Folding@Home Weekly Team News - 27th July 2006

Joe42 said:
Well the bigger the wus the better the ppd. So how much memory do the biggest wus take?
Is it under 256?
at the moment the largest only take around 100MB - at peak I think the QMD units used around 300MB but I think the servers would only issue them to clients with 500MB or more (to leave headroom for your system)

If you look on the server status page you'll see a column called "memory" which is the minimum memory a client must have available to receive work from it

http://vspx27.stanford.edu/serverstat.html

you'll see that most are currently set at 64MB with several around the 250MB mark (probably those using about 100MB physical RAM) and some using 300, 500 and even one looking for 1.5GB - the latter are most likely internal test machines
 
Well only having 512 would save a few pennies.
The problem with the buy a cheap cpu and overclock it strategy is you can't skimp on the board, you have to get a decent one.


Just been looking at the specs for the panaflo 120x38mm fans as i'm planning on getting another one to replace the nexus, and they do 50cfm @ 22dba, the nexus does 36cfm @ 22dba. That extra thickness makes a lot of difference.
I'm not too keen on hanging it off my Ninja tho, its heavy enough as it is. Have to see if i can rig up a mounting mechanism for it. That way i'll have two panaflos in a push pull config on the ninja. Shame the fins are so wide spaced, bit of a waste of the extra pressure.
 
512 should be fine for the foreseeable future. The most demanding WUs which have been publicly released are the QMDs which called for 504MB. The only problem would be if WUs like that were released again and you wanted to run two of them at once. But then again running two WUs which use that much memory at the same time is probably a bad idea anyway because they'd saturate the memory bus and you'd lose the performance improvement.

Although RAM is cheap these days :)
 
holy moly - these 364 pointers may only use about 100MB of ram but look at the size of the upload

[21:42:23] + Attempting to send results
[21:42:23] - Reading file work/wuresults_03.dat from core
[21:42:23] (Read 8549150 bytes from disk)
[21:42:23] Connecting to http://171.64.122.128:8080/

8.5MB??? :eek:
Well I guess that's what classes them as "Large" WUs then :cool:

edit: they're pretty huge to download in the first place, around 4MB - that's a fair few wasted minutes that the backup client will be loving :p

[09:28:28] Connecting to http://171.64.122.128:8080/
[09:28:39] Posted data.
[09:28:39] Initial: 0000; - Receiving payload (expected size: 4067893)
[09:29:26] - Downloaded at ~84 kB/s
 
Last edited:
Yeah, they take some uploading! In terms of points, though, nothing comes close to them across all platforms. I think they're actually pretty overvalued, but not complaining :cool:
 
lay-z-boy said:
Im out for 3 weeks,

Going to gozo/malta, need to make use of the holiday house.


yeah thats right, you have absolutely no idea where that is :p:D


bye.

Where, Malta and Gozo? Been there :p If you mean you're holiday house - not yet ;)

/stalker

I go away for nearly 5 weeks at a time and leave 6 rigs running - keep em on........
You know you should ;)
 
That's it, I've sucumbed to origami.

For the time being add me to the single crunchers list. One day I might get to assimilate the office PCs and server, but for the time being my stock Athlon 3500 XP will have to do.
 
Mattus said:
512 should be fine for the foreseeable future. The most demanding WUs which have been publicly released are the QMDs which called for 504MB. The only problem would be if WUs like that were released again and you wanted to run two of them at once. But then again running two WUs which use that much memory at the same time is probably a bad idea anyway because they'd saturate the memory bus and you'd lose the performance improvement.

Although RAM is cheap these days :)

you should have seen what happened when a dual xeon machine tried to run four at once through a single memory bus :eek:

machine died

HT
 
Back
Top Bottom