Windows SMP client is now open beta.

hmpf. and I have spent all this time getting my windows vmware SMP clients going...
 
Can't run this as a service yet (w/o work arounds).

4x threads, runs just fine on dual core machines.

100% CPU usage unlike the nix client, is some hit on PPD on nix vs. Win clients-- no hard data to back that up yet.

~ 225+ meg RAM usage

Follow the directions for the mpi install stuff, #1 problem causer I think. Need to have password on user account this is run under.

In essance it's kinda like the GPU client with the command box always up. I have not looked into the work around for the service yet.

Thanks for the info Macaholic, you prick.

:p
 
marty9876 said:
~ 225+ meg RAM usage
Click to expand...

Shit. Anyone have two 1gig sticks of DDR2-PC4200 they're looking to sell? I'm already thrashing this machine, another 225 megs and I'll have to bury this disk cause of all the swapping :(

Meanwhile, no work assigned yet.
 
According to the readme, running as a service is not currently supported, this is what I did to get around that:

Here is fah.bat:
Code: 
start cmdow.exe /RUN /HID fah.exe -verbosity 9 -forceasm -advmethods

Place cmdow.exe in the SMP directory along with fah.bat. Put a shortcut to fah.bat in your startup. I had issues doing the whole thing in a shortcut, it wasn't starting in the proper path and thus not using the config files I had already made.

This will start fah and hide the window. No need to have a dos prompt on your taskbar.

cmdow.exe: http://www.commandline.co.uk/cmdow/


Project: 2651 (Run 0, Clone 59, Gen 3)
unitinfo.txt
Code: 
Current Work Unit
-----------------
Name: Protein in POPC
Tag: -
Download time: March 18 00:29:26
Due time: March 22 00:29:26
Progress: 0%  [__________]


Edit: How to unhide and re-hide the SMP window:
hide.bat
Code: 
cmdow.exe "C:\Program Files\Folding@Home Windows SMP Client V1.01\fah.exe" /HID
unhide.bat
Code: 
cmdow.exe "C:\Program Files\Folding@Home Windows SMP Client V1.01\fah.exe" /VIS
 
Sweet! Even though I just got vmware up yesterday :rolleyes:

Is this giving out the same wu's as the nix smp client?
 
SpoogeMonkey said:
Sweet! Even though I just got vmware up yesterday :rolleyes:

Is this giving out the same wu's as the nix smp client?
Click to expand...

This is what is currently running on my Linux SMP box:
[19:47:12] Project: 2605 (Run 6, Clone 93, Gen 17)
unitinfo.txt
Code: 
Current Work Unit
-----------------
Name: Protein in POPC
Tag: -
Download time: March 17 19:46:53
Due time: March 21 19:46:53
Progress: 10%  [|_________]
 
If I only had a Dual Core Windows box now. Gotta get that 805 back up.
 
drizzt81 said:
hmpf. and I have spent all this time getting my windows vmware SMP clients going...
Click to expand...

That's not a bad thing..........Linux on a VM makes more points. Linux straight up makes a good bit more points.
 
SpoogeMonkey said:
Sweet! Even though I just got vmware up yesterday :rolleyes:

Is this giving out the same wu's as the nix smp client?
Click to expand...

No, the windows SMP work units are different from the ones we have been seeing on nix.

Current projects are:
P2610 - 1523 points, deadline 4 days
P2651 - 1760 points, deadline 4 days
P2652 - 1148 points, deadline 3 days
 
Just installed, working fine it looks like.

Crunching 2651 here, hasn't completed a frame yet but I'll post the times when I get them.


 
Killer[MoB];1030783088 said:
That's not a bad thing..........Linux on a VM makes more points. Linux straight up makes a good bit more points.
Click to expand...

Can you post data to support that?

I'd have to assume (key work assume...) that the WIN SMP client would at least equal VM/Nix client. I think the native nix still wins the PPD game, but WIN SMP ain't bad.

I heard tale of a dual dual Woodcrest 2.4 box pulling ~ 2340 PPD I think it was. 8-way Woodcrest 2.66's (dual quads) pull about 2500 PPD on each instance for 5K PPD total under native nix. These numbers are pretty close.
 
