Benchmarks for new Protein 105aa

[IronBits]

Put your benchmarks here. I released the others. IF someone needs them back (I hope not) I'll dig back and find them ;)

[Chinasaur]

Distributed Folding Linux Benchmark Script V1.0

Sample Size: 89120 structures over 32821 seconds.

Structures Per Second: 2.72
Structures Per Minute: 162.92
Structures Per Hour: 9775.20
Structures Per Day: 234604.80

Linux OS - Running Kernel Version 2.4.18
AMD Athlon(tm) Processor @ 1533mhz (256 KB cache)

Libranet Debian (Woody), 512MB, 2xXP1800, S2460, nice -3

[MAD-ness]

Thanks IB.

I grabbed a copy of the ones from a couple times ago (the last time there were quite a few submitted) so it is all good.

I guess I should do something with them. :(

Maybe we will ge ta lot submitted since this protein should last a long time.

More different cpus = better. :)

[Alpha_7]

Sample Size : 53440 structures over 25290 seconds.
Protein Size: 105AA

Structures Per Second: 2.11
Structures Per Minute: 126.8
Structures Per Hour : 7607
Structures Per Day : 182571

OS : Windows 2000 MHz: 1300
CPU: AMD Duron(tm) processor
Client Switches: -df -rt

My Main Box

[pfb]

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.8 Benchmark

Sample Size : 120040 structures over 39254 seconds.
Protein Size: 105AA

Structures Per Second: 3.06
Structures Per Minute: 183.5
Structures Per Hour : 11009
Structures Per Day : 264214

OS : Windows XP MHz: 2709
CPU: Intel(R) Pentium(R) 4 CPU 2.40GHz
Client Switches: -df -rt
------------------------------------------------------------


:thumbs:

[Paratima]

Distributed Folding Windows dfGUI v1.8 Benchmark

Sample Size : 82885 structures over 27006 seconds.
Protein Size: 105AA

Structures Per Second: 3.07
Structures Per Minute: 184.1
Structures Per Hour : 11049
Structures Per Day : 265173

OS : Windows 98 MHz: 1587
CPU: AuthenticAMD AMD Athlon(tm) XP 1900+
Client Switches: -df -qt -rt -s 9000

[FoBoT]

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.8 Benchmark

Sample Size : 19670 structures over 10396 seconds.
Protein Size: 105AA

Structures Per Second: 1.89
Structures Per Minute: 113.5
Structures Per Hour : 6811
Structures Per Day : 163475

OS : Windows XP MHz: 1063
CPU: Intel(R) Pentium(R) III Mobile CPU 1066MHz
Client Switches: -df -qt -i f -rt -s 10000
------------------------------------------------------------

[Scoofy12]

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.7 Benchmark

Sample Size : 124505 structures over 78227 seconds.
Protein Size: 105AA

Structures Per Second: 1.59
Structures Per Minute: 95.5
Structures Per Hour : 5730
Structures Per Day : 137513

OS : Microsoft Windows XP MHz: 945
CPU: AMD Duron(tm) processor
Client Switches: -df -rt
------------------------------------------------------------
Duron 750@945, FSB@270

------------------------------------------------------------
Scoofy's Stolen Distributed Folding Linux Benchmark Script V1.0

Sample Size: 136315 structures over 154061 seconds.

Structures Per Second: 1.77
Structures Per Minute: 106.19
Structures Per Hour: 6371.02
Structures Per Day: 152904.48

Linux OS - Running Kernel Version 2.4.19
Celeron (Mendocino) @ 495mhz (128 KB cache)
------------------------------------------------------------
2x Celeron 366@495 90(?) FSB, -df -rt

[pfb]

leaving mine for a bit longer with nowt else running (first one was after I watched some DVDs and so around 85% CPU usage for DF) I get:



------------------------------------------------------------
Distributed Folding Windows dfGUI v1.8 Benchmark

Sample Size : 252090 structures over 80549 seconds.
Protein Size: 105AA

Structures Per Second: 3.13
Structures Per Minute: 187.8
Structures Per Hour : 11267
Structures Per Day : 270402

OS : Windows XP MHz: 2709
CPU: Intel(R) Pentium(R) 4 CPU 2.40GHz
Client Switches: -df -rt
------------------------------------------------------------

