new client/protein is out, lets hit it

[FoBoT]

:cool:

the new protein is available, lets grab it and crunch away! :jester:

a quick ROUGH benchmark on speed

Pentium 4 1.8 Ghz running red hat linux 7.2 with the icc version does ~ 900 structures in an hour, so it will do between 20,000 to 25,000 structures a day, i am GUESSING

it uploads each 3000 structures

[Angus]

I'm on it!

dfGui is reporting just over 27000 structures per day for my P4-2.0Ghz Win2K box.


:|party|:

[pointwood]

6748 structures/day on a cel300@450 running win2k.

[ECL]

Is anyone else seeing abysmal performance on AMD processors with this new protein?

I've got a 1700+ XP (Win98) which hasn't done better than 7000 / day since the changeover, even after a reboot.

Right next to it is a dual PIII 600 (Linux 2.4.2-smp) which reports 9640 / day / CPU.

High-school math tells us that the dual PIII is now 2.75 times as fast as a machine which was out-crunching it by a serious margin this morning.

I'd like to see numbers from other AMD users to see if this is a fluke or something endemic to this CPU model. If the new protein is AMD-hostile I'll switch this box to another project.

[FoBoT]

i only have intel at work :rolleyes:
i will load up my
1.0 Ghz Athlon
1.2 Ghz Athlon

when i get home and check it out

[Elspuddy]

got all my pc's runing df now, :D
and just up dated them all to the new version :)

down side, i'v swaped teams to jsi hardware ! :moon:

[ECL]

I think I've got it figured out.

It's not AMD v Intel, it's ANSI art v quiet mode.

I switched to quiet mode on the AMD and it's now cruising along at ~25,000 / day, which is modest improvement over 7k.

I'd tried benchmarking verbose v quiet on this box and on the Linux boxes on the old protein and found the difference to be negligible. It isn't negligible now on the Windows box, but a side-by-side comparison on Linux shows that it's still about even.

[Angus]

It seems that the P4 is still at a disadvantage...

A P3-550 on WinNT4.0 is crunching the new protein at almost 14000 structures per day.

If it was strictly a cpu speed function, then I would expect the P4-2.0GHz box to do much better than the 32000 per day where it seems to have settled after an hour and a half running.

Both of those are running under dfGui with the client in hidden mode.

[Terminator]

Here's a couple of mine with the new protein :)

XP1800 @ 1667mhz - Redhat 7.3 - 25,000/day

P4-2.2 - WinXP Pro - 27,500/day

[DATA]

Hi ppl

the S-MDc is up and running with 45 systems 51 cpu meantime - complete manual load - no idea yet the production figures - will make a list tomorrow night and se whats what :cool:

[IronBits]

Out of 4 Mandrake boxen, only 1 made the update correctly. 1 took off when I started it. 1 took off after 4 attempts to re-start it. and the last one, well, I had to manually download the client :(
Both w2k boxes did fine, the win98 box did fine.

[ECL]

Here are some rough benchmarking numbers I've collected for the new protein. They come from my machines plus various posts.

CPU (mHz,OS) Struct/Day
PII(Cel) 333 (333mHz,NT4) 6350
PII(Cel) 300 (450mHz,Win2K) 6748
PII 400 (400mHz,L 2.4.2) 6900
PIII 550 (550mHz,NT4) 14000
PIII 600 * (600mHz,L 2.4.2-SMP) 10500
PIII 700 (700mHz,L 2.4.2) 12400
PIII 700 (700mHz,Win2K) 11077
XP1600+ (1400mHz,?) 28800
XP1700+ (1477mHz,Win98) 23000
XP1800+ (1667mHz,L 2.4.2) 25000
P4 1.5 (1495mHz,Win2K) 27000
P4 2.0 (2000mHz,Win2K) 27000
P4 2.2 (2200mHz,WinXP) 27500


NB: Because the proteins are random, these numbers are not exact. Your milage will vary.

Win98 users (maybe other versions of Windows) should definitely run in quiet mode if they want to maximize their production. Linux users won't see much difference.

* This is an SMP machine, and it's averaging 10500 / day / CPU.

[Paratima]

One Linux box had to be restarted manually, the other 2 were OK. (All SuSE) Most of my boxen are currently W2K, all did fine. My last remaining W98 was OK. (Finally!) Me happy! :D :D :D This changeover was a big :sleepy:.

Oh yeah! :crazy: Last remaining NT4 also did OK.

[Moogie]

Up and running here. Had to select "Y" when I got home to accept the newest version but no probs here.

What's the advantage of running in quiet mode?

