[mikkyo]

I was poking around the various fora today and looking at the team pages but I cant find what I am looking for..

I am wondering which macs (and other platforms) are the best at the various DCs. I want to know how to distribute my machines across several projects and benchmarking info would be very helpful.
So which machine is the best at each of the DC projects?
RC5 - G4 Dual GHz is about the only one I can guess at but how many keys/sec? And how does that compare to the next best thing?
Ubero? Athlon 2100 XP? Running what OS? RAM a factor?

It would be very nice to have a table of various machines and how thye do on each of the projects.
Maybe someone could put up a page where data like this could be entered ala xlr8yourmac's CPU benchmarking pages?

In the mean time how about people just post their benchmarks for individual machines working on only one project?
Something like:

DC Project
DC version (CLUI,GUI)
CPU
OS
RAM
Best Day's Work

[mikkyo]

DC Project - Ubero
DC version - CLUI 2 agents
CPU - G4 Dual GHz
OS - Mac OS X 10.1.3
RAM - 256
Best Day's Work 600 WUs (300/agent)

[Scotttheking]

http://jaffafamily.org/~distcomp/bestmachine.html

Sorry about the weird character, it somehow got in there, and I can't get it out.

[DAlex]

For dfold the only benchmark I can provide is:

Athlon 1800xp, 512 Ram, running XP Pro = 7328 units / hour using the CLI version of the client (average taken over 24 hours)

I'll benchmark my G4 733, iBook and P3 933 later today (they all run dfold)

Update: P3 933 = 2790/hour
G4 733 = 3900/hour
note that both these are work machines and these are daytime (ie in use) figures.

DAlex

[ 04-15-2002: Message edited by: DAlex ]

[slipjack]

ont size="1" face="Geneva, Verdana, Arial">quote:</font><HR>Originally posted by Scotttheking:
<STRONG>http://jaffafamily.org/~distcomp/bestmachine.html

Sorry about the weird character, it somehow got in there, and I can't get it out.</STRONG>[/quote]

Try textSOAP... it's pretty good at getting rid of that kind of thing... and there is a free trial.
http://www.unmarked.com/downloads.html

[ 04-15-2002: Message edited by: slipjack ]

[ 04-15-2002: Message edited by: slipjack ]

[Shaktai]

Following are SETI benchmarks

DC version (CLUI,GUI): CLUI
CPU: 800 mhz G4
OS: OS-X
RAM: 512
Best Day's Work: 2.5 units

DC version (CLUI,GUI) CLUI
CPU: 600mhz G3
OS: OS-X
RAM: 256
Best Day's Work: 1.6

DC version (CLUI,GUI) CLUI
CPU: Celeron 1 ghz
OS: Windows XP
RAM: 256
Best Day's Work : 2.08

DC version: CLUI
CPU: Athlon XP 1600+ (1.4 ghz)
OS: Windows XP
RAM: 256
Best Day's Work: 3.7 units

Following are for Dist. Folding
All are the CLUI client running in "quiet" mode which is the fastest, especially for Macs. All have 256 mb RAM except for the G4 with 512 mb

Power Mac 800
4110 per hour

iBook 600
2910 per hour

Celeron 1 ghz
3512 per hour

Athlon 1600+ (1.4 ghz)
6566 per hour

Your dual G4 1 gig will eat everything else alive at RC-5. Somewhere around 20 millon keys per second.

[ 04-15-2002: Message edited by: Shaktai ]

[Shaktai]

Here are updated benchmarks for the new Distributed Folding Client which folds a much larger protein and takes a lot longer. This is after about 22 hours. The old clients benchmarks are in Parenthesis for easy comparison. There is a huge speed difference, but then a lot more science is being done also.

800 mhz G4 Power Mac ------------ 1310 (4110)
600 mhz G3 iBook------------------- 973 (2910)
Celeron 1 ghz ---------------------- 1384 (3512)
Athlon XP 1600+ 1.4 ghz------------ 2517 (6566)

[ 04-17-2002: Message edited by: Shaktai ]

[SkiBikeSki]

DFold - G4 867 - 1587units/hour

[Shaktai]

Originally posted by SkiBikeSki:
<STRONG>DFold - G4 867 - 1587units/hour</STRONG>

Yeah, you have a few more megahertz, and a lot bigger cache. Only thing I don't like about my Power Mac 800 is its lack of an L3 cache. It does make a difference.

[Shaktai]

SETI GUI benchmark

iBook 600
OS-X 10.1.3
256 mb RAM
15.5 hours as screensaver with screen blanked after 1 minute.