[FoBoT]

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.7 Benchmark

Sample Size : 240720 structures over 93265 seconds.
Protein Size: 105AA

Structures Per Second: 2.58
Structures Per Minute: 154.9
Structures Per Hour : 9292
Structures Per Day : 223001

OS : Microsoft Windows XP MHz: 1392
CPU: AMD Athlon(tm) processor
Client Switches: -df -qt -rt -s 5000
------------------------------------------------------------

[FoBoT]

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.7 Benchmark

Sample Size : 195910 structures over 108767 seconds.
Protein Size: 105AA

Structures Per Second: 1.80
Structures Per Minute: 108.1
Structures Per Hour : 6484
Structures Per Day : 155623

OS : Microsoft Windows XP MHz: 1000
CPU: AMD Athlon(tm) processor
Client Switches: -df -qt -rt -s 2500
------------------------------------------------------------

[Wedge]

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.8 Benchmark

Sample Size : 233840 structures over 75273 seconds.
Protein Size: 105AA

Structures Per Second: 3.11
Structures Per Minute: 186.4
Structures Per Hour : 11184
Structures Per Day : 268410

OS : Windows XP MHz: 1799
CPU: AMD Athlon(TM) XP 2200+
Client Switches: -df -rt -s 10000 -p 20
------------------------------------------------------------

[MAD-ness]

In the interest of collecting data that might give more insite into the peformance and behavior of the client, could anyone do any of the following benchmarks (if you have the ability and the willingness to do the work, for which there is no reward other than knowledge gained):

- an AMD Duron 1.0GHz system dedicated to crunching. It can use SDRAM or DDR, but it should be a relatively "average" 1ghz box with pretty consistent output/usage. The goal here is to get a reliable number for Tellian to use in his cool cross-project stats.

- a Pentium 4 Northwood (though a Willy would be welcome as well) running with the the extra RAM option and also running without the extra RAM option. I am curious if the smaller performance increase on the P4 when enabling the extra RAM option in comparison to the Athlon is still an issue. Also, if anyone has noticed any differences between the Athlon (sort of the standard) and platforms other than the P4 (specifically the P3). Is the Athlon's increase larger than on other CPUs, or is the P4s smaller?

- tests on a single computer (or on seperate boxen with identical hardware) running Windows NT (any flavor, I think they are all real close), Windows9x (ditto in regards to versions) and Linux (I imagine this could vary a lot depending on libraries and distros, etc., anyone have anything useful to point out from thier observations?).

Some questions that someone can probably answer without running tests.

Is the performance anomaly when running the command line client in Windows 9x with the graphics on (as opposed to having the output muted) still present?

I am pretty sure that there still exists a performance difference between Linux and Windows (any version), but can someone confirm this (based upon reasonably reliable data)? If so, does anyone have a guess which they feel is pretty accurate in terms of % difference?

Has anyone ever observed any real performance improvement as a result of increasing the priority level (this is in regards to farm boxen, not computers in use for other activities)?

How about in regards to the frequency of writing to the progress.txt file? Obviously less disk writes would be a good thing, but has anyone been able to document what was even a noticeable difference?

Thanks.

Also, if you have questions regarding this stuff that someone doesn't answer right away, feel free to holler at me and I will see what I can dig up. I don't actually know jack about anything, but I can be persistent and I hassle people who actually DO know what they are talking about. ;)

[Paratima]

Perhaps this will help answer somebody's questions, somewhere.

I'm setting up a system (typing on it now, matter of fact) to test OSes. The hardware is REALLY stable!

The hdd has 4 partitions, one each for Win98, W2K, WXP, and RedHat Linux. Set up with System Commander, none of the partitions can see any of the others. Each one thinks it's all there is. ;) Yes, Bill, they're all legal, fresh copies.

Just finished loading the 3 Winders platforms. Gonna do RedHat tomorrow night, unless work intervenes.

I'm optioning each as close as possible to what I believe to be a "standard" installation.

My plan is to run each one for 24 hours, no-netted, then harvest the results. The way my schedule looks now, I should be able to publish first results in about a week. I hope. THEN I should be able to go back and tweak, as for instance, the graphics in Win98.

Look for a new thread on this soon.

