World Community Grid - View Thread

World Community Grid Forums

Category: Active Research

Forum: Africa Rainfall Project

Thread: Work Available

Quick Go »

Member(s) browsing this thread: 0x40

Thread Status: Active
Total posts in this thread: 3322

[ ]

Author

This topic has been viewed 3364074 times and has 3321 replies

MJH333
Senior Cruncher
England
Joined: Apr 3, 2021
Post Count: 268
Status: Offline
Project Badges:

50 year badge for Mapping Cancer Markers

10 year badge for Africa Rainfall Project

5 year badge for OpenPandemics - COVID-19


Re: Work Available

Mike,
101 is now priority.
Cheers,
Mark

[Nov 4, 2021 8:55:02 AM]

MJH333
Senior Cruncher
England
Joined: Apr 3, 2021
Post Count: 268
Status: Offline
Project Badges:


Re: Work Available

Mike,

I've just noticed that I have program versions 7.27 & 7.32, both ARP1.

I think alanb1951's post in this thread answers your question: https://www.worldcommunitygrid.org/forums/wcg/viewthread_thread,43868

Cheers,
Mark

[Nov 4, 2021 9:11:55 AM]

phytell
Cruncher
Joined: Sep 8, 2014
Post Count: 37
Status: Offline


Re: Work Available

The fastest system I have access to is currently averaging in the 12 hour range (11-14 or so) on 8 cores.
If we could assemble a system tuned for maximum performance for each single thread on current cutting edge hardware, I suspect 8 hours or faster would be viable, especially if it were overclocked (even my system has seen some 9-11 hour units).
Guessing beyond that is pure speculation on my part; maybe somebody else on here has a newer box they can provide statistics from.
Also keep in mind that the number of threads running matters; One core at a time will run much faster than 8, but you'll get a lot less real work done. That's why I don't have single-thread statistics for my box - I'm not going to have the rest of the CPU sit idle for long enough to gather them.

[Nov 4, 2021 2:44:39 PM]

Mike.Gibson
Ace Cruncher
England
Joined: Aug 23, 2007
Post Count: 12440
Status: Offline
Project Badges:

1 year badge for Human Proteome Folding - Phase 2

45 day badge for Discovering Dengue Drugs - Together

14 day badge for Nutritious Rice for the World

180 day badge for Help Fight Childhood Cancer

90 day badge for Help Cure Muscular Dystrophy - Phase 2

14 day badge for Discovering Dengue Drugs - Together - Phase 2

5 year badge for The Clean Energy Project - Phase 2

90 day badge for Computing for Clean Water

1 year badge for Drug Search for Leishmaniasis

180 day badge for GO Fight Against Malaria

45 day badge for Computing for Sustainable Water

20 year badge for Mapping Cancer Markers

5 year badge for Uncovering Genome Mysteries

5 year badge for Outsmart Ebola Together

5 year badge for FightAIDS@Home - Phase 2

2 year badge for Microbiome Immunity Project

10 year badge for OpenPandemics - COVID-19


Re: Work Available

phytell

ARP is the problem with regard to capacity. Not the others. What would be interesting would be using all threads but gradually increasing ARP threads and gradually decreasing the others.

That way there would be no wastage.

RAM is a problem for high ARP usage, but I have 20GB on an 8 thread machine.

Mike

[Nov 4, 2021 8:34:51 PM]

phytell
Cruncher
Joined: Sep 8, 2014
Post Count: 37
Status: Offline


Re: Work Available

@Mike:
In this instance, I'd expect a single arp thread running on an otherwise idle CPU to do significantly better than 1 arp task and 7 mcm (or opn, if you wish) due to turboboost.
The fastest single-thread performance happens when that's the only task given - the speed of each thread is reduced as more cores have work to do.
I might be able to beat my 12-hour average by bringing in other projects (this chip isn't hyperthreaded and I don't know how that changes the speed characteristics), but they'll still keep the CPU busy and the single arp thread won't achieve maximum clock speed.

