[SkiBikeSki]

About the . I had something to say, but then I realized it was answered already much farther up the topic. So rather than look like an idoit, I decided to look cool.

[jbcool]

Originally posted by mikkyo:
<STRONG>Oh and I almost forgot...



Like add another post to my tally</STRONG>

I am guessing that your post numbers never go up because of the great weight on your sholders in holding the Ubero team up.

[jbcool]

Originally posted by SkiBikeSki:
<STRONG>About the . I had something to say, but then I realized it was answered already much farther up the topic. So rather than look like an idoit, I decided to look cool. </STRONG>

Just good natured ribbing. And how could you look like and idiot?
You can't even spell it. I on the other hand am very fluent in the speech patterns of an idiot, learn from birth.

[mikkyo]

I need to post more so people take me seriously
Actually I need to get my Star rating up so I don't look like such a newbie and insignificant member of the community.

Ok so there really isn't a reason to post a lot.

Maybe since the DC world encourages competetion and everyone here is into DCing then they all like to see their forums "stats" rise.

I wish this forum were like Ubero where you can rate the other people... Now you know why several Team MacNN members have 5 stars.

[mikkyo]

Some more numbers:
All Ubero CLUI OS X
G4 400 - 140
G4 450 - 150
G4 466 - 170
G4 533 - 185
G4 450DP 2 clients - 300
G4 466DP 2 clients - 330
G4 533DP 2 clients - 350
G4 800 - 265

[SkiBikeSki]

Originally posted by mikkyo:
<STRONG>Some more numbers:
</STRONG>

Whoa. Just give me a moment to add these to the table. I think you also gave me some the other day, that I haven't yet added to the table. I wish there was some sort of automatic entry form, like on team.macnn.com or something.

[Shaktai]

Originally posted by mikkyo:
<STRONG>Some more numbers:
All Ubero CLUI OS X

G4 533 - 185

G4 800 - 265</STRONG>

Okay mikkyo, what is your secret? Your G4 533 matched my best from my Power Mac 800 quicksilver. Your G4 800 stats are way above mine for Ubero. My iBook 600 only did about 110 on its best day.

iBook 600 256k RAM
Power Mac 800 512k RAM.

[mikkyo]

Originally posted by Shaktai:
<STRONG>
Okay mikkyo, what is your secret? Your G4 533 matched my best from my Power Mac 800 quicksilver. Your G4 800 stats are way above mine for Ubero. My iBook 600 only did about 110 on its best day.

iBook 600 256k RAM
Power Mac 800 512k RAM.</STRONG>

You running the CLUI? With the java memory tweaks?
That might be one big difference.
You log out when not using the computer and leave the agents running in the background?

One big difference between the 533 and 666+ G4 is the L2 cache.
The 533 has double the L2 size which makes a big difference in performance, especially with just one app running reusing the same code.
My iBook 600 has only 128Mb RAM and it does better than yours.
It makes me think your aren't running the java tweaks.
More RAM, better Ubero.

[Shaktai]

Originally posted by mikkyo:
<STRONG>

You running the CLUI? With the java memory tweaks?
That might be one big difference.
You log out when not using the computer and leave the agents running in the background?

One big difference between the 533 and 666+ G4 is the L2 cache.
The 533 has double the L2 size which makes a big difference in performance, especially with just one app running reusing the same code.
My iBook 600 has only 128Mb RAM and it does better than yours.
It makes me think your aren't running the java tweaks.
More RAM, better Ubero.</STRONG>

Well apparently not the correct memory tweaks. I had used the MDC3 app to configure for me. Found the script you posted, downloaded it and tried it out on my iBook. Followed the readme instructions and it worked like a charm. Amazing difference. It will be interesting to see what my PM 800 can do after the dFold competition.

[SkiBikeSki]

I made the appropriate adjustments to the Ubero table. BTW mikkyo where did you get and Dual Processor 466MHz. I had no idea the existed. Was it a DP450 that was overclocked?

[SkiBikeSki]

I though I would try out the GCC and Intel Linux compilations of the Dfold client on the PC. I've got an Athlon XP, and am using Mandrake 8.1. Although the Intel compilation didn't give any errors, it didn't really crunch any data. I saw it work out the strucute for one protein, but then it never advanced to the next one. The more standard GCC compilation worked fine though. No errors, and it seemed to be fast. However after I let the client run overnight I saw that was not the case. It completed 7565 structures in 15 hours, for an average speed of 504/hour. That's shockingly slow, especially since I wasn't even using X, which has been kown to slow down CLI apps. I guess until the comp. is over, I won't be running the Linux version.