[FoBoT]

MAD-ness
i can do some of this, when i get time, probably in a couple of weeks

i have access to many identical boxen that i can load various OS onto and run benchmarks

i know for a FACT that WinXP is slower than Win9X, i have already submitted benchmarks from identical boxen proving that

[pizzaking]

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.8 Benchmark

Sample Size : 18165 structures over 9004 seconds.
Protein Size: 105AA

Structures Per Second: 2.02
Structures Per Minute: 121.0
Structures Per Hour : 7263
Structures Per Day : 174307

OS : Windows 2000 MHz: 1594
CPU: Intel(R) Pentium(R) 4 Mobile CPU 1.60GHz
Client Switches: -df -i f -rt -s 10000 -p 20
------------------------------------------------------------

Shame I only have it for a day

[MAD-ness]

Seems like you already have enough work to do Fobot. :(

Not that I will argue.

Also, if you have results that you have submitted on a recent protein which you feel are pretty accurate (machine not in use, several hour minimum run time, etc.) and you point them out to me I will use them as the data.

It isn't so much that no one has submitted the benchmarks I am looking for as it is that I don't KNOW that is what they are and I don't know how trustworthy the data is.

I figure that even with accurate data I will screw things up, so if i start with unreliable data I will REALLY be in trouble.

Paratima: sounds interesting, methodical and ambitious. I like it. :)

Another thing that I just thought of. Using the gcc compiled version of a client instead of the icc compiled version (it is late again, maybe I am thinking of another project's client offering both versions). Again, there is an assumption that one is faster than the other (with the icc version being the faster of the two) but having a solid number to throw to give people when they ask would be a "good thing."

[FoBoT]

i can give you good data, when i get time

these boxen are not used by anybody else , no other processes going on
they are stock out of the box Compaq pc's, all identical
i have full control of them and can load any OS, run the client anyway i need to (well, execpt with -it)

the only down side is, they are ALL intel, no AMD
and my largest populations are

Pentium III 1Ghz
Celeron 1.1 Ghz
Pentium III 450 Mhz

in january , i could do similar runs with celeron 1.7 Ghz, and maybe another model, but again, no AMD

[MAD-ness]

Fobot: DC is probably the only area more heavily populated by AMD K7 core CPUs than gaming.

There always seem to be Duron, Tbird, and XP results out the wazoo, but not as many on Intel systems.

To me, the massive number of platform variations, even in one "family" of Intel processors (i.e., Pentium, PPro, P2, Celeron, P3, P4, etc.) is both a hassle (trying to figure out what stepping, FSB speed, chipset, cache size/speed, packaging type, pin count, etc.) and something of interest.

Because everything after the Pentium and before the P4 were derivatives of the PPro core (correct me if I am wrong here, but I don't think that I am, the P2 was a reworking of the PPro and the rest of the lines evolved from the P2 core), data taken from across this range naturally conveys information about architecture improvements, client resource demands, etc.

Want to know if more cache has a large impact on a program? Run it on Celeron, a P2, a coppermine P3 and a Xeon P3 with a large cache. Want to know how FSB speed impacts a program? You can go from 66 to 133 and you can cover several configurations with a single motherboard (given the proper chipset, the 440bx being a famous example of a versatile chipset).

Anyways, AMD benchmarks are easy to obtain, in fact, I have no doubt I could obtain just about any info I need on an AMD, if not I could run it myself on one of the 4 various AMD systems I have direct access to. :)

[Darkness Productions]

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.8 Benchmark

Sample Size : 1069300 structures over 339638 seconds.
Protein Size: 105AA

Structures Per Second: 3.15
Structures Per Minute: 188.9
Structures Per Hour : 11334
Structures Per Day : 272019

OS : Windows 2000 MHz: 1625
CPU: AMD Athlon(tm) XP 1900+
------------------------------------------------------------

Commandline:

.\foldtrajlite -f protein -n native -qt -df -rt -g 100 -s 10000 -p 18

[Darkness Productions]

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.8 Benchmark

Sample Size : 173050 structures over 54734 seconds.
Protein Size: 105AA

Structures Per Second: 3.16
Structures Per Minute: 189.7
Structures Per Hour : 11382
Structures Per Day : 273167

OS : Windows 2000 MHz: 1637
CPU: AMD Athlon(tm) XP 2000+
------------------------------------------------------------


Same machine as before, with a 1MHz FSB bump.
I plan to continue on like this until the box stops responding.