I'm a few percent in, here ya go:
Code: 
[00:45:00] Writing local files
[00:45:00] Completed 0 out of 500000 steps  (0 percent)
[00:58:51] Writing local files
[00:58:51] Completed 5000 out of 500000 steps  (1 percent)
[01:12:45] Writing local files
[01:12:45] Completed 10000 out of 500000 steps  (2 percent)

[email protected]. This is all while watching video on VLC and surfing the web.


Doing the math,
P2651 (1760) @ 13:51/f = 1830 PPD
 
From KillerMOB:
X2 3800+ (2 X 512 KB cache) @ 2.65GHz
P2651 (1760) @ 23:57/f = 1058 PPD (399 PPDpGHz)
P2652 (1148) @ 17:53/f = 924 PPD (348 PPDpGHz)

E6600 (4 MB shared cache) @ 3.33GHz
P2652 (1148) @ 11:19/f= 1460 PPD (438 PPDpGHz)

From BillR:
E6400 (2 MB shared cache) @ 3.0GHz
P2652 (1148) @ 14:00/f= 1180 PPD (393 PPDpGHz)

From Dew:
E6400 (2 MB shared cache) @ 3.2GHz
P2651 (1760) @ 13:51/f = 1829 PPD ( 571 PPDpGHz)

From Wraith28K:
Opty170 (2 X 1 MB cache) @ 2.65 GHz
P2651 (1760) @ 21:17/f = 1190 PPD (449 PPDpGHz)

From SpoogeMonkey:
Opty 165 (2 X 1 MB cache) @ 2.9GHz
P2651 (1760) @ 19:30/f= 1299 PPD (447 PPDpGHz)

From ND40oz:
Dual 5160 (4 MB shared per CPU) @ 3.0GHz
P2610 (1523) @ 8:04/f= 2718 PPD (453 PPDpGHz)
 
X2 is VM/native/Win SMP?

X2 comparasions to C2D never work out too well... :D
 
Killer[MoB];1030783126 said:
X2 @ 2650

P2651 (1760) @ 23:57/f = 1058 PPD (399 PPDpGHz)
P2652 (1148) @ 17:53/f = 924 PPD (348 PPDpGHz)

From Dew:
[email protected]
P2651 (1760) @ 13:51/f = 1829 PPD ( 571 PPDpGHz)
Click to expand...


That's good news. I see a HUGE surge in our PpD with the windows smp client. Which reminds me, where's Rich? This needs front page pimpage!!
 
marty9876 said:
X2 is VM/native/Win SMP?

X2 comparasions to C2D never work out too well... :D
Click to expand...

Windows SMP cleint on the X2 as you can see from the project numbers.
Not comparing per se, I just added Dew's (and anyone else's numbers to my post).
 
Killer[MoB];1030783181 said:
Windows SMP cleint on the X2 as you can see from the project numbers.
Not comparing per se, I just added Dew's (and anyone else's numbers to my post).
Click to expand...

10-4

I don't think anyone has doen a direct comparsion against native nix SMP and Windows SMP. VJ stated should be about a 10-20% hit on PPD from nix to Win but... nix uses ~80% of the box and Win is 100% under SMP.

Either way, SMP anything ='s boat load more PPD.
 
PSA:

The deadlines are nuts on SMP WU's. You need to leave your box on at least 12+ hours out of the day to meet deadlines on these. 2-3 day perfered and 3-4 final deadlines on these guys, some less. My P4D 3.0 finishes them with about 50% of the tiem remaining running 24/7. Turning the box off in the evenings would not fly.

If your running a notebook and traveling alot, tends not to work out. FYI hibernation borks the WU.

Frankly if your not running 24/7 on this, skip it.
 
Here's my first numbers, P2651 like I said. Surfing and burned a Linux CD, so that hurt my time a bit. System in signature.

[01:16:19] Protein: Protein in POPC
[01:16:19] Writing local files
[01:16:20] Extra SSE boost OK.
[01:16:21] Writing local files
[01:16:21] Completed 0 out of 500000 steps (0 percent)
[01:38:46] Writing local files
[01:38:46] Completed 5000 out of 500000 steps (1 percent)

And uh, if no one else is interested, I was planning on installing Ubuntu and the Linux SMP tonight anyway(before I saw the Windows SMP news)... I could do a direct comparison if ya'll would like.