[mikkyo]

Originally posted by SkiBikeSki:
<STRONG>I made the appropriate adjustments to the Ubero table. BTW mikkyo where did you get and Dual Processor 466MHz. I had no idea the existed. Was it a DP450 that was overclocked?</STRONG>

Ding! Ding! Cigar to the winna!

[SkiBikeSki]

i thought one had to overclock in increments of half the bus speed; ie 50Mhz. Did you tweak the bus speed too?

[mikkyo]

Originally posted by SkiBikeSki:
<STRONG>i thought one had to overclock in increments of half the bus speed; ie 50Mhz. Did you tweak the bus speed too?</STRONG>

yeah
And anticipating your next question..
I even did some soldering too!

Definately not worth the trouble though.
And knowing you will ask how...

Tweaking the bus

[mikkyo]

DC Project - Ubero
DC version - CLUI
CPU - Athlon 1900 XP
OS - linux Red hat 7.2
RAM - 512 DDR
Bus Speed - 133
Best Day's Work 650 WUs

[Shaktai]

Originally posted by mikkyo:
<STRONG>DC Project - Ubero
DC version - CLUI
CPU - Athlon 1900 XP
OS - linux Red hat 7.2
RAM - 512 DDR
Bus Speed - 133
Best Day's Work 650 WUs</STRONG>

Okay, I've got to ask, because that is way better than mine.

Which Java version, and how do you apply the memory tweak?

1900XP with Mandrake Linux 8.2, 256 mb SDRAM, and no where near your numbers, and yes I am a total Linux newbie.

BTW, your last script worked wonders on my Macs. A huge performance boost.

[ 05-10-2002: Message edited by: Shaktai ]

[SkiBikeSki]

Although I haven't tried Ubero in Linux, I have done a lot with it in Win2K. For any computer to produce those sorts of untis/day it would have to be an Athlon of some kind, running the IBM Jave VM. For me that just about tripled the productivity.

I've yet to see how many units/day my PC can do, now that's it's overclocked to 1.57GHz.

[ 05-11-2002: Message edited by: SkiBikeSki ]

[Shaktai]

Originally posted by SkiBikeSki:
<STRONG>Although I haven't tried Ubero in Linux, I have done a lot with it in Win2K. For any computer to produce those sorts of untis/day it would have to be an Athlon of some kind, running the IBM Jave VM. For me that just about tripled the productivity.

I've yet to see how many units/day my PC can do, now that's it's overclocked to 1.57GHz.

[ 05-11-2002: Message edited by: SkiBikeSki ]</STRONG>

mikkyo's configuration and mine are very similar except for the RAM. We both have Athlon XP 1900+ (1.6ghz) But his performance is exceptional. We use different Linux versions, and of course it is obvious that he understands command lines, java, and all the other stuff that I don't have a clue about. I am going to get a solid benchmark on mine, and then have to figure out how to improve it down the road when Ubero gets active again. I finally just figured out how to get it to use text or quiet mode, maybe that will help. My java is Sun 1.4.

[mikkyo]

Originally posted by Shaktai:
<STRONG>
Which Java version, and how do you apply the memory tweak?
</STRONG>

IBM 1.3 jre is much faster than 1.4 or 1.3 sun.
The memory tweak is the same in ibm java except for the app heap.
java -Xms300m -Xmx300m -Xss100m -Xoss100m -cp ....
Do a java -X to see the settings.

Text client way faster than gui.

I have an ecs k7s5a MB and my RAM is DDR 2700.
I have the bios ram settings at the fastest possible options.
This may be the difference.
If you can run your FSB at 266, you will do much much better.
I haven't been able to yet because I can't boot off a damn floppy to update the bios to get fast bus settings.

Also my linux is in console mode, not X-Windows.

[Shaktai]

Originally posted by mikkyo:
<STRONG>

IBM 1.3 jre is much faster than 1.4 or 1.3 sun.
The memory tweak is the same in ibm java except for the app heap.
java -Xms300m -Xmx300m -Xss100m -Xoss100m -cp ....
Do a java -X to see the settings.

