It has arrived

[Climhazard89]

Well just tossed it into main rig and im going to see how well it goes Kinda weird seeing 4 different command lines under task manager but whatever, it's still less then VMware took up. Also mines using 350megs vs VMware using 775megs

[frozndevl]

Originally Posted by Wuden

jebus 400 megs?

and I hear we need to run PW on our accounts now?

So if i have a powerfailure and my pc reboots it won't fold until I get to the machine to enter the Password ??

or is there way around that

You could create a scheduled task that starts when the computer boots up. I don't know how that would work yet, but I've got it setup on another comp I control. As soon as they get installing as a service working, you won't have that problem.

[Wuden]

don't forget to run install.bat

silly read me files

[AMDGUY]

What are the general thoughts for people running this so far?
Major complaints?/errors?
I see they have a whole sub-section over at Stanford, lol.

[frozndevl]

They've got a sub-section for every type of client over there. My initial reaction is that since I can get more ppd using VMWare and a 64-bit linux distro, I will do that on most rigs. However, I have had it running on my gf's rig, x2 3800, and it has been fine. It's a guaranteed good ppd instead of the crapshoot that two clients will deliver.

[NM_JJ]

Same here. I have a P-D which can't run under VMWare so I have to dual boot Ubuntu. I switch over if I'm going to be out of town a while. Otherwise, I let WinSMP run so I can use my computer.

[KillerBeagle]

Ultimately there will probably be a slight advantage in ppd to a native Linux client since it has a better MPI implementation. Right now Linux is "looking" better because it doesn't have project 2610, which is a killer on the WinSMP side. The other WinSMP projects 2651 and 2652 give ppd very close to the Linux client under VMware.

[NM_JJ]

Where do we find the points for each WinSMP protein? I looked on the Standord site and those proteins don't seem to show up. I found them once in the Stanford Forum, but now I don't remember where.

[KillerBeagle]

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

It's not on Stanford's site at all, but it can be found buried in a few threads at the folding community forum. The most recent one is here. Quite a few complaints have been made but they still haven't been added to the official project summary.

2610 is just plain nasty unless you have 4MB cache. The other two move along pretty well, similar to the Linux SMP projects.

Here are some comparisons from the same system:
e6300 @ 2.8GHz, Windows Vista 32-bit

Windows SMP client
p2610: avg 21mn 40s per frame, 1012 ppd
p2651: avg 16mn 05s per frame, 1576 ppd

Linux SMP client under VMware 1.01
p2604: avg 14mn 20s per frame, 1768 ppd
p2605: avg 14mn 00s per frame, 1810 ppd
p3025: avg 6mn 47s per frame, 1246 ppd
p3027: avg 7mn 27s per frame, 1384 ppd

[DriveEuro]

I was excited for the WinSMP client until I tried it. I was pulling about 3500-3700PPD culmative on my Q6600 with VMWare. On WinSMP I dropped to about 3100PPD. Decided to go back to VMWare.

What sucks is that Windows wasn't meant to run like this. A mod or developer at FAH community forums even said it.

Hopefully they can get it to mature some.

[frozndevl]

Just as an update. I was pulling about 480ppd with my x2 3800+ using the Windows SMP on p2610. I had some mpi issues last night, had to reinstall it. Now I grabbed a p2652 and I am getting 813ppd, 20:20/frame. I also manually set the affinity of the processes so that two of the fah cores are hitting one core and the other two are hitting the other. Reports say a 10-20% gain this way. I am unable to test it at this time as that rig doesn't like to shut down the Windows SMP so I don't to risk losing another WU.

x2 3800 @ 2.6
512mb RAM

More RAM would probably increase the ppd as well.

Edit:
Yeah, no difference in frame times so far.

[NM_JJ]

Let us know if you get an increase by comparing the same WUs. 2610 is known to be poor for ppd and 2652 is known to be good for ppd. If you do get a good increase let us noobs know how to set affinity.

[KillerBeagle]

Someone from Russia said 20% gain, but I haven't seen any confirmation. I tried it on p2610 on my E6300 and it made no difference. That makes sense since the large cache is shared. Maybe on a dual-core processor with separate caches it would help.

You can change the affinity on the fly from task manager - just go to the process view, right click on any of the Fahcore_a1.exe's, and choose "set affinity". Check only the CPU0 box for two fahcores and only CPU1 for the other two.

On a quad-core system, comparing VMware/Linux to Windows SMP isn't quite fair, since you're running 2 2-CPU VMs and running 2-core WU. p2604 & 2605 get much better ppd than the Linux 4-core WU like 3036. In a head-to-head comparison of a 4-core native Linux to a 4-core Windows SMP, Linux is probably a bit ahead because the MPI interface was originated on *nix OS's and is more efficient there. Unfortunately there are no "matching" Linux and Windows WU at the moment so the comparison isn't possible.

[Climhazard89]

Yeah...on that 20% difference. I saw maybe a 1% gain, which im betting is just the placebo effect going on there

[NM_JJ]

I read it only works on Hyperthreaded CPUs if there are like 4 virtual cores and 2 real cores. You have a 33% chance of assigning the second thread to a real core that is already busy. However if you set affinity you can force it to always go to the nonbusy core.