Discussion in 'Folding@Home' started by hat, Aug 30, 2009.
Anyone running SMP on an i7 with HT off? I was just wondering how long it takes to do a work unit.
why would you want to run a smp fold wu with ht off, it slows it down. Those big experimental wu's they came out with needs ht on to finish ontime for the bonus points. Right now i'm just gpu folding with 3 nvidia gpus, until i get linux squared away, then i'll be cpu folding
Because F@H only goes up to 4 threads, and the i7 has 4 cores with 8 threads, so it would be better to have only one thread per core to make sure each core is being used to the fullest.
and it does all 8 threads on a smp fold and hyperthreading does produce faster wu's
One instance will do all 8?
yep, those new wu's that came out will take full advantage of hyperthreading on I7 cpu's
Looks like the real PPD is run under a vmware/ubuntu installation. Found this @ the folding forum:
i7 PPD, threads and clients
Postby mephistopheles » Wed May 20, 2009 10:07 pm
There are reports on this forum saying that running more smp clients (up to 4) in VMWare on i7 increases performance.
I tried it out on my own system: i7 920 @ stock, 3x2GB RAM, native Ubuntu 9.04 (Ubuntu runs directly on the hardware, no VM is involved).
Here are the results:
* 1x smp-4: 5800 PPD
* 1x smp-8: 6100 PPD
* 1x smp-4 + 1x smp-8: 6800 PPD
* 2x smp-4: 7200 PPD
* 3x smp-4: 7200 PPD
* 2x smp-4 + 1 gpu: 7100 PPD
The numbers are for the CPU with hyperthreading enabled. Variation within a work unit is ~100 PPD, variation between two 1920 point WUs about the same.
* 2 smp-4 clients is more efficient than 1 smp-8
* when running native Linux, more than 2 smp-4 clients gives no benefit
* there is a (small) overhead to running the GPU client, I get 100-300 PPD performance loss from the CPU. Not sure if the overhead comes from the client or the Wine emulation.
Well, that would make sense. If you're running VMware Player/Server and limited to 2 CPU's per VM, then you'd probably get the most points out of 3-4 instances on an i7.
But are they showing the numbers of the new wu's they just came out with in july, gives you a bunus when finished under 24 hours, in which a regular quadcore processor just won't cut it unless you have dual processors running?
right, a non-HT quad wouldn't be able to finish the "big" work units in time to get the bonus, but they also cannot run the wu that require 8 cores/threads unless you're running two quads
Separate names with a comma.