| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Active Research
Forum: Africa Rainfall Project
Thread: Work Available |
| Member(s) browsing this thread: dylanht |
|
Thread Status: Active Total posts in this thread: 3212
|
|
| Author |
|
|
geophi
Advanced Cruncher U.S. Joined: Sep 3, 2007 Post Count: 102 Status: Offline Project Badges: 
|
Eight hours? I'm getting 16-20 hours on my Threadripper. Is this project much faster on Redhat, or Intel, or possibly have you made an error? It's, in part, dependent on how many ARPs you are running at a time, and how many other tasks from other projects you might be running alongside. My Ryzen 3600X running only 4 ARPs was completing them in 6-6.5 hours. |
||
|
|
Dayle Diamond
Senior Cruncher Joined: Jan 31, 2013 Post Count: 452 Status: Offline Project Badges: 
|
For the record, I've limited ARP to 2 downloads per queue and MIP to 6.
I'm getting 16 hours on an Intel 6700K, and 20+hours on my 1950x threadripper. If I'm doing something wrong, I sure would love to discover what it is, because I could be running ARP plus eight MCM for the same time as I'm doing one ARP. |
||
|
|
Mike.Gibson
Ace Cruncher England Joined: Aug 23, 2007 Post Count: 12364 Status: Offline Project Badges: 
|
Dayle
The different projects all have different run times, arp being the longest. The checkpointing on arp which occurs every 12.5% is very intensive so should be restricted to a maximum of half of the threads of your machine. MIP also has constraints so should be restricted to one third of your threads. All projects currently running are readily available so no need for high cache settings. 1+1 will be more than enough. If you hold a bigger cache, you would delay others from getting their units validated. Mike |
||
|
|
Jean-David Beyer
Senior Cruncher USA Joined: Oct 2, 2007 Post Count: 335 Status: Offline Project Badges: 
|
Eight hours? I'm getting 16-20 hours on my Threadripper. Is this project much faster on Redhat, or Intel, or possibly have you made an error? If I add the time consumed and estimated time to completion, I am getting 8 hours 41 minutes for a current instance of ARP1. My machine really has an 8-core processor, but it is hyperthreaded and pretends to be 16 cores. I allow it to run a maximum of 8 BOINC processes, and it almost always does. I allow three ClimatePrediction hadam4-216 processes at a time and they take about a week to accomplish. They use 1.3 to 1.4 GBytes of RAM (each) and use lots of L3 cache. I allow at most two ARP1 processes at a time, although only one is running at the moment. My machine is like this: CPU type GenuineIntel Intel(R) Xeon(R) W-2245 CPU @ 3.90GHz [Family 6 Model 85 Stepping 7] Number of processors 16 Coprocessors --- Operating System Linux Red Hat Enterprise Linux Red Hat Enterprise Linux 8.2 (Ootpa) [4.18.0-193.28.1.el8_2.x86_64|libc 2.28 (GNU libc)] BOINC version 7.16.11 Memory 63943.89 MB Cache 16896 KB Swap space 15992 MB Total disk space 117.21 GB Free Disk Space 89.71 GB Measured floating point speed 6529.83 million ops/sec Measured integer speed 31889.23 million ops/sec  |
||
|
|
alanb1951
Veteran Cruncher Joined: Jan 20, 2006 Post Count: 953 Status: Offline Project Badges:
|
A general word of caution for those interested in throughput (ARP1 or otherwise); whilst we expect jobs like ARP1 and HST1 to be cache-hogs, and many folks will be aware that MIP1 is arguably even worse (despite a much smaller memory footprint than ARP1!), do we realize that OPN1 is also L3 cache intensive. Because of its relatively small footprint it doesn't show up too much on light work-loads, but in heavy work loads the L3 miss proportion can get as high as that of MIP1, and then it seems to pull down the performance of other tasks (as one might expect!)...
----------------------------------------The only really "memory access friendly" apps in the current collection seem to be SCC1 and MCM1; they'll just push CPU temperatures and power requirements up, and you can use the cache-hogs to drag those numbers back down again! Cheers - Al. P.S. Any Linux gurus who fancy using proc stat can see these performance influences for themseves - performance measurements independent of total run-time. (Windows and Mac users presumably have equivalent tools!) [Edited for word-choice] [Edit 3 times, last edit by alanb1951 at Jan 5, 2021 12:25:47 AM] |
||
|
|
Mike.Gibson
Ace Cruncher England Joined: Aug 23, 2007 Post Count: 12364 Status: Offline Project Badges:
|
I am temporarily not running arp so I would appreciate it if someone would post here when 044s started flowing and for subsequent iterations.
I can then keep the target updated. Mike |
||
|
|
PMH_UK
Veteran Cruncher UK Joined: Apr 26, 2007 Post Count: 769 Status: Offline Project Badges: 
|
My first 44 and last (original) 43 arrived on the 6th.
----------------------------------------Only running 1 each on 4 PCs so small sample. Paul.
Paul.
|
||
|
|
Mike.Gibson
Ace Cruncher England Joined: Aug 23, 2007 Post Count: 12364 Status: Offline Project Badges:
|
Paul
That was the earliest date that I expected. 045s might start on 11 January. Target for completion is now 3 May 2023. Mike |
||
|
|
Mike.Gibson
Ace Cruncher England Joined: Aug 23, 2007 Post Count: 12364 Status: Offline Project Badges:
|
My first 045 was today, so we will soon be at 24.6%, and my estimated completion is now June 2023.
Mike |
||
|
|
|