Ryzen and Threadripper

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

To post messages, you must log in.

Previous · 1 . . . 57 · 58 · 59 · 60 · 61 · 62 · 63 . . . 69 · Next

AuthorMessage
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15184
Credit: 4,362,181
RAC: 3
Netherlands
Message 2044349 - Posted: 12 Apr 2020, 7:58:22 UTC - in response to Message 2044258.  

Simply solved with a script that hardcodes the usernames for all of the past posts without link. Can do the same for past postIDs as well, just going to be a lot of scrolling. Just need threadID and fipo of that thread is the first person who posts in the new thread. 😁
ID: 2044349 · Report as offensive     Reply Quote
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13727
Credit: 208,696,464
RAC: 304
Australia
Message 2044359 - Posted: 12 Apr 2020, 9:41:37 UTC - in response to Message 2044342.  

Yes I assumed that would be the case with the 2990wx but it is cheerfully running 60 threads at present so I can see how it responds. As you say the occasional thread runs out of memory but picks up later. Does the calculation program run to completion or for a set time period?
It's for a set time- 8 hrs is the default. It may bail out early (600 Decoys from 600 attempts seems to be the limit), but rarely goes much more than 5 min over. They do have a watchdog timer so if a Task goes 4 hours over the Target CPU runtime it will bail out then anyway.


Also the log shows that Rosetta often does not accept download requests with the response ' system running 97% of the time and would not finish in the time available' [I paraphrase].
Never had that particular response myself (but my systems are on 24/7 & i never suspend BOINC).
With the short deadlines & long Runtimes it's best to have a small cache, "Store at least 1 days of work" & "Store up to an additional 0.02 days of work" works for me doing Rosetta only. If you do other projects as well, it'd suggest "Store at least 0.5 days of work" or less.
They occasionally have dud Tasks, and it scrambles the Estimated completion time (huegly). Likewise releases of new applications result in very low Estimated completion times. The very small cache stops you from getting more than you can process and allows the Estimated completion times a chance to recover.


I am also having some problem with the team I joined - Isle of Man (Onchan). I do not live in Onchan and would rather that was dropped but whatever I cannot reconcile my membership with the output from Free-DC.
You could always join up with the Seti orphans team...
:-)
Grant
Darwin NT
ID: 2044359 · Report as offensive     Reply Quote
Profile ML1
Volunteer moderator
Volunteer tester

Send message
Joined: 25 Nov 01
Posts: 20254
Credit: 7,508,002
RAC: 20
United Kingdom
Message 2045319 - Posted: 17 Apr 2020, 0:25:49 UTC
Last modified: 17 Apr 2020, 0:27:04 UTC

AMD are adding a large chunk of EPYC compute in the name of pushing the frontiers of Science in the race against Coronavirus:


AMD Donates $15M of EPYC Processing Power To Fight Coronavirus
wrote:
Having recently joined the COVID-19 HPC Consortium, AMD has created a $15 million COVID-19 HPC fund to equip research institutions with the necessary computing power to find a cure for coronavirus and other diseases.

Research institutions will receive the best of what AMD has to offer – high-performance systems that leverage the chipmaker's core-heavy EPYC processors and speedy Radeon Instinct accelerators...



An interesting move... Might they join Boinc/Rosetta or Folding@Home?

Stay safe!

Keep searchin',
Martin
See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)
ID: 2045319 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5124
Credit: 276,046,078
RAC: 462
Message 2045388 - Posted: 17 Apr 2020, 13:11:09 UTC - in response to Message 2045319.  

AMD are adding a large chunk of EPYC compute in the name of pushing the frontiers of Science in the race against Coronavirus:


AMD Donates $15M of EPYC Processing Power To Fight Coronavirus
wrote:
Having recently joined the COVID-19 HPC Consortium, AMD has created a $15 million COVID-19 HPC fund to equip research institutions with the necessary computing power to find a cure for coronavirus and other diseases.

Research institutions will receive the best of what AMD has to offer – high-performance systems that leverage the chipmaker's core-heavy EPYC processors and speedy Radeon Instinct accelerators...