 
Now that this thing is a open beta I would like to request someone try something and report back. I have been biting my tounge on this for a couple weeks but now I really would like someone to try running this Win SMP client on a dually AthlonXP box. I know there are some out there still folding and this could be a good boost for them if it works and they are capable of making the deadine. I think a pair of 2 ghz XPs could make it.

There is no mention of 64 bit requirements either in hardware or OS for the WinSMP client.

If it works, it could brings some of those good old asus and msi duallys out of retirement for someone.
 
Wraith28k said:
And uh, if no one else is interested, I was planning on installing Ubuntu and the Linux SMP tonight anyway(before I saw the Windows SMP news)... I could do a direct comparison if ya'll would like.
Click to expand...

It would be hard to directly compare because the work units are not the same. I would be very interested in a accurate time from your opty on that current work unit because it is running at the same clock as my test X2. I would like to see what kind of difference the extra cache on the opty makes. Thanks!
 
Wouldn't p2651 be a good comparison to p2604/2605 since they are both 1760 pointers?

I'll give this a shot tomorrow and compare times. I'm 60% done with a p2605 that should be done around 9am (guess I'll have to get up early to make sure I don't get started on another wu)
 
SpoogeMonkey said:
Wouldn't p2651 be a good comparison to p2604/2605 since they are both 1760 pointers?

I'll give this a shot tomorrow and compare times. I'm 60% done with a p2605 that should be done around 9am (guess I'll have to get up early to make sure I don't get started on another wu)
Click to expand...

be close enough for me! :D
 
Hope our Linux 64bit smp's dont take a dive in points now that the windows client it out....
 
I have always felt that because of the hardware requirements and the short deadlines that the bonuses will remain in place.
 
4DoorGTZ said:
Hope our Linux 64bit smp's dont take a dive in points now that the windows client it out....
Click to expand...

I've been debating that... I vote SMP points will take a huge wack not that wim SMP is out. Others swear it won't, the SMP clients return results are too valuable to mess with.

Something is coming per VJ himself so... ???

VJ is a happy, happy guy with the SMP stuff from what I hear.
 
Killer[MoB];1030783088 said:
That's not a bad thing..........Linux on a VM makes more points. Linux straight up makes a good bit more points.
Click to expand...

ICE_9 said:
If I only had a Dual Core Windows box now. Gotta get that 805 back up.
Click to expand...

Yep, what Killer[MoB] said, if points is your goal, keep running the linux smp client. I've got
both going right now....only reason I'm running the Win SMP client is because that's my game machine & I don't want to mess with dual boot.



 
Killer[MoB];1030783234 said:
Now that this thing is a open beta I would like to request someone try something and report back. I have been biting my tounge on this for a couple weeks but now I really would like someone to try running this Win SMP client on a dually AthlonXP box. I know there are some out there still folding and this could be a good boost for them if it works and they are capable of making the deadine. I think a pair of 2 ghz XPs could make it.

There is no mention of 64 bit requirements either in hardware or OS for the WinSMP client.

If it works, it could brings some of those good old asus and msi duallys out of retirement for someone.
Click to expand...

http://forum.folding-community.org/ftopic18545.html

your answer I think, Dual AMD 1800+ (1.53ghz) [I'm not an AMD guy so I don't even know off hand if them are the Athlons or not... :) ]

Project 2652 (Run 0, Clone 5, Gen 11)
It took me about 41 minutes to do 1 percent. At this rate I'll be done with this work unit in 2.85 days. Well within the 4 day deadline I've been reading about (for some workunits, I'm not sure if it applies to the one I'm crunching, to be honest I've never concerned myself with deadlines)

It seems to be working 100% okay so far, I'll be monitoring it very closely. The machien does nothing but very light gaming and folding, I have a MacbookPro as my main machine so if the F@H Staff needs me to run anything, let me know.
 
Ok, I've been running the client for a bit now, here's a better measure of my times.

