windows smp.... very slow

[enhanced08]

ok, so i just installed the smp client on friday and went away for the weekend, came back today and found it still on the same WU! its only 37% done! its takin about 1.5hours per frame. i saw the thread where members posted frame times and from that it looks like i should be around 30min a frame. i have an amd 4800+ x2 @ 2676mhz. task manager is showing 4 cores running, but i would think it should only be 2, 1 per core. got any ideas?

 

[Jon855]

ok, so i just installed the smp client on friday and went away for the weekend, came back today and found it still on the same WU! its only 37% done! its takin about 1.5hours per frame. i saw the thread where members posted frame times and from that it looks like i should be around 30min a frame. i have an amd 4800+ x2 @ 2676mhz. task manager is showing 4 cores running, but i would think it should only be 2, 1 per core. got any ideas?

Make sure you aren't running it as a service. And you are suppose to have four cores running. The SMP clients have four cores running at any given time. and the the mpiexec.exe should be at 0%

I'm sure once you set it to run as an app, by setting a shortcut of it to the start up and what not, it should start up automatically after rebooting.

Hope that helps.

 

[enhanced08]

ok, well all 4 cores are showing about 20-30% cpu usage and the mpiexec.exe is at 0%. the client is not installed as a service and must be opened via clicking the fah.exe file.

 

[Killer[MoB]]

Make sure SSE boost shows as OK in the log. If it is, then try changing your team number to 33. Proteins always fold faster the [H]ard way.

 

[THL3]

Also only 1 instance of mpiexec.exe not 2, and that csrss.exe isn't going nuts.

 

[enhanced08]

ok, i got 1 instance of mpiexec.exe. SSE is running, and i tried changing the team number, but it gave me some error that said "team not sufficient, team pcmech > [H]" or something similar

but anyway, im gonna restart the client and see if that does anything...

 

[ND40oz]

ok, i got 1 instance of mpiexec.exe. SSE is running, and i tried changing the team number, but it gave me some error that said "team not sufficient, team pcmech > [H]" or something similar

but anyway, im gonna restart the client and see if that does anything...

If that's the case, you might want to try their forums for help

j/k, have you done a reboot, if so, without manually starting the client, do you see anything folding related running?

Also, is it installed as a service? Try disabling the service and manually restarting the client to see what your times are.

 

[APOLLO]

How does one know if they'll make the deadline? My dual [email protected] is taking nearly 50min a frame. Hope that's fast enough. Sheesh, to think that it wasn't that long ago such a setup was cutting edge...

 

[enhanced08]

restarting the client did nothing but after 2 system restarts it finally looks to be taking about 28min/frame. only 2 frames have finished tho, so it could be wrong.... we'll see what its doin come morning.

 

[APOLLO]

restarting the client did nothing but after 2 system restarts it finally looks to be taking about 28min/frame. only 2 frames have finished tho, so it could be wrong.... we'll see what its doin come morning.

I think it might be necessary to restart Windows. Mine was giving me errors until I restarted.

 

[vraa]

How does one know if they'll make the deadline? My dual [email protected] is taking nearly 50min a frame. Hope that's fast enough. Sheesh, to think that it wasn't that long ago such a setup was cutting edge...

Current Windows SMP projects

Project 2610: 1523 points, deadline 4 days
Project 2651: 1760 points, deadline 4 days
Project 2652: 1148 points, deadline 3 days

You can obviously do the math to see if you'll complete it in time or not.
I believe the 2610 project favors high-L2 cached endowed processors. So AMD's will be cutting it even closer.

Remember, the client is in beta right now and designed around four threads = four core systems or setups.

Since you're on a dual core you're going to be shorting yourself + the fact that you have an AMD and right now their cache limited which hurts them with the more complex WU's.

No sweat though, I've read that once the Windows SMP client is out of beta, some of the deadlines will be relaxed a little it. I really want to put my dual AMD 1800+'s back into the game

 

[APOLLO]

Current Windows SMP projects

Project 2610: 1523 points, deadline 4 days
Project 2651: 1760 points, deadline 4 days
Project 2652: 1148 points, deadline 3 days

You can obviously do the math to see if you'll complete it in time or not.
I believe the 2610 project favors high-L2 cached endowed processors. So AMD's will be cutting it even closer.

Well, my dual Barton didn't make it even though it was finishing frames every 49-50 minutes. If I have the math correct, it should have barely squeezed in at that rate. However, the client stalled overnight once and it hiccupped a few other times, which probably delayed it long enough to fail completing the WU before the deadline unfortunately.

Remember, the client is in beta right now and designed around four threads = four core systems or setups.

There are not that many users with a total of 4 or more processing cores in their systems. A fast enough dual core setup should make the deadline irrespective. Are you saying that the SMP client is solely targeted for an elite segment of the F@H community?

Since you're on a dual core you're going to be shorting yourself + the fact that you have an AMD and right now their cache limited which hurts them with the more complex WU's.

Well, my old Barton system was still managing significantly less than an hour a frame. The problem arises when something happens to either the client or the system that forces a restart and precious time is lost, especially when unattended. Otherwise, there shouldn't be a problem making the deadline.

No sweat though, I've read that once the Windows SMP client is out of beta, some of the deadlines will be relaxed a little it. I really want to put my dual AMD 1800+'s back into the game

So do I, because I can't really see it justified to run my older systems with the associated power consumption and frequent maintenance. I don't see why Stanford must have such tight deadlines for running 4 simultaneous threads. That requires a heck of a lot of processing power which doesn't come easy or cheap. Even one extra day will make a huge difference.

In any case, I just installed the SMP client on my dual uni-core Opteron system, my fastest system to date. It is not much faster in clockspeed than my older dual Barton (2.3GHz vs 2.2GHz) but it should be much better overall. Originally, I didn't want to install the SMP client on this system because I do everything else on it, and some of my apps utilize nearly all of the processing power of each processor. I can just imagine the slowdowns when I really multitask... Oh well, looking forward to a 'cheap' dual Barcelona (4X4), whenever that will be.