Frankenbot rebirth. The cass farm of DOOM.

Come join the... uh... er... fold.

Moderators: just brew it!, farmpuma

Postposted on Sat Nov 06, 2004 8:34 pm

don't you have a node you are trying to start.... without cat play

If you have a spare processor, we have 5 empty boards here awaiting processors.

Just FYI, the litte 650 duron will do a 240 pointer in a little over 4 days.

chat with emkubed and Idchafee about the equipment donations, I'm just a grunt assembler.
cass
Minister of Gerbil Affairs
 
Posts: 2266
Joined: Mon Feb 10, 2003 9:12 am

Postposted on Sun Nov 07, 2004 12:12 am

OK, diskless node #4 is running... epox 8kha+ 256M ram, athlon1200. OC'd up to 1400.... It is using 3dnow, and stuck on a P1089.... good lord that will take forever.. these things take 48mins/% on an axp2200. I'll see, if it looks slow, I will cast the WU asunder.

I have the 1000 tbird loaded, but the acoup web site won't respond so I can get jumper settings.

as soon as the site comes up all I have to do is push the start button, because the server is configured and the net cable is plugged up.

Right now, I have 5 TRFrankebots running

km3m w/axp1600 256M freeBSD4.8 Diskless
***********************
Msi nforce2 w/axp2400 512M FreeBSD 5.3 server.
**I am just loaning this to set up FBot The disks go with Fbot but the board stays. It folds under Fbot though***
************************
k7t w/650 (748) duron 256 pc133 freeBSD 5.3 diskless
ecs nforce2 w/axp1800 256M freeBSD 5.3 diskless
epox 8kha+ w/1200 (1400) tbird 256M freeBSD 5.3 diskless
cass
Minister of Gerbil Affairs
 
Posts: 2266
Joined: Mon Feb 10, 2003 9:12 am

Postposted on Sun Nov 07, 2004 10:20 am

cheesyking wrote:PS nice try Frankenbot, but I don't think you managed to get past me even for one update icon_lol.gif

You better add some power in a hurry... TRFbot is real close to you now.
cass
Minister of Gerbil Affairs
 
Posts: 2266
Joined: Mon Feb 10, 2003 9:12 am

Postposted on Sun Nov 07, 2004 11:38 am

Cool .. five bots in the field .. five bots in the farm .. and residuals from Halloween .. no wonder he's climbing the top 100 like a monkey going for a banana :D
WTG cass and all who donated
Image Image
.* * M-51 * *. .The Whirlpool Galaxy.
farmpuma
Minister of Gerbil Affairs
Silver subscriber
 
 
Posts: 2319
Joined: Mon Mar 22, 2004 12:33 am
Location: Soybean field, IN, USA, Earth .. just a bit south of John .. err .... Fart Wayne, Indiana

Postposted on Sun Nov 07, 2004 5:28 pm

cass wrote:
cheesyking wrote:PS nice try Frankenbot, but I don't think you managed to get past me even for one update icon_lol.gif

You better add some power in a hurry... TRFbot is real close to you now.



Noooooooooooooooooooooooooooooooooooooooooooooooooooooooooo :lol:
Fernando!
Your mother ate my dog!
cheesyking
Minister of Gerbil Affairs
 
Posts: 2276
Joined: Sun Jan 25, 2004 7:52 am
Location: That London (or so I'm told)

Postposted on Sun Nov 07, 2004 5:35 pm

Dang, I guess it is time to get more boxen up and running agin...seems like everyone is passing me now.... :-?
Image
Abit BP6 2X 300A@450 mhz, 1.5 GB PC133 ECC, Renditon Verite V2200, SB AWE32, Adaptec 2940U2W, 4.3GB Quantum Atlas 10K, Plextor Ultraplex
Starfalcon
Gerbilus Supremus
 
Posts: 11716
Joined: Mon Oct 14, 2002 10:43 am
Location: Evergreen Park, IL

Postposted on Sun Nov 07, 2004 11:45 pm

TRFrankenbot #4 diskless is now up and running.

Acorp 7kta MB w/256MB pc133 athlon tbird 1000. Actually only running 750mhz.

Numbnuts here pencil unlocked the proc only to find that the MB does not have Mulitplier selection so evidently it defaults to the lowest mult.

