climateprediction.net (CPDN) home page
Thread 'Type of processor affecting PNW results?'

Thread 'Type of processor affecting PNW results?'

Message boards : Number crunching : Type of processor affecting PNW results?
Message board moderation

To post messages, you must log in.

AuthorMessage
old_user69295

Send message
Joined: 6 Apr 05
Posts: 17
Credit: 744,057
RAC: 0
Message 40995 - Posted: 9 Nov 2010, 20:17:27 UTC

First, let me say it really doesn't matter to me whether someone uses an Intel or AMD box, so no flames please. This is a simple observation that I am curious about.

I've noticed that on a high percentage of the PNW WU's ending with a computation error, the platform was Intel.

The problem is I don't know what the ratio of Intel to AMD users is among crunchers of PNW WU's. I guess I'm suggesting that the team might do a little data mining to see if there actually is some kind of trend.

If the ratio of Intel to AMD error WU's is the same as the ratio of Intel to AMD users in general, then I plan to have a cup of tea and go back to sleep.

If the error ratio is lopsided either way, maybe it's something to look into.

=Mike
ID: 40995 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 40996 - Posted: 9 Nov 2010, 20:20:10 UTC - in response to Message 40995.  

This is one of several differences that the project knows about, and which has been, and no doubt will continue to be, discussed on the beta test site.


Backups: Here
ID: 40996 · Report as offensive     Reply Quote
Profilemo.v
Volunteer moderator
Avatar

Send message
Joined: 29 Sep 04
Posts: 2363
Credit: 14,611,758
RAC: 0
Message 40997 - Posted: 9 Nov 2010, 21:41:52 UTC

There are no reasons for flames. CPDN has superb computers with both types of CPU crunching for it.

I think that if either AMD or Intel were more likely to crash any of the three regional model types we'd have noticed this on Beta. I have an Intel quad and two computers with AMD (three AMD cores) and have found all the regional model types very stable on all the computers.

My very approximate guesses are that CPDN has about 70-80% of computers with Windows, 15% with Linux and 10% Mac. I'd also guess that around 80% of the CPUs crunching CPDN are Intel.

Iain Inglis has collected statistics for completion rates on HadSM models, but by OS not CPU type.

Apart from FAMOUS whose parameter values mean that quite a few model crashes are inevitable and HadSM where some iceworlds are inherent in the models, I've found that most model crashes are caused by software misconfiguration (eg Linux owners not getting the 32-bit libraries or AVs blocking file access) or by user behaviour (usually not exiting from Boinc before shutting down the computer).
Cpdn news
ID: 40997 · Report as offensive     Reply Quote
ProfileIain Inglis
Volunteer moderator

Send message
Joined: 16 Jan 10
Posts: 1084
Credit: 7,827,799
RAC: 5,038
Message 40999 - Posted: 10 Nov 2010, 0:19:51 UTC

The figures when looking at HADSM3 slabs were:

Windows/Intel = 56%
Windows/AMD = 17%
Linux/Intel = 12%
Linux/AMD = 6%
Darwin = 10%.
ID: 40999 · Report as offensive     Reply Quote
Profilemo.v
Volunteer moderator
Avatar

Send message
Joined: 29 Sep 04
Posts: 2363
Credit: 14,611,758
RAC: 0
Message 41002 - Posted: 10 Nov 2010, 8:49:27 UTC
Last modified: 10 Nov 2010, 8:51:40 UTC

Thanks for the details Iain. I'd remembered the proportions by OS fairly accurately. I'd forgotten that for slab iceworlds you obviously had to take CPU type into account.

That must mean nearly 78% are Intel and nearly 23% Linux. I think these figures must be the proportions by core not by computer.

Linux people are more likely to have AMD than Windows users.
Cpdn news
ID: 41002 · Report as offensive     Reply Quote

Message boards : Number crunching : Type of processor affecting PNW results?

©2024 cpdn.org