[wheeles]

dFold - G4 DP 450 - 2000 units/hour (flat out on current protein)

edit: &lt;--- 50 posts. Yay!

[ 04-18-2002: Message edited by: wheeles ]

[Shaktai]

Originally posted by wheeles:
<STRONG>dFold - G4 DP 450 - 2000 units/hour (flat out on current protein)

edit: &lt;--- 50 posts. Yay!

[ 04-18-2002: Message edited by: wheeles ]</STRONG>

That is pretty impressive from a dual 450.

Congrats on the 50 posts. Don't do like me though. Get a life while you're still young.

[Arkham_c]

50 posts a day? Sheesh, don't you people have a life? I always wondered how people got to have 3000+ posts, I guess I know now. At the rate I am going (1 or 2 per day), it'll be a LONG time before I hit 1000.

There's one guy with over 10,000 posts. Hhas multicolored stars and is labeled as "Clinically Insane". I think that is an accurate description. I'd get fired if I were on here enough to post 50 times a day.

[Grozni Majmun]

Originally posted by Arkham_c:
<STRONG>50 posts a day? Sheesh, don't you people have a life? I always wondered how people got to have 3000+ posts, I guess I know now. At the rate I am going (1 or 2 per day), it'll be a LONG time before I hit 1000.
</STRONG>

Arkham_c, are you wearing your glasses?
wheeles wrote "50 posts, Yay" not "50 posts a day". He clocked up those 50 posts since April 2001, so that's 50 in over a year. Wheeles, congrats on losing your "new member" status!

[wheeles]

Originally posted by Grozni Majmun:
<STRONG>

Arkham_c, are you wearing your glasses?
wheeles wrote "50 posts, Yay" not "50 posts a day". He clocked up those 50 posts since April 2001, so that's 50 in over a year. Wheeles, congrats on losing your "new member" status!</STRONG>

I will admit that most of those 50 posts were actually in the last 3 weeks or so, so I am almost as sad as Shaktai on the post rate. However, I shall be starting a new contract soon, so my post rate will drop.

The G4 DP 450 speed that was achieved was measured over a 5 hour period. I did the following:
(i) use &gt;console mode with 2 dFold processes backgrounded and very little else running.
(ii) I lowered the frequency of writing to the progress.txt files.
(iii) To record the status I set up a cron job to write the contents of the progress.txt files to a separate log file once an hour and used this to work out the work rate. I didn't have a while loop keep churning out status every few seconds.
(iv) I uploaded outstanding work just prior to the run thus keeping memory leaking in check.

I haven't played with the niceness as if there isn't much else running the dFold processes' cpu percentages are in the upper 90s. I might be able to squeeze out a little bit more by reducing the writing to progress.txt files but I don't think I would gain much.

FYI I just bought a secondhand G4 cube for a little iTunes jukebox/DVD player I have in mind. I do like the idea of a funky piece of cubic design that makes minimal noise (no fan!) and runs iTunes. I can hook it up to the HiFi, remove the keyboard, mouse and monitor and using airport with VNC can operate it from my G4 DP450. It'll also be crunching.

[SkiBikeSki]

I am compiling all the benchmarks posted here into an excel sheet and exporting it to html and putting on my iDisk. So far I have done the dFold stats.

[SkiBikeSki]

Up went the SETI@Home stats.

[SkiBikeSki]

Here are the Ubero stats. I took the liberty of pulling some of the stats from the team page on Ubero.com by looking for agents with CPU+MHz in the name.

[Shaktai]

Originally posted by SkiBikeSki:
<STRONG>I am compiling all the benchmarks posted here into an excel sheet and exporting it to html and putting on my iDisk. So far I have done the dFold stats.</STRONG>

I think that 733 must have been the old protein, and all the other figures the new protein. An important distinction.

FYI: My Athlon XP 1600+ (1.4 ghz) and my Celeron were both running Win XP.

[SkiBikeSki]

Now there are RC5 stats. Everything I put into it is a subset of the main speeds page on DNet's website. I plan on replacing those benchmarks with our own as soon as people can give me some.

[ 04-19-2002: Message edited by: SkiBikeSki ]

[SkiBikeSki]

Ya I should get rid of that 733, but it will wait until I also add something to the chart, like stats for a 733 with the new protein.

I am naming Win NT, or 2000, or XP as "Win NT" on the table since it's the kernel that is most inportant in the performance of the OS. I will however make the distinction between OS 9 and OS 8-7 since I care more about Mac's than PC's.

I'm still looking for some good Folding@Home stats.