All that said, I think the answer to the original question of whether or not folks needed to buffer for the maintenance was "yes, but not hugely". I buffered around 15 or so units before disabling networking, and had three left waiting to run when checking in afterward. Everything had returned to normal not long after that.

[Nov 5, 2021 5:52:30 PM]

adriverhoef
Master Cruncher
The Netherlands
Joined: Apr 3, 2009
Post Count: 2181
Status: Offline
Project Badges:

5 year badge for Human Proteome Folding - Phase 2

90 day badge for Nutritious Rice for the World

2 year badge for Help Fight Childhood Cancer

2 year badge for Help Cure Muscular Dystrophy - Phase 2

180 day badge for The Clean Energy Project - Phase 2

1 year badge for Computing for Clean Water

1 year badge for GO Fight Against Malaria

100 year badge for Mapping Cancer Markers

1 year badge for Uncovering Genome Mysteries

20 year badge for Outsmart Ebola Together

2 year badge for FightAIDS@Home - Phase 2

20 year badge for Smash Childhood Cancer

5 year badge for Microbiome Immunity Project

50 year badge for OpenPandemics - COVID-19


Re: Work Available

What is the fastest an ARP WU can be done?

I have seen one ARP1 today completing under 8 hours:

ARP1_0005761_103_0  Linux Ubuntu  P. Valid.  2021-11-05T07:37:54  2021-11-05T15:40:06    7.97/7.99     397.9/0.0   
	OS-Version: Ubuntu 20.04.2 LTS [5.8.0-63-generic|libc 2.31 (Ubuntu GLIBC 2.31-0ubuntu9.2)]
	Devicename: boinc-ripper
	Logfile:
	<core_client_version>7.16.6</core_client_version>
	<stderr_txt>
	INFO: Initializing
	INFO: No state to restore.  Start from the beginning.
	Starting WRFMain
	[04:32:05] INFO: Checkpoint taken at 2019-01-23_06:00:00
	[05:42:10] INFO: Checkpoint taken at 2019-01-23_12:00:00
	[06:48:03] INFO: Checkpoint taken at 2019-01-23_18:00:00
	[07:36:38] INFO: Checkpoint taken at 2019-01-24_00:00:00
	[08:30:38] INFO: Checkpoint taken at 2019-01-24_06:00:00
	[09:41:06] INFO: Checkpoint taken at 2019-01-24_12:00:00
	[10:44:28] INFO: Checkpoint taken at 2019-01-24_18:00:00
	[11:36:23] INFO: Checkpoint taken at 2019-01-25_00:00:00
	INFO: Simulation complete compressing output.
	11:37:50 (1255639): called boinc_finish(0)
	
	</stderr_txt>

[Nov 5, 2021 6:30:09 PM]

Mike.Gibson
Ace Cruncher
England
Joined: Aug 23, 2007
Post Count: 12440
Status: Offline
Project Badges:


Re: Work Available

Hyperthreading increases the output but decreases the speed of each unit. They would be somewhere between 20% & 50% slower, but doubling the units means an increase of output by between 33% & 66%.

As I think all the current projects are worthwhile, I use all threads with a combination of projects, restricting ARP to half my threads.

Mike

[Nov 5, 2021 6:58:29 PM]

Mike.Gibson
Ace Cruncher
England
Joined: Aug 23, 2007
Post Count: 12440
Status: Offline
Project Badges:


Re: Work Available

Generation 101 is now priority.

Mike

[Nov 6, 2021 1:40:04 PM]

Mike.Gibson
Ace Cruncher
England
Joined: Aug 23, 2007
Post Count: 12440
Status: Offline
Project Badges:


Re: Work Available

Just received my first generation 110.

Mike

[Nov 7, 2021 4:35:32 AM]

MJH333
Senior Cruncher
England
Joined: Apr 3, 2021
Post Count: 268
Status: Offline
Project Badges:


Re: Work Available

Mike,
102 is now priority.
Cheers,
Mark

[Nov 7, 2021 9:26:22 AM]

[ ]