CASP5 Protein #2 Performance Benchmarks

[MAD-ness]

In a thread on the Free-DC forum I am collecting benchmark submissions for the DF client running on various machines/CPUs/etc. Most are using DFGUI's built in benchmark scripts but you can also use Jeff's (Digital Parasite) Windows Perl script or Dyyryath's windows or linux scripts to get benchmark information.

http://bane.free-dc.org/forum/showthread.php3?s=&threadid=1020

I also have a link to a temporary page with the currently submitted benchmarks in a chart, but there are only 9 so far. I will hopefully get the benchmark page hosted somewhere non-sucky tommorow, but for now you can take the link in that thread to see the small sample set we have so far.

Thanks.

[Eaglechild]

Posted in other thread. Nice job.:thumbs:

[MAD-ness]

I added in the additional submitted benchmarks and have a new version of the page up. I will just gank what I typed over in the benchmark thread on the Free-DC DF forum:


ok, added in new benchmarks.

Forgot to add in jdhaeden's laptop and I think someone at Ars posted a K6-3 bench and I have had my celeron266 running for 24 hours, but oh well, next protein I will grab all them, the info is here for anyone who is really interested anyways.

Fastest bench submitted here was MashRinx's P4 2.2ghz Northwood.

Collin's two heavily overclocked AMD CPUs had great results though. I don't know if I got the stock fsb/multiplier info right on those cpus though Collin. I guessed on the XP1500+ frequency since it is one of the XP models I do not know in terms of frequency. I assumed that you left the multipliers locked and went for straight FSB overclocking.

The processors that have "(1)" or "(2)" in the Processor name were the first and second cpus of duallies. This isn't the most efficient way of doing this, but I thought the info on how the second cpu performs (reallly close to the first CPU and not much slower than a single CPU set-up) would be useful.

http://www.geocities.com/m4dn3ss2000/CASP5Protein2.html

I have sent off the file to pointwood and we will be getting it up on www.teamstirfry.com but I don't know jack about HTML or any other web nonsense (or computers in general) so I don't know exactly how long it will take for us to play email tag over about 10 timezones and get it set up.

Thanks for the submissions, I will try to get it up faster next week, I had a ton of real life issues this week not to mention no internet access for a while.

Hopefully I can get something set up on the TSF site that is easy to navigate and which I can update more quickly and painlessly.

As always, if someone wants to help out by setting this up and running a database, or by sending me a better looking table, or offering suggestions or tips or whatever, I appreciate it. I am just trying to put together another resource for our community to use that will hopefully encourage people to participate in DF and to stick with it.

[MAD-ness]

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.0 Benchmark

Sample Size: 1500 structures over 17261 seconds.

Structures Per Second: 0.09
Structures Per Minute: 5.3
Structures Per Hour : 316
Structures Per Day : 7580

OS : Microsoft Windows XP MHz: 350
CPU: AMD-K6(tm) 3D processor
------------------------------------------------------------

[Paratima]

Hey, thanks, MAD-ness! You're doing a good service here. :thumbs:

I've already made a couple of changes based on the numbers I've seen and just from the exercise of running the numbers on my own boxen. Very helpful.

Also toots on the horn for Dyyryath for the Perl script and Digital Parasite for dfGUI, so we can get these measurements to begin with!

[collin]

Hi MAD-ness,

The stock speed you specified for the XP1500 is correct (1333, 10 multiplier).

I've unlocked that chip, but my best guess is that the df client would love the extra fsb for the memory access.

Here's another gem of mine (this chip's still locked though)

Athlon XP 1600+ (10.5x169) Base Speed (1400MHz)
WinXP (~1775MHz)
256Mb DDR266
Running as a service
-----
Structures Per Second: 0.81
Structures Per Minute: 48.3
Structures Per Hour : 2900
Structures Per Day : 69591

Sample size : 228050 structures

[Burgergold]

tbird 900@945 FSB 200
MSI kt266
3x128mb DDR PC2100
Windows 2000

0.44 / second

[Eaglechild]

I noticed there were some info gaps on the Macs, so thought I would fill them in.

The G4(1) and G4(2) should have 133mhz frontside bus. They also have either 1 or 2 mb of L3 cache per processor This compares with the Power Mac 800 (also a G4) which does not have a L3 cache, and so runs noticeably slower.

The iBook 600 has a 100mhz FSB. The operating system on all 4 is the same OS-X v10.1.5. The iBooks times are a little slower than they should be, as it is my "primary" computer, and received a lot of other use at the same time it was benchmarked crunching. Otherwise it would have been well into the 600's

Thanks for the great work!:cheers:

[baja27]

Pentium 4 2.26 @2652mhz 17x156 asus p4t533cl rambus@936

Structures Per Second: 1.25
Structures Per Minute: 75.0
Structures Per Hour : 4.500
Structures Per Day : 108.000

this is the best show in 1 min i saw

Win XP Pro

[jamesa]

Originally posted by Eaglechild
I noticed there were some info gaps on the Macs, so thought I would fill them in.

The G4(1) and G4(2) should have 133mhz frontside bus. They also have either 1 or 2 mb of L3 cache per processor This compares with the Power Mac 800 (also a G4) which does not have a L3 cache, and so runs noticeably slower.

The iBook 600 has a 100mhz FSB. The operating system on all 4 is the same OS-X v10.1.5. The iBooks times are a little slower than they should be, as it is my "primary" computer, and received a lot of other use at the same time it was benchmarked crunching. Otherwise it would have been well into the 600's

Thanks for the great work!:cheers:

EagleChild, how do you run the benchmarks under OS X? I had a brief look through the forums but couldn't see it. :crazy:

Thanks

-- james

[Welnic]

Originally posted by jamesa


EagleChild, how do you run the benchmarks under OS X? I had a brief look through the forums but couldn't see it. :crazy:

Thanks

-- james

James

Edit the foldit and add a -df and -i f to the startup line.
Remove the foltrajlite.lock file so the process will stop.
Once the program stops start it with ./foldit & in the proper folder.
Run the program for as long as you feel like it.
The foldtrajlite.lock file has the starting time.
When you are ready to stop then delete the foldtrajlite.lock file.
The filelist.txt file has the end time.
Read the filelist.txt file for the number of folds done.
Remove the -i f and restart.

Clint