[Shaktai]

Originally posted by SkiBikeSki:
<STRONG>Ya I should get rid of that 733, but it will wait until I also add something to the chart, like stats for a 733 with the new protein.

I am naming Win NT, or 2000, or XP as "Win NT" on the table since it's the kernel that is most inportant in the performance of the OS. I will however make the distinction between OS 9 and OS 8-7 since I care more about Mac's than PC's.

I'm still looking for some good Folding@Home stats.</STRONG>

Awsome job with the other projects, but folding at home is a different bird. For the benchmarks to have any validity, you must also define them by the protein and project, because the numbers vary so dramatically from one to another.

So anyway, here are the only approx stats I can remember, and not exactly certain of the tenths. Too many other projects that I can't directly compare. Times should be considered approximate.

Projects 312 &313, GNRA hairpin 1/2 point
Power Mac 800mhz G4, 6.5 hours
iBook 600mhz G3, 7.7 hours

Projects 111, 112, 113 and 114 -- Staphylococcal protein A, 5 point.
Celeron 1 ghz, 35 hours
Athlon 1600+ 1.4 ghz, 22 hours

[SkiBikeSki]

Originally posted by Shaktai:
<STRONG>

Awsome job with the other projects, but folding at home is a different bird. For the benchmarks to have any validity, you must also define them by the protein and project, because the numbers vary so dramatically from one to another.</STRONG>

Good Point. I had entirely forgot about that. Thanks for your approx. becnhmarks though. Give me some time to come up with a decent average method or table-of-proteins or something. For now I'm thinking of not keeping track of average time, but only the best and worst proteins, and their time.

[wheeles]

posted 04-19-2002 02:33 AM

Hey, guys, unless there's some time difference going on here it looks like you're burning the midnight oil a little. Get some sleep or you'll end up like Scott.

edit: It's 8:30am here before anyone comments on the time of my post.

[ 04-19-2002: Message edited by: wheeles ]

[reader50]

On Folding@home, the "score" values have already been benchmarked (on a P3) by the project, to regularize the value of work units. A CPU should produce a consistent score per week, regardless of protein. It may not be quite as consistent on a PPC, but score is the only thing I can think of that's accurate, barring a massive protein/CPU list.

[Shaktai]

I work a swing shift now. I always burn the midnight oil. It is 12:32 Pacific daylight time.

[SkiBikeSki]

I really was buring the midnight oil. like it says by my name, I'm in Ohio, which is in the eastern time zone. So it really was 2:33 AM when I did all that. I do most of my posting either after class, like now, or late at night on a weeknight when there's nothing else to do, like last night.

I likely will end up like scott. Not because I'm addicted to MacNN, even though I am, but because I'm in college and have very funyy sleep habits. I also, because of college, have very funny eating habits, I'm about to go eat a dinner of 6 bowls of Cap'n Crunch!

[wheeles]

Originally posted by SkiBikeSki:
<STRONG>I likely will end up like scott. Not because I'm addicted to MacNN, even though I am, but because I'm in college and have very funyy sleep habits. I also, because of college, have very funny eating habits, I'm about to go eat a dinner of 6 bowls of Cap'n Crunch!</STRONG>

It's about 9 years since I left college and my eating habits haven't changed much since then. Only yesterday my lunch consisted of breakfast cereal. As for the sleeping habits I tend to go nocturnal between work contracts. If a day had 26 hours and not 24 then I'm sure I'd always be in bed on time.

As for benchmarking, once I've set my cube up I'll post some stats on that.

[jbcool]

Originally posted by SkiBikeSki:
<STRONG>I really was buring the midnight oil. like it says by my name, I'm in Ohio, which is in the eastern time zone. So it really was 2:33 AM when I did all that. I do most of my posting either after class, like now, or late at night on a weeknight when there's nothing else to do, like last night.

I likely will end up like scott. Not because I'm addicted to MacNN, even though I am, but because I'm in college and have very funyy sleep habits. I also, because of college, have very funny eating habits, I'm about to go eat a dinner of 6 bowls of Cap'n Crunch!</STRONG>

Those sound like normal eating and sleeping habits.
What is this sleeping thing people do anyway.

Back to the thread subject, sorry:

G4 DP 800MHZ 1.5GB-133 CL3, Dfold 2700/hr
G4 DP 450MHZ 1.0GB-133 CL3, Dfold 1800/hr
G4 400MHZ 1.5GB-100 CL3, Dfold 880/hr
G4 400MHZ .5GB-100 CL3, Dfold 860/hr
AMD 1900+ .5GB-DDR , Dfold 2700/hr
AMD 1.2 .5GB-SDRAM , Dfold 1700/hr

[SkiBikeSki]

WOW! I wish I had access to all those high power CPU's, all with GB's of RAM.