An interesting move... Might they join Boinc/Rosetta or Folding@Home?


My impression is there is a way for "Super Computers" to join Folding@Home as well as regular PC's. And F@H was getting alot of publicity.
But the announcement may actually be so that more "local processing" rather than distributed processing can be made available? Otherwise why target research institutions?

Tom
A proud member of the OFA (Old Farts Association).
ID: 2045388 · Report as offensive     Reply Quote
Profile ML1
Volunteer moderator
Volunteer tester

Send message
Joined: 25 Nov 01
Posts: 20254
Credit: 7,508,002
RAC: 20
United Kingdom
Message 2047925 - Posted: 2 May 2020, 23:02:27 UTC
Last modified: 2 May 2020, 23:06:22 UTC

AMD continues to lead the way whilst Intel burns their TDP with Marketing games:

Intel is offering (yet) more 14nm Skylake desktop processors, we repeat: More 14nm Skylake desktop processors
wrote:
... There are 32 combinations all told in this fresh batch of desktop-grade [Intel] processors, code-named Comet Lake, and too many to list here...

... Meltdown mitigations are built into the silicon, though variant 4 requires operating system support. The Spectre data-leaking flaws require software and firmware-level mitigations...

... Keep an eye on the TDP, too, as some will say they go up to 125W, others 65W, some 35W, and so on – yet you'll find, for example, the 125W parts may hit 250W in turbo mode....

... Meanwhile, around the same price points as the Comet Lake Cores, AMD's touting up to 12-core (24 thread) 7nm Ryzen parts clocked up to around 4.6GHz with faster RAM and PCIe 4. [All at far more honest and reasonable TDPs!]

Note also that AMD CPUs need far fewer security mitigations for a much smaller impact on performance.

All for a slam-dunk AMD win!?


So, why hasn't Intel patched up their silicon for all the security fixes for their new CPUs?!


Happy cool crunchin' - on whichever project!

Stay safe,
Martin
See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)
ID: 2047925 · 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 2047929 - Posted: 3 May 2020, 0:47:26 UTC - in response to Message 2047925.  

So, why hasn't Intel patched up their silicon for all the security fixes for their new CPUs?!

They can't. Or more to the point, not worth the complete microarchitecture redesign to remove the security leak points.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2047929 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5124
Credit: 276,046,078
RAC: 462
Message 2048127 - Posted: 4 May 2020, 18:44:08 UTC

@jsm

Well I have "joined" you in running a cpu-only cruncher :)
https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=4308859

Tom M
A proud member of the OFA (Old Farts Association).
ID: 2048127 · Report as offensive     Reply Quote
Profile Kissagogo27 Special Project $75 donor
Avatar

Send message
Joined: 6 Nov 99
Posts: 715
Credit: 8,032,827
RAC: 62
France
Message 2048141 - Posted: 4 May 2020, 21:30:35 UTC

\o/
ID: 2048141 · Report as offensive     Reply Quote
jsm

Send message
Joined: 1 Oct 16
Posts: 124
Credit: 51,135,572
RAC: 298
Isle of Man
Message 2048168 - Posted: 5 May 2020, 7:36:49 UTC - in response to Message 2048127.  

Well I am quietly adding computers to the Rosetta project as they finally run out of stored SETI tasks. Both myself and the team I have joined (isle of man (onchan)) are climbing up the tables faster than Jack up his Beanstalk. However despite having joined in the last month I have now started getting errors saying that this project is using an out of date URL. Remove project and rejoin. Now the 2990 has 60 (yes 60) threads running in 32gb memory and presumably all work on them would be lost - and for what? A better URL??
jsm
ID: 2048168 · Report as offensive     Reply Quote
Profile Wiggo
Avatar

Send message
Joined: 24 Jan 00
Posts: 34744
Credit: 261,360,520
RAC: 489
Australia
Message 2048169 - Posted: 5 May 2020, 7:50:22 UTC - in response to Message 2048168.  

