841 PPW from one cpu?

haysdb · Post by **haysdb** » Mon Jan 05, 2004 11:08 am

I'm sure the overclocking guys see these kinds of numbers all the time, but this is a first for me. My P4 2.66/533 machine is on a pace to produce 841 PPW according to F@H LogStats.

This is thanks to protein p803_p53dimer803. This is a 32.0 point Gromacs that this machine will crunch through in just over 6 hours. Give me more of these!

David

ColdFlame · Post by **ColdFlame** » Mon Jan 05, 2004 2:06 pm

Bah me! My brand spanking new Barton only got close to something like 750 PPW on that 32.0 point Gromacs. Maybe it even got close to 800 PPW I dont remember. But 2.66/533

Is it a hyper-threaded CPU? I feel spanked with my array of Athlon CPUs

Right now I'm enjoying 556 PPW on that Barton, on a stinky 60.0 point Gromacs.

By the way, I've noticed that the deviation in my PPW output is huge depending on what kind of proteins I've got. Sometimes I'd get only half of what I expect.

haysdb · Post by **haysdb** » Mon Jan 05, 2004 4:56 pm

Not Hyper-Threaded, no.

This is the first time in memory that the 2.6 has out-produced the 2.4 with HT enabled. It'd be interesting to see what the HT machine could do with a couple of these proteins.

David

Zyzzyx · Post by **Zyzzyx** » Mon Jan 05, 2004 5:00 pm

I've got a P4-2.2ghz currently working ~550ppw on a p845 protein. There have been a couple times I've seen it into the 600ppw range, and other times down below 350ppw. Quite the variance, but I don't mind... not like I can upgrade or tweak it much. Its my work machine at the office.

TRC-13 · Post by **TRC-13** » Tue Jan 06, 2004 10:31 am

Not bad at all! My Barton 3200+ with a 400MHz FSB only pounds out an estimated 841.46 PPW in FAH logstats. I do agree though that the estimate is only on this group of work units and only on some frames.

Protein p813_p53dimer813
Protein Core Gromacs
Credit 32.00
Deadline 8.00
Current Frame 90 of 100 (10 left )
Time Per Frame 3 mins, 50 sec
Time Left 38 mins, 20 sec
Client Version 4.00
Core Version Version 1.55 (December 22, 2003) SSE Enabled
User Name TRC-13
Team Number Team 31574
Est. PPD 120.21 (3.76 WUs)
Est. PPW 841.46 (26.30 WUs)
Uploaded Projects 368

I'm thinking about sharing my fahstats online. If I get positive feedback on this I might just do it.

dukla2000 · Post by **dukla2000** » Tue Jan 06, 2004 11:43 pm

Slightly aside, but I am becoming a big fan (

) of qd for measuring boxen performance. Calculates a forecast points/year based on all 10 entries in your queue.dat, and available for Win & Linux. Eliminates the temporary excitement of a good or bad frame of WU but basically seems more meaningful to me.

Zyzzyx · Post by **Zyzzyx** » Wed Jan 07, 2004 12:13 am

TRC-13 wrote:I'm thinking about sharing my fahstats online. If I get positive feedback on this I might just do it.

Go for it, I'm rather curious to see your collection.

I've got my four home systems up and displayed at folding.zyzzyx.com

haysdb · Post by **haysdb** » Wed Jan 07, 2004 12:16 am

OK, I give up, what am I supposed to glean from this?

Code: Select all

C:\Program Files\Folding@Home>qd -i
qd released 5 January 2004
Queue version 4.01
Current index: 2

 Index 3: finished 51.40 pts (3.555 pt/hr) 59.8 X min speed
   server: 171.67.89.149:8080; project: 1026
   Folding: run 2, clone 5, generation 32; benchmark 6168; misc: 500, 300
   issue: Fri Jan 02 10:12:59 2004; begin: Fri Jan 02 10:12:56 2004
   end: Sat Jan 03 00:40:28 2004; due: Sat Feb 07 10:12:56 2004 (36 days)
   core URL: http://www.stanford.edu/~pande/Win32/x86/Core_78.fah
   assignment info: Fri Jan 02 10:12:58 2004; ACD24B60
   user: David_Hays; team: 31574; ID: 6383C60E3D6CD3DD; mach ID: 1
   work/wudata_03.dat file size: 197645; WU type: Folding@Home

