
ID: 2034542 · Report as offensive     Reply Quote
Previous · 1 . . . 53 · 54 · 55 · 56 · 57 · 58 · 59 . . . 69 · Next

Message boards : Number crunching : Ryzen and Threadripper


 
©2025 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.