Celeron D VS Q9400

[Sticky]

Shouldn't one core on the Q9400 be faster than the Celeron D 2.6? They are running a nearly identical Amber unit worth 225 credits. Yet the Celeron is 2 PPD faster.

[cdolphin]

Test it out via super Pi to ensure you are correct. I would think it is only different points within the unit, small variance/etc.

[PatonB]

Unless a single core losses some speed as it shuffles around the cpu...

[johnerz]

as the q9400 supports it, have you thought of trying using notfred and vm ware?

[Sticky]

Originally Posted by cdolphin

Test it out via super Pi to ensure you are correct. I would think it is only different points within the unit, small variance/etc.

Tried it, a single core on the q9400 did:
512K @ 8

The Cel did:
512K @ 31

So the q9400 is practically 4 times faster.

I had 3 clients running. Two as console services and one as the Tray Client. The Celeron had a Console service running. I compared my Tray Client with the Console service. (Maybe thats where the issue is) Three cores where 100%, but maybe it was still shuffling the processes around on those three only. Not sure. Trying to get the SMP client working, but its giving me issues. Not sure if it will bog the system down. Work is more important. Also why im hesitant to try the VM route. I may give it a try though if i cant run smp native in xp.

[jevans64]

I had problems getting the Deino SMP client to work, so I just stuck with the MPICH version but that was when the Deino client was first released.

VMWare + Notfreds takes a little bit more work to get going properly but you'll get the same amount of points on TWO cores as you would get on FOUR cores with WinSMP. A WinSMP 1920-pointer will fetch about 2000 PPD at 4 * 3.2GHz and a Linux 1920-pointer will get 2500 PPD on 2 * 3.2GHz.

[Sticky]

Still havent figured out the winsmp issue. Took no time to setup the vm + notfreds though. It's currently running. Didn't get a chance to see the first few frames though. Will know how it's doing in the morning. I'll try to figure out the winsmp issue for the sake of i may not be able to keep running the vm setup and the that it may help out others as i've spent close to 5 hours learning how it doesn't work. Nice to know though that will will put out a lot more ppd.

[jevans64]

Originally Posted by Sticky

Still havent figured out the winsmp issue. Took no time to setup the vm + notfreds though. It's currently running. Didn't get a chance to see the first few frames though. Will know how it's doing in the morning. I'll try to figure out the winsmp issue for the sake of i may not be able to keep running the vm setup and the that it may help out others as i've spent close to 5 hours learning how it doesn't work. Nice to know though that will will put out a lot more ppd.

Well I'd say don't bother with WinSMP if you got VM+Notfreds running and aren't too peeved about the slowdown when running big tasks. One thing I did with my Notfred clients since I have so many of them is to give them a static IP address 10 greater than the host IP address -- which are also static. This is done in the same browser interface you use for changing the folding name and team number. I also have a process manager ( Process Lasso ) installed just to make sure the GPU2 clients get priority over the VM.

Just a note on the VM. The log file ( FAHlog.txt ) and work unit count is wiped out every time the client is stopped / restarted. I like to keep copies of the logs for reference, so I copy the text file to another location on the drive before I stop the client. Not really a big deal but you can do the same if you want to look at the logs at a later date. You'll also see this message ( attempt to access beyond end of device ) pop up in the VM window. This is a known error and doesn't interfere with folding.

There aren't too many different work units for Linux SMP so it is easy to gauge performance.

WinSMP a1 -- q9550 @ 3.166 -- p2665 -- 13m16s per frame -- 2084 PPD
VM SMP a2 -- q6600 @ 3.204 -- p2671 -- 5m06s per frame -- 5421 PPD