Text client way faster than gui.

I have an ecs k7s5a MB and my RAM is DDR 2700.
I have the bios ram settings at the fastest possible options.
This may be the difference.
If you can run your FSB at 266, you will do much much better.
I haven't been able to yet because I can't boot off a damn floppy to update the bios to get fast bus settings.

Also my linux is in console mode, not X-Windows.</STRONG>

Thanks that is a big help, and a big difference in settings. I will have to look to see if I can bet a bios update to allow 266 FSB. I am at 133 currently. I will also try booting into console mode and not using XWindows. I need to learn that anyway, to get Distributed Folding production up.

When you only have a small garden, you have to make the best of what you have.


[ 05-12-2002: Message edited by: Shaktai ]

[mikkyo]

Another one...

DC Project - Ubero
DC version - CLUI
CPU - Athlon 1900 XP
OS - linux Red hat 7.2
RAM - 512 DDR
Bus Speed - 133
Best Day's Work 733 WUs

[SkiBikeSki]

I updated the Ubero table.

It's been a long time since anyone else submitted some benchmarks. In fact since I created the unofficial benchmark tables, barely anyone has submitted stuff.

[ 05-18-2002: Message edited by: SkiBikeSki ]

[mikkyo]

How sad.
Come on people!
We need to know what machines can do how many WUs/day.
And if some has tweaks or exceptionally good stats we want to know why.

[Shaktai]

Well I posted what I have, but I did just download the IBM java for Linux, and get it running on my Athlon tonight. Once I have some numbers I will post them, probably within 24 hours. I'll find out how much faster that really is, and see if I did it right.

Thanks for your help Mikkyo. Now if I could just type accurately the first time, and remember to CD to the correct directories, I wouldn't have wasted an hour trying to get things going.

Seems like what we need is more SETI & RC-5 benchmarks, along with a bigger variety of dFold benchmarks while we are on the same protein.

[ 05-18-2002: Message edited by: Shaktai ]

[enola]

This is for dFold.

2xG4 1GHz 1770/hr

This is over a 9 hour period running two clients, so each client was doing this rate. Finder and Terminal only apps running. OS X 10.1.4 w/512MB RAM. Running in quiet mode with a nice of 1. I made my calculation using cpu time (what you get in the top or ps commands) not chronological time. I've noticed that these two times don't correspond with each other.

[SkiBikeSki]

Here's the difference between CPU time and chronological time.
Hopefully everyone knows what chronological time so I won't go into explainging that.

CPU time is the total amount on time on the CPU a program has spent. In many cases an app isn't the only thing running on a CPU, there are also OS's and other app's. So it makes sense that for all these things to share the CPU each one of them can only get the CPU for a portion of a sec, and then the others get the CPU. So CPU time is the time that the app was actually running. EG If an app runs at 50% CPU power for 2 hours chronologically, it racks up 1 hour of CPU time.

BTW Enola, I'm pretty sure all my stats are in chronological time. Which is nice since that gives a better idea of real world performance.

[enola]

SkiBikeSki,

On your benchmark for the dual G4's are you running one client? Or running two and aggregating up the numbers? Also, are those numbers for the previous protein or the current one?

I started another benchmark session this morning to get a sense of the production rate using chronological time. My problem is I forget to write down when I started.

[SkiBikeSki]

I'd like the benchmarks to be the best-possible yet still real-world benchmarks. So I guess in the case of a DP PowerMac, I want the total stats for 2 instances running simultaneously.

When I started collecting stats, the old protein was running. But when the switch was made I threw out all that data and started from scratch. So the short answer is, all the benchmarks are for the current protein.

[enola]

Okay, here's my amended benchmarks for dfold:

2xG4 1GHz 3205/hr

This is averaged over 12 real time hours, 2 clients running simultaneously, results summed together. The computer was used as normal so the environment wasn't very optimized.

I'm going try at some point to benchmark using console mode to see what difference that will make.

[typo: Make that 3205 not 3250.]

[ 05-19-2002: Message edited by: enola ]

[SkiBikeSki]

OK. I've added Enola's stats to my dfold page Let's keep these numbers coming!