Well I am quietly adding computers to the Rosetta project as they finally run out of stored SETI tasks. Both myself and the team I have joined (isle of man (onchan)) are climbing up the tables faster than Jack up his Beanstalk. However despite having joined in the last month I have now started getting errors saying that this project is using an out of date URL. Remove project and rejoin. Now the 2990 has 60 (yes 60) threads running in 32gb memory and presumably all work on them would be lost - and for what? A better URL??
jsm
That is not an error. Just set NNT and finish running your current tasks, once you have done that remove the Rosetta project and then reattach to the project just using the supplied new url (you should notice that the new url uses "https" whereas the old 1 is just "http"), I did mine yesterday morning here. ;-)

Cheers.
ID: 2048169 · Report as offensive     Reply Quote
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13727
Credit: 208,696,464
RAC: 304
Australia
Message 2048170 - Posted: 5 May 2020, 7:59:28 UTC - in response to Message 2048168.  

Now the 2990 has 60 (yes 60) threads running in 32gb memory and presumably all work on them would be lost - and for what? A better URL??
Set No New Tasks. When all Tasks are competed & reported, then Detach from the project, then Re-attach using the new HTTPS address.
Or just continue with things as they are.


And i would suggest you set a much smaller cache, say 0.4 & 0.02- you are trashing huge numbers of Tasks by missing the deadlines (the recently released new application won't have helped as it would have reset the Estimated completion times).

And also i'd suggest adding plenty more RAM if you want to run all those cores & threads without running in to out of memory issues, or reduce the number of threads processing work.
16GB wasn't enough for my 6c/12t system to avoid memory problems- some Tasks can use 3GB of RAM, 1-1.5GB for a Task isn't unusual.
I'd suggest for 60 threads you'd want at least 64GB of RAM as a bare minimum to avoid running in to RAM issues (70GB+ to be sure). And in the BOINC Computation settings, increase the amount of RAM available to BOINC to 95% or so.
Grant
Darwin NT
ID: 2048170 · Report as offensive     Reply Quote
jsm

Send message
Joined: 1 Oct 16
Posts: 124
Credit: 51,135,572
RAC: 298
Isle of Man
Message 2048175 - Posted: 5 May 2020, 9:25:55 UTC - in response to Message 2048170.  

Ok I will comply via NNT.
There doesnt seem to be a need to reduce the threads so far as there is only an occasional message saying waiting to run (or for memory) and it clears on that thread within a minute. Thus with 32GB they cannot be using substantial memory allocation. However I will monitor and reduce if the reports increase.
I had not changed the prefs from SETI operation where i very much needed as much in the cache as possible because the 2990 ran out of work and sulked every time SETI went down for maintenance. I will look at the cache shortly.
jsm
ID: 2048175 · 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: 22184
Credit: 416,307,556
RAC: 380
United Kingdom
Message 2048177 - Posted: 5 May 2020, 9:39:13 UTC

A quick observation on Rosetta's memory use - it is very dependant on the particular model and being worked on, some of the models are very frugal coming in at about 300k, while others are well over 1G.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 2048177 · Report as offensive     Reply Quote
jsm

Send message
Joined: 1 Oct 16
Posts: 124
Credit: 51,135,572
RAC: 298
Isle of Man
Message 2048222 - Posted: 5 May 2020, 18:54:33 UTC - in response to Message 2048177.  

Perhaps the allocation of tasks is a bit more sophisticated and takes available host memory into consideration?
I confirm that I have set all computers to NNT and will remove/rejoin as they MT the cache. I have also rest the prefs from their default on joining of 5;5 TO 1:1 for the time being and will monitor the effect.
With respect to the actual memory per task I looked at that on two of the computers and the average was around 300K.
jsm
ID: 2048222 · Report as offensive     Reply Quote
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13727
Credit: 208,696,464
RAC: 304
Australia
Message 2048229 - Posted: 5 May 2020, 19:36:35 UTC - in response to Message 2048222.  
Last modified: 5 May 2020, 19:46:17 UTC