I'll edit the spreadsheet as soon as I can.

[ 04-21-2002: Message edited by: SkiBikeSki ]

[SkiBikeSki]

I made the changes. jbcool, I suggest we both double check our Athlon XP benchmarks. It seems like my 1466 outperformed your 1600 by too much. It might be the OS, I've got Windows XP, what do you have?

&lt;----- HOLY COW! I've made another 25 posts in no time. I guess I'm very addicted to MacNN now that we have more projects, and the next competition is almost upon us.

[ 04-21-2002: Message edited by: SkiBikeSki ]

[SkiBikeSki]

Anyone else find it odd that mikkyo started this topic, but now I'm the one who's the most active in it. Besides the spreadsheets, I also made the most posts in this topic.

[jbcool]

Originally posted by SkiBikeSki:
<STRONG>Anyone else find it odd that mikkyo started this topic, but now I'm the one who's the most active in it. Besides the spreadsheets, I also made the most posts in this topic.</STRONG>

Same OS type that you are using.
Checked it and it had done 60000 in 18.5 hours. Which comes up to 3243 per hour.
None of my machines are overclocked.

[ 04-23-2002: Message edited by: jbcool ]

[Shaktai]

Originally posted by jbcool:
<STRONG>

Same OS type that you are using.
Checked it and it had done 60000 in 18.5 hours. Which comes up to 3243 per hour.
None of my machines are overclocked.

[ 04-23-2002: Message edited by: jbcool ]</STRONG>

That is pretty impressive. Over a 30 hour period, my 1600+ (1.4 ghz) only averaged 2385 per hour, also with XP. What kind of motherboard and RAM do you have? Are you running the client in quiet mode or normal or as a service? Maybe I should have upgraded my Athlon after all.

[jbcool]

Originally posted by Shaktai:
<STRONG>

That is pretty impressive. Over a 30 hour period, my 1600+ (1.4 ghz) only averaged 2385 per hour, also with XP. What kind of motherboard and RAM do you have? Are you running the client in quiet mode or normal or as a service? Maybe I should have upgraded my Athlon after all.</STRONG>

All motherboards that are running the AMD chips are the ECS - K7S5A
this machine is running 512 MB of DDR ram. This is the fastest box of the bunch. The other machines are using SDRAM chips, go to cut the cost somewhere.

Sorry, forgot to add that I am running quiet mode. Anybody do a benchmark to see if service mode is faster than quiet mode?
Got to do something to catch slipjack.

[ 04-23-2002: Message edited by: jbcool ]

[gumby5647]

Originally posted by Arkham_c:
<STRONG>

There's one guy with over 10,000 posts. Hhas multicolored stars and is labeled as "Clinically Insane". I think that is an accurate description. I'd get fired if I were on here enough to post 50 times a day.</STRONG>

Yeah, we started a thread about what his new title would be called.

Someone suggested "post whore"

but some how he got "clinically insane"

[SkiBikeSki]

Originally posted by jbcool:
<STRONG>

Same OS type that you are using.
Checked it and it had done 60000 in 18.5 hours. Which comes up to 3243 per hour.
None of my machines are overclocked.

[ 04-23-2002: Message edited by: jbcool ]</STRONG>

3243 Sounds much better. I'll edit the table ASAP. BTW I've got an Epox 8K7A mobo with 512DDR.

[SkiBikeSki]

Originally posted by Shaktai:
<STRONG>

That is pretty impressive. Over a 30 hour period, my 1600+ (1.4 ghz) only averaged 2385 per hour, also with XP. </STRONG>

So your Athlon XP 1.4GHZ averaged 2517units/day one time and 2385units/day another time. Guess I'll put 2 data points in the table.

[ 04-23-2002: Message edited by: SkiBikeSki ]

[SkiBikeSki]

Originally posted by Shaktai:
<STRONG>

Are you running the client in quiet mode or normal or as a service?</STRONG>

How does one go about running it as a service? Is that where there is no window for it at all, and it runs entirely in the background? Sorta like "&lt; /dev/null"?

[ 04-23-2002: Message edited by: SkiBikeSki ]

[Shaktai]

Originally posted by SkiBikeSki:
<STRONG>

So your Athlon XP 1.4GHZ averaged 2517units/day one time and 2385units/day another time. Guess I'll put 2 data points in the table.