[enola]

Originally posted by SkiBikeSki:
<STRONG>OK. I've added Enola's stats to my dfold page Let's keep these numbers coming!</STRONG>

Geez, you're quick! Sorry but I transposed the last two digits of the numbers. It should be 3205/hr.

[SkiBikeSki]

I made the fix to my dFold page.

[mikkyo]

How are you guys checking your crunch rate for dfold?

[ 05-19-2002: Message edited by: mikkyo ]

[enola]

Originally posted by mikkyo:
<STRONG>How are you guys checking your crunch rate for dfold?
</STRONG>

There is a progress.txt file that gets created when the client is running. It's usually in the same folder as the client binary and support files.

[SkiBikeSki]

Or you can use my script which repeatedly outputs the contents of that progress.txt file, and the uptime. Make it easy to find the work done and the time it took. It makes it easy to keep tabs on the progress of the client, but not was CPU time for WindowManager or Dock. The Dock can use as much as 30% just trying to draw the window contents of the client.

[enola]

Originally posted by SkiBikeSki:
<STRONG>Or you can use my script ... </STRONG>

Thanks for the reminder about your script. I was intending to try it after the race but I forgot. I'm using it right now!

[SkiBikeSki]

Originally posted by enola:
<STRONG>

Thanks for the reminder about your script. I was intending to try it after the race but I forgot. I'm using it right now! </STRONG>

Another satisfied customer.

Here's a benchmark that I'm not going to add to my table. A single PIII running XP with 128MB 133 RAM. Running un-interupted for 24 hours, produced 5000! I feel really sorry for my friend who has to use that computer.

[Grozni Majmun]

Here's a benchmark that I'm not going to add to my table. A single PIII running XP with 128MB 133 RAM. Running un-interupted for 24 hours, produced 5000! I feel really sorry for my friend who has to use that computer.

Whoa! That sounds pretty slow. What's the speed in MHZ of that PIII, if you know?

[slipjack]

Anyone want to write a benching program? I'll make the UI for it if it's in Cocoa... I've learned that much.

[SkiBikeSki]

oops! I for got to mention its a 1GHz. It's brand new and a custom-build job. A friend of his put it together from new parts all bought separately. Here's a better rundown of it's components

PIII 1GHZ - Best available at the time
128MB 133MHz SDRAM
20GB 5400RPM ATA66 HD
AOpen Mobo - not sure about it's model #
TNT2 Ultra Graphics
15" Monitor - again not sure
Simple 105-key keyboard
Simple 2-button mouse
3 piece speakers - I am really surprised that these poor speakers come with a sub-woofer

I'm not sure how much he paid for it, but I hope it's not more than $500.

[ 05-20-2002: Message edited by: SkiBikeSki ]

[TNproud2b]

Originally posted by SkiBikeSki:
<STRONG>oops! I for got to mention its a 1GHz. It's brand new and a custom-build job. A friend of his put it together from new parts all bought separately. Here's a better rundown of it's components

PIII 1GHZ - Best available at the time
128MB 133MHz SDRAM
20GB 5400RPM ATA66 HD
AOpen Mobo - not sure about it's model #
TNT2 Ultra Graphics
15" Monitor - again not sure
Simple 105-key keyboard
Simple 2-button mouse
3 piece speakers - I am really surprised that these poor speakers come with a sub-woofer

I'm not sure how much he paid for it, but I hope it's not more than $500.

[ 05-20-2002: Message edited by: SkiBikeSki ]</STRONG>

If it doesn't complete 60,000 structures per day - it's the only 1GHz P3 in the world that doesn't.

edited to add: you are using the commandline client, right?

[ 05-22-2002: Message edited by: TNproud2b ]

[Shaktai]

While it is too early to be certain, it appears initially that the new FAH3 Console Beta Windows & Linux clients, may indeed be faster, even on the old Science Cores.


Protein: proteinAg29a -- 5 point unit
CPU: Celeron 1.3 ghz
OS: Windows XP
FAH2 time: 26.7 hours
FAH3 time: 21.6 hours

Protein: 127 p127_wwext -- 2 point unit
CPU: Athlon 1900+ on a slow mother board
OS: Mandrake Linux 8.2 with KDU windows environment
FAH2 time: (from memory) 9-10 hours.
FAH3 time: 7.7 hours (two protein average)

There is no Mac Beta Client to test yet.