Code: 
[01:16:10] Project: 2651 (Run 0, Clone 70, Gen 1)
[01:16:10]
[01:16:12] Entering M.D.
[01:16:18] Rejecting checkpoint
[01:16:19] Protein: Protein in POPC
[01:16:19] Writing local files
[01:16:20] Extra SSE boost OK.
[01:16:21] Writing local files
[01:16:21] Completed 0 out of 500000 steps  (0 percent)
[01:38:46] Writing local files
[01:38:46] Completed 5000 out of 500000 steps  (1 percent)
[02:01:00] Writing local files
[02:01:01] Completed 10000 out of 500000 steps  (2 percent)
[02:22:29] Writing local files
[02:22:29] Completed 15000 out of 500000 steps  (3 percent)
[02:43:50] Writing local files
[02:43:51] Completed 20000 out of 500000 steps  (4 percent)
[03:05:07] Writing local files
[03:05:07] Completed 25000 out of 500000 steps  (5 percent)
[03:26:24] Writing local files
[03:26:24] Completed 30000 out of 500000 steps  (6 percent)
[03:47:41] Writing local files
[03:47:41] Completed 35000 out of 500000 steps  (7 percent)

Looks like around 21 minutes per frame.. not sure how to work it out into PPD.

It's gotta be better than the horrendous 300PPD I've been getting with dual console clients.



 
Wraith28k said:
Looks like around 21 minutes per frame.. not sure how to work it out into PPD.

It's gotta be better than the horrendous 300PPD I've been getting with dual console clients.
Click to expand...

21:17 per frame making 1190 PPD (449 PPDpGHz)

Thanks again.
 
The curiosity got to me so I couldn't wait till tomorrow. I loaded up the win client and went to the grocery store to get some beers.

Opty 165 @ 2.9ghz 2x1gb ram
Code: 
[02:55:00] Project: 2651 (Run 0, Clone 76, Gen 1)
[02:55:00] 
[02:55:00] t)
[02:55:00] - Starting from initial work packet
[02:55:00] 
[02:55:00] Project: 2651 (Run 0, Clone 76, Gen 1)
[02:55:00] 
[02:55:01] Entering M.D.
[02:55:18]  packet
[02:55:18] 
[02:55:18] Project: 2651 (Run 0, Clone 76, Gen 1)
[02:55:18] 
[02:55:18] Entering M.D.
[02:55:24] Rejecting checkpoint
[02:55:25] Protein: Protein in POPC
[02:55:25] Writing local files
[02:55:26] Extra SSE boost OK.
[02:55:27] Writing local files
[02:55:27] Completed 0 out of 500000 steps  (0 percent)
[03:15:08] Writing local files
[03:15:08] Completed 5000 out of 500000 steps  (1 percent)
[03:34:50] Writing local files
[03:34:50] Completed 10000 out of 500000 steps  (2 percent)
[03:54:30] Writing local files
[03:54:30] Completed 15000 out of 500000 steps  (3 percent)
[04:14:09] Writing local files
[04:14:09] Completed 20000 out of 500000 steps  (4 percent)

19:30 / frame....only :30 more than the vm/nix client. I'm liking this alot more. My pc is so much more respopnsive again. The vm really takes over, where as the win client does not.


Wraith28k said:
Looks like around 21 minutes per frame.. not sure how to work it out into PPD.
Click to expand...

To calc ppd, use this formula:

protein pts x (1440 / (1% x 100))
1760 x (1440 / 2100) = 1206
1% = time to complete 1 frame
 
marty9876 said:
http://forum.folding-community.org/ftopic18545.html

your answer I think, Dual AMD 1800+ (1.53ghz) [I'm not an AMD guy so I don't even know off hand if them are the Athlons or not... :) ]
Click to expand...

Thanks Marty. That's my answer. There were a lot of dually AMD MPs and modded XPs built back in the day. Quite a few were 2 GHz or more so those would now be pretty good points producers.
His 1.53 GHz makes 403 PPD on the 2652. If I scale that out to a 2.1 GHz dually, that's 553 PPD. Not bad for a pair of XPs. ;)
 
Yeah, I'd say that's a bit better than my 395 PPD average right now.

I'm making more points per Ghz than I was total before.


SMP FTW!


 
It's taking me 8:04 per step with the 2610 project. Box is dual 5160s and I gave it a gig of ram in the setup.
 
You must log in or register to reply here.
Back
Top