Brings the total nuber of Fbots here to 6... (1) 5.3 server, (1) 4.8 diskless and (4) 5.3 diskless
cass
Minister of Gerbil Affairs
 
Posts: 2266
Joined: Mon Feb 10, 2003 9:12 am

Postposted on Mon Nov 08, 2004 12:16 am

can't update the BIOS

or erase the pencil :wink:
Usacomp2k3
Gerbil God
 
Posts: 21311
Joined: Thu Apr 01, 2004 4:53 pm
Location: Orlando, FL

Postposted on Mon Nov 08, 2004 9:37 am

update bios to v1.8 from 1.6.. that added some stuff but no multiplier choice.

I may swap procs around and put the 650 duron in the acorp board, since I have yet been unable to get the mults to change on it. The k7t board that the 650 is currently in has fsb and mult setting.

eraser.... lol... I took some mineral spirits to it, but no joy. The board my be locked at a max mult. or 7.5 . I don't know.. it does run fine.
cass
Minister of Gerbil Affairs
 
Posts: 2266
Joined: Mon Feb 10, 2003 9:12 am

Postposted on Mon Nov 08, 2004 5:44 pm

received the gfx cards from idchafee today. box looked fine, I just took the top 4 out and looked at them.... look OK
cass
Minister of Gerbil Affairs
 
Posts: 2266
Joined: Mon Feb 10, 2003 9:12 am

Postposted on Mon Nov 08, 2004 6:06 pm

Swimming in cards there, are ya?

Wish we were swimming in $50 Athlons...
emkubed
Gerbil Elder
 
Posts: 5848
Joined: Mon Jun 02, 2003 9:28 am
Location: Limbo

Postposted on Mon Nov 08, 2004 11:26 pm

I got the problem with my sempron and the problem with the axp1800 in the Fbot farm straightend out. I was NFS mounting the same (folding) directory twice over the same place thus corrupting the files and causing the lock up.

I fixed the extra mount, compiled a new optimized kernel, deleted all the files in the folding dir and everything works beautifully.

I will get a comparative speed fix of freeBSD 5.3 vs 4.8 now that I have my sempron runnng in the fbot farm.

the little 1000 tbird is on a P1412 and is taking a cool 2hrs/% that means 8 days for 230 points.. I have noticed that the P14xx series suck for points. them and the p8xx are bad news.
cass
Minister of Gerbil Affairs
 
Posts: 2266
Joined: Mon Feb 10, 2003 9:12 am

Postposted on Tue Nov 09, 2004 9:03 pm

switched the 650 duron and 1000 athlons around, put the 1000 on the k7t MB so it would OC with multiplier, and put the 650 duron on the acorp board since the acorp won't change multiplier, but will change fsb. Currently the 650 is somewhere around 750, and the 1000 athlon is running 1200. Now I must turn my attentions to OC the axp1800 and axp 1600.

ran cvsup last night for server, did buildworld and build kernel with -O2 -funroll loops on compiler, I have ran install world, install kernel and merge master... just need to reboot.
cass
Minister of Gerbil Affairs
 
Posts: 2266
Joined: Mon Feb 10, 2003 9:12 am

Postposted on Wed Nov 10, 2004 3:16 am

cass wrote:switched the 650 duron and 1000 athlons around, put the 1000 on the k7t MB so it would OC with multiplier, and put the 650 duron on the acorp board since the acorp won't change multiplier, but will change fsb. Currently the 650 is somewhere around 750, and the 1000 athlon is running 1200. Now I must turn my attentions to OC the axp1800 and axp 1600.

ran cvsup last night for server, did buildworld and build kernel with -O2 -funroll loops on compiler, I have ran install world, install kernel and merge master... just need to reboot.

w00t .. you da man :D
farmpuma
Minister of Gerbil Affairs
Silver subscriber
 
 
Posts: 2319
Joined: Mon Mar 22, 2004 12:33 am
Location: Soybean field, IN, USA, Earth .. just a bit south of John .. err .... Fart Wayne, Indiana

Postposted on Thu Nov 11, 2004 2:13 am

I've been Folding with LWD now for a while. I just can't bring myself to jump ship with the systems I have now but I'd be willing to start up a Frankenbot machine if it's okay with you. If it is I'll work on getting a MB to get a 1.3 Athlon online.
oz_eod
Gerbil In Training
 
Posts: 8
Joined: Thu Jan 17, 2002 7:00 pm

Postposted on Thu Nov 11, 2004 10:03 am

oz_eod wrote:I've been Folding with LWD now for a while. I just can't bring myself to jump ship with the systems I have now but I'd be willing to start up a Frankenbot machine if it's okay with you. If it is I'll work on getting a MB to get a 1.3 Athlon online.