[digital_me]

P4 2.0A @ 2400mhz [20x120mhz] w/ Rambus running XP ~34500 S/Day

P4 1.8A @ 1800mhz [18x100mhz] w/ Rambus running Win2k ~30400 S/Day

DM

[MAD-ness]

We are attempting to compile a collection of benchmarks for the Distributed Folding text client and a test page can be found here ( http://tsf.dbestern.net/bench.html ).

The page is very empty, contains info from the previous protein and is in a very beta stage, but I will be collecting benchmarks that people have been kind enough to post on our forum and over at Free-DC as well, to help us get an initial set of data. Any suggestions, submissions or offers of help are appreciated. :)

Dyy's Linux & Windows Benchmark scripts
http://www.free-dc.org/forum/showthread.php3?s=&threadid=855

Distributed Folding GUI v0.5 for Windows
http://gilchrist.ca/jeff/dfGUI/

Send results to the email addy on the benchmark page, to me at madnessNOSPAM@flashmail.com or post the results in the TSF Perpetual Thread on the Ars Technica Open Forums (or here).

--------

Just ganked all that from the post at the TSF site.

None of us are web design gurus and we haven't put together submission forms or what-not, nor have we even really done much work on the page, but I will be collecting what benchmark info I can find and attempting to fill in the page. When we change proteins next time, I don't know what the plan is, though archiving the page and starting a new one sounds like a decent plan if nothing intelligent comes to mind before then.

Any help is appreciated, I will hopefully get up 20 or so results this weekend some time. If you have a better idea/page design/system for collecting and displaying benchmark info feel free to share it. :) This is just a hacked together thing to get some info out there for consumption by the hungry masses (hey, two people is enough for it to be "masses," right?).

[Dyyryath]

I've got a couple of suggestions, but no time to implement them right now. I've been considering this for Free-DC, but I've already got too much on my plate for the immediate future so I've pushed it back down my 'To-Do' list a ways.

First, use a database on the backend to hold the data. Linux/MySQL is always a good choice. It's free, it's fast, and it's well supported.

Second, rather than have users enter the stuff manually into a form, you write a small Java applet that you embed in a web page. The applet basically has a "Get Benchmark Now" button. When selected, it does exactly what my benchmark scripts do, it finds the user's Distributed Folding directory (you'd probably have a default path with an option to let the user browse to it if their's is different than the default) and calculates the speed right then and there. Then it uploads the data to the database.

This makes uploading easy. The user just needs a browser. No fuss, no muss.

Once the data's in the database, the number of ways to format it and display it is pretty limitless.

Want to show every benchmark for a particular processor/OS/memory combo? No biggie.

Want to graph the difference between different combos dynamically? Again no problem.

How about showing a single averaged result based on all submissions for a combo? Yup, once again, no problem. ;)

Like I said, I've considered doing this for Free-DC, but I just haven't gotten to it. You wouldn't believe the number of things like this I have on my 'to-do' list. :D :D

[Dyyryath]

MAD-ness: I'm going to make a sticky thread for people to just post benchmarks of the current protein.

Feel free to grab whatever you need from there.

[MAD-ness]

I had already thought of most of that, though the java script to 'grab' thier info is a very nice feature (though horribly ambitious from my POV).

I have been trying to get Apache/mySQL/PHP set up and running here on a Windows box but uh...well, been busy and a little bit intimidated by the whole process. Not so much PHP, it looks relatively logical and the syntax is similar enough to C to make it somewhat familiar, I have been reading various PHP guides, tutorials, etc. and it doesn't appear to be impossible or anything. ;) However, configuring an Apache server is a bit daunting. Then I have to tie PHP and mySQL into it.

Sounds like a goal for this weekend. I guess I could just use Access but I would really rather do it the non-gay way. Who knows how long it will be before I get it all set up and figure out how to do anything useful. I imagine someone will beat me to it, which is fine.

If I ever actually get to the point of having a database up and running and all that jazz, I might holler at you if I get stuck. Already looked through PHP scripts to output database data to HTML tables, etc. PHP forms a bit, etc. I have some CSV to HTML table converters for fooking with a stats collection program I have.

Either way, I will likely put the info into a database (access if nothing else, though the last time I worked with an MS database it was on a 286 with windows 3.0 as the OS) even if I don't do anything dynamic with the information.

[Dyyryath]

I'm doing quite a bit of tinkering with benchmarks right now, so I may set aside some time this weekend to slap something together.

I'll let you know if I do...

[MAD-ness]

I now understand the whole "java script to actively retreive thier DFGui info.

Manually putting the benchmark info into a SQL friendly format is a pain in the arse.