Also note, the memory timings as I have them are about as good as they will get.


Commandline:

.\foldtrajlite -f protein -n native -qt -df -rt -g 10 -s 10000 -p 18

[dtsang]

------------------------------------------------------------
Distributed Folding OS X Benchmark Script V1.0

Sample Size: 3600 structures over 4708 seconds.

Structures Per Second: 0.76
Structures Per Minute: 45.88
Structures Per Hour: 2752.76
Structures Per Day: 66066.24

Mac OS X 10.1.5 build 5S66 - Running Kernel Version 5.5
1 ppc7400 (PowerPC 7400) @ 466mHz (1024Kb L2 cache) (0Kb L3 cache)
------------------------------------------------------------
-qt on, progress.txt written every 100 proteins, -rt on

[guru]

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.7 Benchmark

Sample Size : 810 structures over 266 seconds.
Protein Size: 105AA

Structures Per Second: 3.11
Structures Per Minute: 182.7
Structures Per Hour : 10962
Structures Per Day : 263098

OS : Microsoft Windows XP MHz: 1575
CPU: AMD Athlon(tm) processor
Client Switches: -df -qt -rt -g 30
------------------------------------------------------------

[Darkness Productions]

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.8 Benchmark

Sample Size : 1656890 structures over 616519 seconds.
Protein Size: 105AA

Structures Per Second: 2.69
Structures Per Minute: 161.2
Structures Per Hour : 9675
Structures Per Day : 232199

OS : Windows ME MHz: ???
CPU: AuthenticAMD AMD Athlon(tm) Processor
Client Switches: -df -qt -i f -rt -g 10 -s 10000
------------------------------------------------------------


It's a 1.4TBird on WinME (as you can see). It's actually a damn good cruncher for what I need it for.

[pizzaking]

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.8 Benchmark

Sample Size : 1635150 structures over 790447 seconds.
Protein Size: 105AA

Structures Per Second: 2.07
Structures Per Minute: 124.1
Structures Per Hour : 7447
Structures Per Day : 178731

OS : Windows 2000 MHz: 1202
CPU: AMD Duron(tm) processor
------------------------------------------------------------

using extra RAM switch

[pizzaking]

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.8 Benchmark

Sample Size : 1121100 structures over 534745 seconds.
Protein Size: 105AA

Structures Per Second: 2.10
Structures Per Minute: 125.8
Structures Per Hour : 7548
Structures Per Day : 181143

OS : Windows 2000 MHz: 1194
CPU: AMD Duron(tm) Processor
------------------------------------------------------------

using Extra RAM switch

[pizzaking]

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.8 Benchmark

Sample Size : 521800 structures over 260197 seconds.
Protein Size: 105AA

Structures Per Second: 2.01
Structures Per Minute: 120.3
Structures Per Hour : 7219
Structures Per Day : 173267

OS : Windows XP MHz: 1210
CPU: AMD Duron(tm) Processor
------------------------------------------------------------

using extra RAM


------------------------------------------------------------
Distributed Folding Windows dfGUI v1.7 Benchmark

Sample Size : 8785 structures over 10462 seconds.
Protein Size: 105AA

Structures Per Second: 0.84
Structures Per Minute: 50.4
Structures Per Hour : 3024
Structures Per Day : 72585

OS : Microsoft Windows 2000 MHz: 500
CPU: GenuineIntel x86 Family 6 Model 8 Stepping 3
Client Switches: -df -rt -i f -s 10000-p 10
------------------------------------------------------------

[bwkaz]

------------------------------------------------------------
Distributed Folding Linux Benchmark Script V1.0

Sample Size: 3192980 structures over 1219500 seconds.

Structures Per Second: 2.62
Structures Per Minute: 157.10
Structures Per Hour: 9425.77
Structures Per Day: 226218.48

Linux OS - Running Kernel Version 2.4.19
AMD Athlon(tm) XP 1800+ @ 1532mhz (256 KB cache)
------------------------------------------------------------Yeah, that's decent, I'd say ;) Using the extra ram switch as well. On my firewall:


------------------------------------------------------------
Distributed Folding Linux Benchmark Script V1.0

Sample Size: 3107795 structures over 2174948 seconds.