 Index 4: finished 51.40 pts (3.595 pt/hr) 60.4 X min speed
   server: 171.67.89.149:8080; project: 1025
   Folding: run 4, clone 68, generation 13; benchmark 6168; misc: 500, 300
   issue: Sat Jan 03 00:41:17 2004; begin: Sat Jan 03 00:41:17 2004
   end: Sat Jan 03 14:59:03 2004; due: Sun Feb 08 00:41:17 2004 (36 days)
   core URL: http://www.stanford.edu/~pande/Win32/x86/Core_78.fah
   assignment info: Sat Jan 03 00:41:17 2004; ACD33EE7
   user: David_Hays; team: 31574; ID: 6383C60E3D6CD3DD; mach ID: 1
   work/wudata_04.dat file size: 197284; WU type: Folding@Home

dukla2000 · Post by **dukla2000** » Wed Jan 07, 2004 11:47 am

haysdb wrote:OK, I give up, what am I supposed to glean from this?

My rocks get off on the bottom line (from my P3 mobile)

Code: Select all

 
Annual score benchmark: 15160

In your case the 2 quoted units ran similar 3.5x points/hour on p1026 and p1025, and as per your first post you know a p803 would give more. But averaged over 10 WU is more meaningful (to me at least - takes out the highs of 803s and lows of Tinkers). That box of yours is presumably around 31k/year?

haysdb · Post by **haysdb** » Wed Jan 07, 2004 6:57 pm

dukla2000, there was a total line at the bottom, which I should have included in my post. Is that what would tell me the PPY?

Let me ask this another way: Which lines in the qd output do you look at?

I haven't tried using qd across the network, but I assume I could get the results for any machine who's folder is accessible on my local area network? I'd like to write a script to tally the results for all cpu's. One of my beefs with LogStats is it looks at only the current WU, so the numbers fluctuate wildly. That's nice to see, but I'd also like the option of seeing a longer term average.

David

dukla2000 · Post by **dukla2000** » Thu Jan 08, 2004 12:28 am

haysdb wrote:dukla2000, there was a total line at the bottom, which I should have included in my post. Is that what would tell me the PPY?

Yup - my interpretation of the docs is the last line is PPY based on the 10 entries in queue.dat.

haysdb wrote:Let me ask this another way: Which lines in the qd output do you look at?

PPY from the bottom line, and pph for each unit. I have worked out my XP2400+ can do 28k PPY. Then again if I am doing stuff that means FAH doesn't get 100% (including reboots, hardware/cooling/silencing upgrades!) the qd report drops to say 27k, but I see it creeping back towards 28k when I stop playing. The PPH is slightly useful to explain how the PPY is moving: if the current unit is slow (below my expected 3.2PPH) then it helps figure.

haysdb wrote:I haven't tried using qd across the network, but I assume I could get the results for any machine who's folder is accessible on my local area network?

Me neither, but as there are Win & Linux versions at least on Linux it should be possible. There is also a plugin for gkrellm but I cant get that (plugin) to compile. To summarise for your farm you may need to pipe the results for each box to a local file then harvest those files?

haysdb · Post by **haysdb** » Thu Jan 08, 2004 12:58 am

So if it says "Annual score benchmark: 31341" that means 31K points per year? That's my P4 2.4/533.

Well I'll be damned. I can run it under WinXP for any other machine on the LAN, WindowsXP or Linux. Here is an example:

> qd -f \\Fahserv\Folding@Home\ws001

31K P4 2.4/533
34K P4 2.6/533
37K P4 2.6/800/HT (combined score of 2 Consoles)
23K 2000+ Palomino (Linux Server)

The scores for my Athlon 2500+ and 2600+ blades are foobar because of some "instability" over the last day or two. One of them claims 73K

I assume I can do the same thing from the Linux box, which would make it easy to "script".

Edit: silly question but how do I refer to other machines on the network in Linux? I have tried //pcnut/Folding@Home and smb://pcnut/Folding@Home.

David