[ 04-23-2002: Message edited by: SkiBikeSki ]</STRONG>

Yeah, weird isn't it. I can only guess that I got a lot of random proteins that took longer. I do see a greater variance with the newest proteins. Every now and then you get one that just sits there going back and forth trying to figure a possible combination. I have seen a couple that get stuck at around 60 or 70 and may take several seconds to complete the structure, where the average is 1 second or less. I figure the 30 hour benchmark is the most realistic.

Any way it goes, I need more power.

[SkiBikeSki]

With the screensaver I see tons of proteins that hover at about 60, and it takes the software a while to decide what to do. But with the CLI I see that happen less. Maybe that's just a coincidence.

Benchmarks should definitely be taken over a 24 hours or more, period to reduce the effects of bazzar proteins.

[Shaktai]

Originally posted by SkiBikeSki:
<STRONG>With the screensaver I see tons of proteins that hover at about 60, and it takes the software a while to decide what to do. But with the CLI I see that happen less. Maybe that's just a coincidence.

Benchmarks should definitely be taken over a 24 hours or more, period to reduce the effects of bazzar proteins.</STRONG>

I think I found the problem. I had a Window stats program running in the background that not only monitored the Athlon's client, but also the Celeron over the network. I have turned it off and will see if that improves production any over the next 24 to 48 hours. I suspect it was stealing a lot of CPU cycles especially from the Athlon.

[jbcool]

Originally posted by SkiBikeSki:
<STRONG>

How does one go about running it as a service? Is that where there is no window for it at all, and it runs entirely in the background? Sorta like "&lt; /dev/null"?

[ 04-23-2002: Message edited by: SkiBikeSki ]</STRONG>

Yes it is where there is no window at all.
More like a background process that needs no intervention.
No no like "&lt; /dev/null" you use that when the program actually produces output that you have no use for.

See the readme1st.txt file!
That is where this came from.

To setup the program to work as a service:

1. Open a DOS prompt to the directory you installed the client.
2. If it is not already there, create a new file called "handle.txt" and in it type your 8 character handle followed by Enter (or alternatively, run foldit.bat first, enter your handle when asked for it, then quit)
3. Type: foldtrajlite /install (and hit Enter)
4. The next time you reboot your machine it will start automatically

[alphamatrix]

DC Project: Ubero
DC version: CLI
CPU: 500Mhz G3
OS: OS 10.1.3
RAM: 384
Best Day's Work: 116 Units

Not bad for a iMac

[Shaktai]

Originally posted by jbcool:
<STRONG>

To setup the program to work as a service:

1. Open a DOS prompt to the directory you installed the client.
2. If it is not already there, create a new file called "handle.txt" and in it type your 8 character handle followed by Enter (or alternatively, run foldit.bat first, enter your handle when asked for it, then quit)
3. Type: foldtrajlite /install (and hit Enter)
4. The next time you reboot your machine it will start automatically</STRONG>

Okay, I read that, but I am a command line rookie. Heck, I've only had a windows machine at home for less than a month. 100% Mac before that and only Windows at work where IT would hang me out to dry if I opened a DOS prompt to anything. How do I open a DOS prompt to the directory?. Also, is running as a service any faster?

[jbcool]

Originally posted by Shaktai:
<STRONG>

Okay, I read that, but I am a command line rookie. Heck, I've only had a windows machine at home for less than a month. 100% Mac before that and only Windows at work where IT would hang me out to dry if I opened a DOS prompt to anything. How do I open a DOS prompt to the directory?. Also, is running as a service any faster?</STRONG>

Since your using XP look for command prompt under the menu.
"Start", "All Programs", "Accessories", "Command Prompt"

Once you get the command prompt open then you will need to change your directory to where the progarm is located. Which is the "cd" command followed by the directory path where the file is stoared. Which under XP might be long enough where it will just confuse you. If you really want to do this e-mail me as it will take a few e-mail or a phone call to help you get through this.

Is it faster than running it in quiet mode. Doubtful. Over the length of the contest it might give you a few extra structres but not enough to make a difference.

[SkiBikeSki]

[ 04-24-2002: Message edited by: SkiBikeSki ]

[jbcool]

Originally posted by SkiBikeSki:
<STRONG>

[ 04-24-2002: Message edited by: SkiBikeSki ]</STRONG>

Hey look everyone. We were able to leave SkiBikeSki speechless!!!
I'm guessing we won't see this happen very often.

[mikkyo]

Sounds like the athlons do very well on all the DC projects.
Guess they designed the thing to wrok best with general types of code.

As to postings.. Well some people need direction.
Give them a topic and off they go.

[mikkyo]

Oh and I almost forgot...



Like add another post to my tally