Hey, the more the merrier...

Welcome to the TR team (and to the Frankenbot effort).
(this space intentionally left blank)
just brew it!
Administrator
Gold subscriber
 
 
Posts: 37824
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Postposted on Thu Nov 11, 2004 10:24 am

oz_eod wrote:I've been Folding with LWD now for a while. I just can't bring myself to jump ship with the systems I have now but I'd be willing to start up a Frankenbot machine if it's okay with you. If it is I'll work on getting a MB to get a 1.3 Athlon online.


That'd be great, and we'd be appreciative. TRFrankenbot is the name.
emkubed
Gerbil Elder
 
Posts: 5848
Joined: Mon Jun 02, 2003 9:28 am
Location: Limbo

Postposted on Fri Nov 12, 2004 8:13 am

I bought an 18 port switch for $1.25 off ebay last night. If it works I will put it with the FBot.

The switch I bought is 16 10mbit ports and 2 100mbit ports. That should be fine, and I will test it. If it hurts production much we don't have to use it.

That switch emkubed sent is a godsend for trouble shooting. That thing has saved me a lot of trouble.

I have two more processors coming to put in fbot. I have a 1400 mobile xp and I am probably going to switch out one of my 1500/1600's for a 2200 and put the old proc. in fbot.

I am down to one problem, and several small optimizing things now. The one big problem is the diskless nodes starting before the server if you turn them all on at once. If the Diskless machine finds no dhcp, they stop. I think I can configure the nics to auto restart in that case, but I have not looked into that yet. Anyhow worst case you will have to turn on the server 30 seconds before the diskless.

I noticed my windows xp box with axp 2700 is getting a p113x frame done in 5 mins. The fastest machines I have in any of my FreeBSD farms take 5:47/frame. I am building a 3100 sempron with a disk to see if the NFS is slowing the thing down, or if the OS is. I have a 3100 sempron diskless unit up and running under freeBSD 5.3, and I have OC'd it all I can. The clock is cranked from 200 up to 232. I have moved the memory from 166 to 200 with no change in frame time. I set the memory back to 166 for stablility. HT bus I have cranked from 400 to 800Mhz with no change in frame time (maybe 3 seconds in 6 mins) I left the ht bus at 800 since it is running fine there.

Even with the processor running at 2088mhz, it is still 47 seconds/ frame slower than the axp 2700 and some slower than my axp 2200 under windows.

I am playing with compiler settings, and getting ready to try polling for the network to see if there is any way to speed up.
cass
Minister of Gerbil Affairs
 
Posts: 2266
Joined: Mon Feb 10, 2003 9:12 am

Postposted on Fri Nov 12, 2004 4:44 pm

Got the mb today. The machine should be up this weekend.
oz_eod
Gerbil In Training
 
Posts: 8
Joined: Thu Jan 17, 2002 7:00 pm

Postposted on Sat Nov 13, 2004 1:01 am

Machine is up but Folding not loaded yet. Might use DamnSmallLinux. Any suggestions on the OS? It's an MSI KM3M mboard with a 1.33 T-bird. I'll probably get a new CPU soon. I'm leaning on an XP 2800 but what's your thoughts on a Sempron?
oz_eod
Gerbil In Training
 
Posts: 8
Joined: Thu Jan 17, 2002 7:00 pm

Postposted on Sat Nov 13, 2004 1:31 am

The only sempron I have used is a 3100 which is socket 754. I like it though.

deciding between processors of the same breed, depends on what you want to do. For folding raw MHZ is all that matters. cache, memory speed, and fsb speed are irrelevant.

Far as OS goes, I have only folded on win98, win xp, and FreeBSD. FreeBSD is 15% slower than windows on tinkers. Gromacs there is very little difference. Setup whatever OS you are comfortable with and can get going. You can always change it if you don't like it.

