AMD Athlon XP question

[Logan[TeamX]]

Hey there, I just signed up today.

I'm not new to distributed computing, but I am to this project.

Question for all of you knowledgeable types: How does the text client perform on an AMD Athlon XP as opposed to a P4 Northwood? What optimizations are being used?

Reason I am asking: I've decided to take a leave of absence from SETI@Home. I was running their text client on my work PC (P3-933MHz Coppermine, 256MB SDRAM, Win2K Pro SP3, Intel motherboard in a Dell Optiplex GX150), and averaged 8 hours per workunit.

On my home PC (AMD XP 1800+ overclocked to 146.7x2 FSB, 1686 MHz, 256MB DDR-2100, Windows ME, ECS K7S5A motherboard, NORMAL RAM timings) I averaged 3:10 to 3:45 hours per workunit.

Here's how I check out with the Distributed Folding text client and DF-GUI for the Distributed Folding text client:

Work PC: 2973 structures / hr, 71351 structures / day

Home PC (only been running for about 10 mins): 5473 / hr and 123526 /day.

Should I expect more?

Thanks from a new member,

Logan

[FoBoT]

most people only run the txt/command line client in DF , so most benchmarks are for it. it has an excellent service install for w2k/XP windows pc's

here are some benchmarks for similar cpu's to yours
as you can see , your results are right in the ball park
happy crunching! :)


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

Sample Size : 3610 structures over 3951 seconds.
Protein Size: 129AA

Structures Per Second: 0.91
Structures Per Minute: 54.8
Structures Per Hour : 3289
Structures Per Day : 78943

OS : Windows XP MHz: 1063
CPU: Intel(R) Pentium(R) III Mobile CPU 1066MHz
Client Switches: -df -qt -rt -s 3000
------------------------------------------------------------



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

Sample Size: 24660 structures over 17257 seconds.

Structures Per Second: 1.43
Structures Per Minute: 85.74
Structures Per Hour: 5144.35
Structures Per Day: 123464.40

Linux OS - Running Kernel Version 2.4.19
AMD Athlon(tm) XP 1800+ @ 1532mhz (256 KB cache)
------------------------------------------------------------

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

Sample Size : 2150 structures over 2616 seconds.
Protein Size: 129AA

Structures Per Second: 0.83
Structures Per Minute: 49.7
Structures Per Hour : 2985
Structures Per Day : 71639

OS : Windows NT 4.0 MHz: 866
CPU: GenuineIntel x86 Family 6 Model 8 Stepping 10
------------------------------------------------------------



------------------------------------------------------------
Distributed Folding Windows dfGUI v1.9 Benchmark

Sample Size : 3340 structures over 2375 seconds.
Protein Size: 129AA

Structures Per Second: 1.41
Structures Per Minute: 84.4
Structures Per Hour : 5065
Structures Per Day : 121557

OS : Windows 2000 MHz: 1607
CPU: AMD Athlon(tm) XP 1900+
Client Switches: -df -rt -s 999 -p 15
------------------------------------------------------------

[Logan[TeamX]]

Thanks for the quick reply!

What are these switches for each client, and what do they mean?

I have the GUI app set to run the client quietly, and to use as much RAM as it can. I've increased the buffer size, and that's about it. What other optimizations can I make?

Thanks!

Logan

[Paratima]

Check out this linky (http://www.free-dc.org/forum/showthread.php3?s=&threadid=1437). ;)

[MAD-ness]

Increased buffer size should always be used, IMO, and if you can use the "use ram" option then that is also a major plus.

All of the other 'switches' can be ignored by the vast majority of users. They have little performance impact.

However, if you are running the text you might try some benchmarks using the "-qt" option (no text ouptut) as it used to be that case that this improved performance on Windows9x. I haven't checked in a long time to see if this is still the case.

Welcome to the project.

[FoBoT]

the switch settings are in the readme1st.txt file that comes with the text client, they do the same things as you describe for the gui one (i have never used the gui one, unless you mean dfGUI)
dfGUI (http://gilchrist.ca/jeff/dfGUI/) is a great way to run the text version on windows 9x/ME (or when you can't run it as a service on w2k/XP)

[FoBoT]

Originally posted by MAD-ness
Increased buffer size should always be used, IMO

i just noticed that howard made a comment about this in the readme1st.txt file at the last protein change


Because data packets are now much smaller than they used to be, this option
is deprecated and it is suggested you always use '-df'. This will be the
new default or this option may be removed/ignored in a future release.

[Logan[TeamX]]

Thanks everyone :)

I'm hovering around the 130,000 structure a day count with my XP 1800+ system, so I'm pleased.

Logan :D :cheers:

[MAD-ness]

Logan: that new number is a lot more in line with what I see from my AthlonXP1800+. Your other number seemed a tiny bit slow, especially for the slight overclocking you had.

[Logan[TeamX]]

MAD-ness

It's the fault of Windows ME and terrible large-memory management, to be sure.

I can't get my vid card to initialize properly in Win2K, or I'd be running that with 1GB of DDR-2100 in a heartbeat. As it is I'm going to get another 256MB stick of DDR-2100 ASAP.

Logan

[SSJ4_Son_Goku]

Originally posted by FoBoT
------------------------------------------------------------
Distributed Folding Linux Benchmark Script V1.0

Sample Size: 24660 structures over 17257 seconds.

Structures Per Second: 1.43
Structures Per Minute: 85.74
Structures Per Hour: 5144.35
Structures Per Day: 123464.40

Linux OS - Running Kernel Version 2.4.19
AMD Athlon(tm) XP 1800+ @ 1532mhz (256 KB cache)
------------------------------------------------------------

------------------------------------------------------------
Distributed Folding Windows dfGUI v1.9 Benchmark

Sample Size : 3340 structures over 2375 seconds.
Protein Size: 129AA

Structures Per Second: 1.41
Structures Per Minute: 84.4
Structures Per Hour : 5065
Structures Per Day : 121557

OS : Windows 2000 MHz: 1607
CPU: AMD Athlon(tm) XP 1900+
Client Switches: -df -rt -s 999 -p 15
------------------------------------------------------------

Hello everyone,

I'm new to DF. I've been to several other projects before, but then I decided to try DF.

And I have a question about the quote above.

I'm running an AMD Athlon 2200+ XP @ 1.8 GHz. And I average at about 0.94 structures per second. Now according to the quote above slower computers than mine are completing structures faster. Why?

And I also have another question. I just downloaded the dfGUI a few minutes ago. Is dfGUI the "spy" for DF. I know that for example TSC and SETI@HOME have spys, and I was looking for a spy for DF, since I'm running the text client. The closest I could get to a DF spy was the dfGUI, so, is that the DF spy?

Thx in advance.

SSJ4_Son_Goku

[bwkaz]

For the first question, that benchmark was done using a VERY old (and much, much shorter, and quicker) protein. Everyone's machines are running much slower now. Notice that the date on the post you quoted from is seven months ago? We've gone through at least ten different proteins since then.

For the second question, what do you mean by "spy"? dfGUI monitors the DF client's progress, and allows you to start and stop it (and do a few more things), if that's what you're talking about.

[SSJ4_Son_Goku]

To 1. Ok, thx, that makes sense.

To 2. Yes, that's what I mean, in other projects they call the things that monitor the projects spy, because that is what a spy does, it watches things.

Thx for the quick reply.