Sadly, I just discovered that I setup the PC's with an incorrect Team ID, and my credits from today, went to another team.
Darned uncoordinated fingers! I need a computer that will do what I want it to do, not what I tell it to do. Still can't figure out how I messed up typing two digits and having it become four or five. Definitely my error though. Too many medicines in my bloodstream I guess.

I would like to see some preliminary comparisons from other F@H users. I think though that Reader50 was right in an earlier post. For F@H, you probably need to get a weeks overall Point average for a particular, OS & CPU combination. Need to get that KFold program downloaded again and re-installed, so I can track the proteins and time over a longer time basis. Supposed to be someway I can get Linux to communicate with the Macs, so that I could eventually monitor them as well remotely from the Linux machine. I just don't know how to do it yet, or what communication software and configurations are needed.


[ 05-22-2002: Message edited by: Shaktai ]

[SkiBikeSki]

I certainly hope that F@H3 for OS X has a more reliable and consistant score/week. I'd like to collect benchmarks for it.

[SkiBikeSki]

Originally posted by TNproud2b:
<STRONG>


If it doesn't complete 60,000 structures per day - it's the only 1GHz P3 in the world that doesn't.

edited to add: you are using the commandline client, right?

[ 05-22-2002: Message edited by: TNproud2b ]</STRONG>

I've set it up to run as a service, since he's using Win XP.

BTW TNproud2b, preemptive congrats on becoming Addicted to MacNN! (almost 1000 posts)

[Microns]

[QUOTE]Originally posted by Shaktai:
[QB] Following are SETI benchmarks

...snip...

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

...snip...

Recently I recruited a 1.8Ghz WinTel box to the Team MacNN camp but it is very slow (1.5 DAYS per WU) and I can't figure out why. Any ideas greatly appreciated!

The box is:

Dell Optiplex GX240 DHS w/[email protected], 1gig RAM, BIOS A03 3/1/02, 256k level 2 cache, 400Mhz front side bus.

OS is:

W2K Professional ver 5.00.2195 Service pAck 2 Build 2195

Later...

[ 05-24-2002: Message edited by: Microns ]

[SkiBikeSki]

Maybe it's running other apps in the back ground. Use the 3-fingered-salute to see what other apps are eating up CPU power. It also might just e the fact that the first ever unit takes a little longer than the rest. Crunch some more and see if it gets any faster.

[Microns]

Originally posted by SkiBikeSki:
<STRONG>Maybe it's running other apps in the back ground. Use the 3-fingered-salute to see what other apps are eating up CPU power. It also might just e the fact that the first ever unit takes a little longer than the rest. Crunch some more and see if it gets any faster.</STRONG>

Forgot to mention that other processes use less than 1% of the CPU. This is the first WU so your suggestion to wait for others may be informative.

Why is the first WU slower than the rest? Is it something unique to Wintel boxes?

Later...

Originally posted by Microns:
<STRONG>

Forgot to mention that other processes use less than 1% of the CPU. This is the first WU so your suggestion to wait for others may be informative.

Why is the first WU slower than the rest? Is it something unique to Wintel boxes?

Later...</STRONG>

You got a problem.
Check the system, it's running dog slow. (Although dells do run slow, that's REALLY slow)

[SkiBikeSki]

I have a friend with a 1.7GHz P4 from Gateway, and his first ever work unit to 20 hours. But all the ones after that only took about 6 hours. However on Athlons and any PPC processor I have not seen this, and I can't explain it at all.

[Microns]

Originally posted by SkiBikeSki:
<STRONG>I have a friend with a 1.7GHz P4 from Gateway, and his first ever work unit to 20 hours. But all the ones after that only took about 6 hours. However on Athlons and any PPC processor I have not seen this, and I can't explain it at all.</STRONG>

Well, the first WU doesn't seem to be the problem. After 20hrs we're only 55% into the second WU. So both the first and second WU's will be much slower than I'd expect for a [email protected] (30+hrs).

The box otherwise appears to be running fine which makes me wonder if Dell has cripled this model somehow (small L2 cache?) in order to save money. I'd expect more general symptoms of distress if the box had a problem.

I'll re-install the S@H software (for no good reason other than I can do it easily). Any of the Wintel experts here have any other thoughts? Appreciate your help.

Later...