You can probably pencil unlock the tbird and get 20% more out of it too.
cass
Minister of Gerbil Affairs
 
Posts: 2266
Joined: Mon Feb 10, 2003 9:12 am

Postposted on Sat Nov 13, 2004 3:28 am

Off the top of your head will the pencil unlock work with the KM3M? I haven't had a chance to check out the BIOS much yet with this mb but I didn't see any multiplier settings...
oz_eod
Gerbil In Training
 
Posts: 8
Joined: Thu Jan 17, 2002 7:00 pm

Postposted on Sat Nov 13, 2004 3:32 am

By the way, thanks for all the help. Any recommends on the linux command line switches?
oz_eod
Gerbil In Training
 
Posts: 8
Joined: Thu Jan 17, 2002 7:00 pm

Postposted on Sat Nov 13, 2004 8:21 am

oz_eod wrote:By the way, thanks for all the help. Any recommends on the linux command line switches?

The defaults are pretty sane now... you don't really need any switches other than maybe -forceasm.
(this space intentionally left blank)
just brew it!
Administrator
Gold subscriber
 
 
Posts: 37824
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Postposted on Sat Nov 13, 2004 10:07 am

km3m-v is this the right MB?

if so, what bios version.

I have two of these, I can play around and see if there is a multiplier. There may not be I don't know. Never tried one with unlocked proc... I throwed the pencil unloock out there without looking at the MB. Sorry if I have confused you.

JBI- I will PM this to you too, but for info purposes I will ask here too.

1. What processors do you have running under linux?

2. What is your frame time on a P112, or P113x,

3. What is your frame time on a p13xx?


For me, with an axp2400 frame time for a P112 or p113x is about 6mins 20 secs +- 5 secs, axp1800 runs about 7:30

A P1308 goes about 23mins 15 secs for me

both above times are diskless stock clocks
cass
Minister of Gerbil Affairs
 
Posts: 2266
Joined: Mon Feb 10, 2003 9:12 am

Postposted on Sat Nov 13, 2004 12:38 pm

Yeah, that's the right mb. I have a km2m and it let's you change the FSB a bit. Not sure what bios the km3m is. I'll have more time to mess around with it today.
oz_eod
Gerbil In Training
 
Posts: 8
Joined: Thu Jan 17, 2002 7:00 pm

Postposted on Sat Nov 13, 2004 4:58 pm

Got a few numbers for you, cass. My Linux systems are a crazy hodge-podge of CPU and memory speeds... mostly older Socket A systems, with one Athlon64 thrown into the mix.

=====

Athlon64 3000+ (2GHz), 512MB PC-2100/PC-3200, Fedora Core 2 64-bit
(This system got upgraded from PC-2100 to PC-3200 memory a few weeks ago... I've given separate frame times for the different memory speeds, where possible.)

p112: 6:17 (PC-2100); 5:51 (PC-3200)

p1131: 6:14 (PC-3200)

p1301: 16:06 (PC-2100)

p1308: 20:22 (PC-2100); 19:15 (PC-3200)

=====

Athlon XP 2000+ (1.67GHz), 64MB PC-133, LTSP 3.x

p112: 7:36

p1132: 7:24

=====

Athlon XP 1800+ (1.53GHz), 512MB PC-2100, Fedora Core 2

p112: 8:05

p1132: 7:54

p1308: 29:05

=====

Athlon XP 1700+ (1.47GHz), 256MB PC-2100, LTSP 3.x

p112: 8:55

p1131: 8:39

=====

Duron 1.1GHz, 256MB PC-2100, LTSP 3.x

p112: 11:19

p1132: 11:07

=====

A few things we can conclude from the above:

- It looks like the F@h assignment servers don't hand out p13xx WUs to slower CPUs. Only the AXP 1800 and A64 3000+ had p13xx WUs in their log files.

- Memory speed (going from PC-2100 to PC-3200) makes a noticeable, but not huge difference.

- It really does look like raw clock speed is the main factor in determining F@h performance, at least for AMD processors. Clock-for-clock, the crusty old 1.1GHz Duron performs pretty close to the Athlon64.
(this space intentionally left blank)
just brew it!
Administrator
Gold subscriber
 
 
Posts: 37824
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Postposted on Sat Nov 13, 2004 9:04 pm

I concur with your findings, I have not went from 2100 to 3200 on memory, I only went from 166 - 200 on speed with the 3100 sempron, and 2100-2700 on a couple of the Axp 2400's.

I noticed 5-10 sec (6min frame) difference is speed when changing memory timings, but I got a larger overall increase by down clocking the memory and OC'ing the piss out of the procs. It is close though. I would say your memory observations are correct. I don't use good memory either. kingston value ram is top of the line for me. I must be living good too, because I have not had any memory compatibility problems.

