• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Has anyone tried CPDN on Zen2?

amd6502

amd6502

Senior member
I'm wondering how long they take on Zen2?

On my 9720p A12 it's taking just over 8 days @2.3ghz (for 2,417,459 GFLOPs; i.e., just under 3.5 GFLOPs/s/core). I probably should upgrade when Zen2 goes mobile.

Project url is: http://climateprediction.net/
 
Last edited:
Sorry for the delay, and thanks for stopping in!

I don't think anyone on the TeAm has run that yet on the newer hardware.

I am headed out the door for work, but I will start a few tasks up for you when I get home and check back in later.

Here is from Sunday's stats, the 3 names I know don't have Zen2's that I know of:

Climate Prediction overall position - 27
TeAm total for the week - 2,350,072
TeAm rank for weekly production - 4

__Credit/week _ UserName
1_______2,212,887____[H]Skillz
2_______70,674_______Hothead
3_______37,256_______Bremen
4_______18,238_______Fardringle
5_______12,178_______zzuupp
6_______759__________Ingleside
 
It took a while to receive tasks, but finally my 3800X got a batch of them.

It is running 15 tasks, and after 12 hours of running it estimates 3 Days-17 Hours to go. So not quite half the time as your CPU, but close.

I will continue to keep an eye on them to determine the actual run time, versus the estimate. But you can do the same (after they finish)
 
Last edited:
TennesseeTony said:
It took a while to receive tasks, but finally my 3800X got a batch of them.

It is running 15 tasks, and after 12 hours of running it estimates 3 Days-17 Hours to go. So not quite half the time as your CPU, but close.

I will continue to keep an eye on them to determine the actual run time, versus the estimate. But you can do the same (after they finish)
Click to expand...

Wow, awesome, thank you!! And your machine is running almost 8x the number of tasks (I only have two tasks running as this laptop is my only usable computer at the moment and it gets bogged down on multitasking if I run over two boinc tasks).

Right now i looks like my machine is going to take a total of just under 8 days and 13 hours to complete. It will be done in four days. (Amazingly this gives your single chiplet Zen2 a perfomance factor lead of ~ 16x divided by the CMT gain; an order of magnitude difference!)


Edit: found my work units:

climateprediction.net | University of Oxford

Website for climateprediction.net (CPDN), a volunteer computing project and the world’s largest climate modelling experiment
www.cpdn.org www.cpdn.org
 
Last edited:
FINALLY! Whew, that batch of tasks I tested has completed, 8 days and 20 hours (had to pause for a day to let some WCG finish before the deadline expired). I would say the estimated time to completion was off by a hair. 😀

amd6502 said:
...On my 9720p A12 it's taking just over 8 days...
Click to expand...

Was that the estimated time, or did they actually finish in that time?

My valid results (the rest of them will finish shortly). It's hard work and the pay is terrible. 🙂

EDIT: All my tasks are "UK Met Office HadAM4 at N144 resolution v8.09 i686-pc-linux-gnu"

These guys are getting much better pay and for less time with a different task set: (First one is a 3800X CPU)


 
Last edited:
  • Like
Reactions: IEC
TennesseeTony said:
FINALLY! Whew, that batch of tasks I tested has completed, 8 days and 20 hours (had to pause for a day to let some WCG finish before the deadline expired). I would say the estimated time to completion was off by a hair. 😀



Was that the estimated time, or did they actually finish in that time?

My valid results (the rest of them will finish shortly). It's hard work and the pay is terrible. 🙂

EDIT: All my tasks are "UK Met Office HadAM4 at N144 resolution v8.09 i686-pc-linux-gnu"

These guys are getting much better pay and for less time with a different task set: (First one is a 3800X CPU)


Click to expand...


Yeah the pay is terrible at CPDN!! I'm giving it a break as I'm about to upgrade my OS.

Last week I added the work unit links in the edit above.

Both my units took pretty much 700,000 seconds to complete (so over 5x longer than yours). My machine was up the entire 8+ days. I think for a project like CPDN a lot of progress is lost when people shut down their system or reboot.

You really loaded up you box with a ton of tasks. How about just 7 tasks, possibly with SMT turned off? (Would be a way to test SMT efficiency/scaling)
 
I have set CPU usage to 50%, poor man's way of SMT off, if you will, and am running two CPDN tasks, along with 6 WCG tasks (MIP, which is known to hammer the L3 cache). With only 6 WCG-MIP, the 3800X's L3 cache should have plenty of reserve (8-10MB).

Personal question: For what team if any do you crunch? If you're TeAm AnandTech, what is your user name (maybe I am blind and just didn't see you in the weekly stats). If for another team, that's fine too, we are all in this for the science! 🙂
 
TennesseeTony said:
I have set CPU usage to 50%, poor man's way of SMT off, if you will, and am running two CPDN tasks, along with 6 WCG tasks (MIP, which is known to hammer the L3 cache). With only 6 WCG-MIP, the 3800X's L3 cache should have plenty of reserve (8-10MB).

Personal question: For what team if any do you crunch? If you're TeAm AnandTech, what is your user name (maybe I am blind and just didn't see you in the weekly stats). If for another team, that's fine too, we are all in this for the science! 🙂
Click to expand...

I think a very large physical simulation like CPDN can really use a very large cache. There is so much grid point data that would benefit hugely from fitting within the L3. I think this is why in Zen3 the L3 was unified (rather one L3 per CCX times two CCX's).

So while Zen2 is HPC focused, Zen3 should see even greater progress.

Near the last few days of the computation I taskset the two CPDN tasks each to their own core, which prevents them hopping around from this to that core once to a few times a second (accumulating context change penalties as well as forcing the L2's one the two modules to constantly re-learn).

I pretty much don't accumulate any significant BOINC loads these days and I'm on a team of like slackers (team Russia, https://www.cpdn.org/team_members.php?teamid=94&offset=0&sort_by=expavg_credit ).
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top