With respect to the actual memory per task I looked at that on two of the computers and the average was around 300K.
Like Seti with VLARs & VHARs, different Tasks result in different system impacts.
My system RAM usage using all 6c/12t can vary between 10GB & 16GB, depending on the work mix. It's probably around 12GB most of the time. Even for Tasks that may use 3GB of RAM, that level of usage might only be for an hour or 2 (if that) of the overall 8 hour Runtime. But get 6 of them running at the same time, and that's 18GB just for them, let alone any others (and the systems own needs).
At least on your system when it runs out of RAM, the Tasks aren't erroring out, they just stop with a "Waiting for memory" message. For me when i had only 16GB of RAM, i would get 1 or 2 Tasks error out each day (and that was an improvement after allowing BOINC to use more RAM than the default settings do) till i upgraded to 32GB.
Grant
Darwin NT
ID: 2048229 · Report as offensive     Reply Quote
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15184
Credit: 4,362,181
RAC: 3
Netherlands
Message 2048416 - Posted: 7 May 2020, 16:37:02 UTC

Linus tests the new R3 3100X and 3300X versus the Intel comparables: https://www.youtube.com/watch?v=vD8Yk7JrBL8
ID: 2048416 · Report as offensive     Reply Quote
Profile ML1
Volunteer moderator
Volunteer tester

Send message
Joined: 25 Nov 01
Posts: 20254
Credit: 7,508,002
RAC: 20
United Kingdom
Message 2048623 - Posted: 9 May 2020, 2:15:00 UTC
Last modified: 9 May 2020, 2:15:53 UTC

So much for Intel Marketing flaunting unsustainable clock speeds...


See AMD fly away with:

AMD Ryzen 3 3100 Overclocked to Nearly 6 GHz
wrote:
Just how far can you push a budget CPU?...



All good fun!

Happy cool crunchin'!
Martin
See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)
ID: 2048623 · Report as offensive     Reply Quote
Profile Wiggo
Avatar

Send message
Joined: 24 Jan 00
Posts: 34744
Credit: 261,360,520
RAC: 489
Australia
Message 2048628 - Posted: 9 May 2020, 3:23:40 UTC

Ah yes, we'll all have to run our rigs on liquid nitrogen now. ;-)

Cheers.
ID: 2048628 · Report as offensive     Reply Quote
Bobzilla

Send message
Joined: 10 Dec 18
Posts: 3
Credit: 130,831
RAC: 0
United States
Message 2048656 - Posted: 9 May 2020, 11:26:30 UTC

Wanted to say thanks to those that helped me with info. I have built my Ryzen 3900x and to say it hauls ass is an understatement. What used to take me 2 days, now takes 5 hrs. Seti is down I guess and am sad to see it gone, but Einstein is up and running great!
Will make another build in next few months of same computer (or better) but for now, Gigabyte x570, 64GB ram 3200mhz 16cl, rx570 video card because I don't need awesome graphics for this ( next time 2070) and ryzen 9 (stock no over clocking) Enough Noctua fans to blow back most peoples hair, (bald guys excluded). If you want to see performance differences? Try going from a single core 1.8 mhz to a Ryzen 9.
All of this is built by me, the last computer I built was an Ibm 386. Ha I could probably get money for that thing if I had it laying around. I'm sure some museum would want to put that on display. Anyway, thanks for the help guys, sorry I don't remember names, not as young as I use to be, and it's been a while since I have been on here.

Two beer or not 2 beer... "Hell yes.. it's beer"!
ID: 2048656 · 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 2048683 - Posted: 9 May 2020, 15:33:54 UTC

Glad to see another Ryzen 3900 user. AMD really hit it out of the park for performance and value. I'm still waiting on my final Threadripper upgrade. Purchased the new Optimus block for my 2920X back in December. But the Covid-19 outbreak had impacted vendor supply lines and I am still waiting. Hope it gets here this month finally. Should be able to push on to my new overclock with its excellent cooling. Its younger sibling is doing outstanding cooling on the Ryzen 3900X and 3950X.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2048683 · Report as offensive     Reply Quote
Previous · 1 . . . 57 · 58 · 59 · 60 · 61 · 62 · 63 . . . 69 · Next

Message boards : Number crunching : Ryzen and Threadripper


 
©2024 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.