I have a standalone 3100 sempron system running now with a HDD to see if that makes a large diff erence in speed, but I don't think it will.

I did get a P1308 on the TRFbot 650 duron honestly... must have been all that was available.

The only thing that bothers me is that gromacs are far better for us linux people than tinkers. we are taking about a 15% hit over the same processor in windows, but running slightly better with gromacs.

The 3100 semprons I have will OC like no tomorrow, but even cranking from 1.8 Ghz to 2.2 does not help as much as it should. Of course at advanced speeds it kicks some butt on gromacs.

evidently freeBSD and Your linux are running the same linux subsystem.

Code: Select all
52server# ./rpm -q glibc
glibc-2.2.4-33


so far I have tried recompiling the world and kernel using -02 -funrollloops and -03 with no noticeable difference.

the new GCC is supposed to net an 11% increase in speed, but I have not figured out how to get it installed yet.

JBI, are you running polling on your network server, or are U using interrupts.? I am not sure if that question will make any sense to you, but if freeBSD, it is an option to force the kernel to check every so often for other interrupts needing servicing rather than just handing control over to the 1st interrupt and letting it have control (however scheduled) till finished. It is said polling is good for a heavy load to keep from starving processes (clients) in a heavy load net server environment.
cass
Minister of Gerbil Affairs
 
Posts: 2266
Joined: Mon Feb 10, 2003 9:12 am

Postposted on Sat Nov 13, 2004 9:19 pm

Thanks a bunch for taking the time to get the information JBI.
cass
Minister of Gerbil Affairs
 
Posts: 2266
Joined: Mon Feb 10, 2003 9:12 am

Postposted on Sat Nov 13, 2004 9:41 pm

cass wrote:I don't use good memory either. kingston value ram is top of the line for me. I must be living good too, because I have not had any memory compatibility problems.

Yeah, I think there's a stick of Kingston Value Ram somewhere in my mess of machines too... most of the crate farm is actually running generic PC-133 from my junk box (old sk00l!! :lol:). IIRC one of them is even running PC-100 overclocked to PC-133 speeds (because the mobo didn't have an option to run FSB at 133 and RAM at 100)... :lol:

I did get a P1308 on the TRFbot 650 duron honestly... must have been all that was available.

I have not yet checked all of my systems... but it sure seemed like the 1308s were preferentially being handed out to the faster ones.

The 3100 semprons I have will OC like no tomorrow, but even cranking from 1.8 Ghz to 2.2 does not help as much as it should. Of course at advanced speeds it kicks some butt on gromacs.

I've been thinking of throwing together another couple of diskless nodes... been checking the mobo listings at Newegg etc. for cheap micro-ATX boards that support the Sempron.

evidently freeBSD and Your linux are running the same linux subsystem.

Well... I've actually got quite a diverse range of Linux versions here. Redhat 8/9, Fedora Core 2, and LTSP 3. Fedora Core 2 is a 2.6.x kernel, the Redhat ones and LTSP are 2.4.x... glibc versions are all over the map as well. It's a bit of a mess, but I figure if it ain't broke... :lol:

so far I have tried recompiling the world and kernel using -02 -funrollloops and -03 with no noticeable difference.

The F@h client probably spends 99+ percent of its time inside its own highly optimized SIMD assembly code. So libraries, kernel, etc. are going to be a very minor contributor to overall F@h performance.

the new GCC is supposed to net an 11% increase in speed, but I have not figured out how to get it installed yet.

Unless you can get Stanford to rebuild the client with the new GCC, it probably won't make much difference.

JBI, are you running polling on your network server, or are U using interrupts.? I am not sure if that question will make any sense to you, but if freeBSD, it is an option to force the kernel to check every so often for other interrupts needing servicing rather than just handing control over to the 1st interrupt and letting it have control (however scheduled) till finished. It is said polling is good for a heavy load to keep from starving processes (clients) in a heavy load net server environment.

I am using the stock kernels... I imagine it is using normal interrupts. I also suspect that the only time there is a heavy load on the server is when I power cycle the entire crate farm and all of the nodes are trying to boot over the network simultaneously (which happens very infrequently).
(this space intentionally left blank)
just brew it!
Administrator
Gold subscriber
 
 
Posts: 37824
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

PreviousNext

Return to TR Distributed Computing Effort

Who is online

Users browsing this forum: No registered users and 4 guests