Message boards : LHC@home Science : 64 Bit proccessing
Message board moderation

To post messages, you must log in.

AuthorMessage
jon

Send message
Joined: 30 Sep 05
Posts: 5
Credit: 29,115
RAC: 0
Message 20045 - Posted: 9 Sep 2008, 0:08:17 UTC

Is there any advantage in running an O/S in 64 Bit config with the workload in coming months,I run Ubuntu/Linux at moment in 32 Bit mode.
jon
ID: 20045 · Report as offensive     Reply Quote
Nuadormrac

Send message
Joined: 26 Sep 05
Posts: 82
Credit: 112,933
RAC: 0
Message 20071 - Posted: 9 Sep 2008, 21:22:21 UTC

Previously, I ran 64-bit Vista in it\'s beta 2 through it\'s RC days, and also a 64-bit Linux. My observations are that it was faster up to the point one ran out of memory. This is to say (mind you the apps were 32-bit), things appeared snappier and more responsive.

This said, BOINC is 32-bit typically and in any case the client files are. From a scientific standpoint if they had a 64-bit client, this could result in more precision being allowable in the calculations without a performance tole on the CPU. This said, type double (64-bit floats) are probably already being used; only thing is it takes a lil more for a 32-bit proc to calculate 64-bit floats then a 64-bit proc.

Memory is an issue however, depending on how much you have. Under Linux I didn\'t run out of so much, but on Vista, 1 GB RAM copuld be eaten awful fast. Once it swaps, the performance drops, and the 64-bit vista did use a ton of RAM. I never looked at Vista-32 to compare. As you\'re running, when I ran slamd64 on this box would probably be more representative; and the memory footprint on that was lower then win64.
ID: 20071 · Report as offensive     Reply Quote
jon

Send message
Joined: 30 Sep 05
Posts: 5
Credit: 29,115
RAC: 0
Message 20074 - Posted: 9 Sep 2008, 22:15:48 UTC - in response to Message 20071.  

Previously, I ran 64-bit Vista in it\\\'s beta 2 through it\\\'s RC days, and also a 64-bit Linux. My observations are that it was faster up to the point one ran out of memory. This is to say (mind you the apps were 32-bit), things appeared snappier and more responsive.

This said, BOINC is 32-bit typically and in any case the client files are. From a scientific standpoint if they had a 64-bit client, this could result in more precision being allowable in the calculations without a performance tole on the CPU. This said, type double (64-bit floats) are probably already being used; only thing is it takes a lil more for a 32-bit proc to calculate 64-bit floats then a 64-bit proc.

Memory is an issue however, depending on how much you have. Under Linux I didn\\\'t run out of so much, but on Vista, 1 GB RAM copuld be eaten awful fast. Once it swaps, the performance drops, and the 64-bit vista did use a ton of RAM. I never looked at Vista-32 to compare. As you\\\'re running, when I ran slamd64 on this box would probably be more representative; and the memory footprint on that was lower then win64.


OK thanks for reply.
I will stay with 32bit for the time being, my box has 4gig which I think is max for a 32bit system,No nont know anything about Vista, moved to linux about 3yrs ago, What a learning curve! but thats another story.
jon
ID: 20074 · Report as offensive     Reply Quote
Nuadormrac

Send message
Joined: 26 Sep 05
Posts: 82
Credit: 112,933
RAC: 0
Message 20079 - Posted: 10 Sep 2008, 0:58:21 UTC - in response to Message 20077.  
Last modified: 10 Sep 2008, 0:59:41 UTC

This said, BOINC is 32-bit typically and in any case the client files are.


There is a 64 bit version of BOINC for Linux, Mac and Windows. There are several projects that have 64 bit applications. Some of those are much faster than the 32 bit version, see ABC@home for example.




Yeah, but not typical; as the 32-bit version of BOINC is most often downladed, and most projects don\'t offer a 64-bit version of the project's app.

Allowing the proc to calculate things in 64-bit mode WOULD be significantly faster where type double and long doubles (80 bit floats) are used for certain fundamental reasons. But to get into this, one would have to get into the mechanics of how an x86 proc handles floating point, and specifically how it is handled in 32-bit mode.
ID: 20079 · Report as offensive     Reply Quote

Message boards : LHC@home Science : 64 Bit proccessing


©2020 CERN