Structures Per Second: 1.43
Structures Per Minute: 85.73
Structures Per Hour: 5144.06
Structures Per Day: 123457.44

Linux OS - Running Kernel Version 2.4.19-pre8
Pentium III (Coppermine) @ 797mhz (256 KB cache)
------------------------------------------------------------Not horrible for that processor. That one's using -rt also. The reason it's doing about half of what my main box is, is that on my main box I do other fun things that take CPU time, like recompiling winex and such. zsnes also takes quite a bit of processor to run, for some reason. And the ut2k3 demo isn't probably good for my production, either. ;)

Edit: Both are the icc-compiled version. Both are LFS, pretty standard versions of everything. On the P3, it's all compiled with -O3 -march=i686, and on the Athlon, it's all compiled with the -O3 -march=athlon-xp -mmmx -msse -m3dnow -mfpmath=sse flags.

[Paratima]

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.8 Benchmark

Sample Size : 691585 structures over 215358 seconds.
Protein Size: 105AA

Structures Per Second: 3.21
Structures Per Minute: 192.7
Structures Per Hour : 11561
Structures Per Day : 277459

OS : Windows 98 MHz: 1,592
CPU: AuthenticAMD AMD Athlon(tm) XP 2200+
Client Switches: -df -qt -rt -s 5500
------------------------------------------------------------
FINALLY got that box running mostly right! This is the MSI-KT3 Ultra with the XP2200+ in it.

Ran like a rabbit when I first slapped it together. Then, as the proteins got faster, I started getting "Signal 11" errors in Linux. Howard said "memory". Swapped out the DDR, that wasn't it.

Switched to Win98. Problems "sorta" went away. A few BSD's, and some lockups, but better. Protein got faster (the current one). Arrrrrrgh! Almost wouldn't run, even with the BIOS set to "safe" defaults.

The BIOS! Ay-yi-yi! I FINALLY took the time to play around with it. Downloaded the lastest from MSI. The upgrade, including writing out a floppy, rebooting, etc, etc, took maybe 15 minutes.

VOILA! With slight tweaking, this board now runs like a striped-assed ape. These numbers are the best I can get without playing with voltages. This is with the CAS left at the default 2.5, so's I can run higher FSB.

Check for upgrades for your BIOS! :p :p :p (This is my new theme song!)

[HaloJones]

------------------------------------------------------------
Distributed Folding Linux Benchmark Script V1.0

Sample Size: 2925 structures over 2425 seconds.

Structures Per Second: 1.21
Structures Per Minute: 72.37
Structures Per Hour: 4342.27
Structures Per Day: 104214.48

Linux OS - Running Kernel Version 5.8
@ mhz ( cache)
------------------------------------------------------------

This is a Sun 6800 so actually a 900MHz Sparc 3 and Solaris 8. Pretty crappy really!

[cygnussphere]

:help:

Distributed Folding Windows dfGUI v1.8 Benchmark

Where can I get this benchmarking proggie???

[Alpha_7]

http://gilchrist.ca/jeff/dfGUI/

Go For It
:D
:cheers:
Alpha_7

[cygnussphere]

:cool:

Thanks!

:cheers:

[Paratima]

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.8 Benchmark

Sample Size : 77190 structures over 21128 seconds.
Protein Size: 105AA

Structures Per Second: 3.65
Structures Per Minute: 219.2
Structures Per Hour : 13152
Structures Per Day : 315658

OS : Windows 98 MHz: 1870
CPU: AuthenticAMD AMD Athlon(tm) XP 2200+
Client Switches: -df -qt -rt -s 6000
------------------------------------------------------------
My final post for this protein.

Yes, I did get around to tweaking the 2200+ on the MSI KT3 board. :D

She's running a tad warm now, so I probably won't let her stay long at this level. But ain't she sweet?

[pizzaking]

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.8 Benchmark

Sample Size : 187985 structures over 74456 seconds.
Protein Size: 105AA

Structures Per Second: 2.52
Structures Per Minute: 151.5
Structures Per Hour : 9089
Structures Per Day : 218141

OS : Windows 2000 MHz: 1999
CPU: Intel(R) Pentium(R) 4 CPU 2.00GHz
Client Switches: -df -i f -rt -s 10000 -p 20
------------------------------------------------------------