The State of the SMX Project

muddocktor

I haven't been active with the team for the last 9 months or so, but that doesn't mean that I don't care what the team does either. I've seen that the SMX committee has been putting new machines out quite a bit lately, so I wanted to see what the SMX rigs are doing now. Here's what I've come up with, from the EOC stats.

SM1 - 81 POINTS/DAY AVG
SM2 - 0 POINTS/DAY AVG, Last WU returned 6/26/05
SM3 - 75 POINTS/DAY AVG
SM4 - 122 POINTS/DAY AVG
SM5 - 73 POINTS/DAY AVG
SM6 - 0 POINTS/DAY AVG, Last WU returned 10/18/05
SM7 - 0 POINTS/DAY AVG, Last WU returned 2/12/04
SM8 - 54 POINTS/DAY AVG
SM9 - 187 POINTS/DAY AVG
SM10 - 314 POINTS/DAY AVG
SM11 - 130 POINTS/DAY AVG
SM12 - 115 POINTS/DAY AVG
SM13 - 61 POINTS/DAY AVG
SM14 - 34 POINTS/DAY AVG
SM15 - 95 POINTS/DAY AVG
SM16 - 111 POINTS/DAY AVG
SM17 - 59 POINTS/DAY AVG
SM18 - 54 POINTS/DAY AVG
SM19 - 75 POINTS/DAY AVG
SM20 - 171 POINTS/DAY AVG
SM21 - 205 POINTS/DAY AVG
SM22 - 128 POINTS/DAY AVG
SM23 - 127 POINTS/DAY AVG
SM24 - 74 POINTS/DAY AVG
SM25 - 5 POINTS/DAY AVG
SM26 - 76 POINTS/DAY AVG
SM27 - 386 POINTS/DAY AVG
SM28 - 66 POINTS/DAY AVG

As you can see, there are a few inactive SMX machines. SM7 has been down for over a year due to an asshat stealing the original machine. I see that you guys have reassigned it to Crypto, who hopefully will have it up and running soon.:)

The second machine that is down is SM6, which hasn't turned in any work for half a month. I guess one of you SMX committee members needs to go over to Icrontic and PM Citrix to check the machine out.

Last we have SM2 that has been MIA since June of this year. Has anyone been able to get in touch with redoulent about why he hasn't had it up and folding? Red used to be one of our most steady hosts, but he is also one of the original hosts that refused to sign a contract too, after the first few machines were stolen. :shakehead I hope that nothing bad has happened to him, but I would hate to think he's turned into a frigging thief also.

Now, to the active machines, it looks to me like some of them are performing below their potential. After all, if you are going to foot the electric bill to keep it running, then why not maximize the points return for your bucks.

For AMD systems with the present work being sent out, set the machine up to do timeless wu's, which are all Tinker units now. Most of them are 239 pointers or 241 pointers and they give a good steady points return, much better on a Tbird or AXP processor than a regular Gromacs wu. I have both my remaining AXP machines configured this way and they return around 150-200 points/day/processor. This is with a NF7 rig at 2400 and a MP dually at 2133. Another plus is that the Tinker wu's are not ram or memory bandwidth intensive, which is great for older machines that have only 128 MB ram and/or SDRAM.

For P3 machines, Tinker work isn't the greatest choice. Setting it up to do regular Gromacs work is probably your best option, unless you have 512 MB ram onboard. Then you can try the -advmethods flag and hopefully grab some BigWU work.

For a P4 based machine, 512 MB ram and using the -advmethods flag with bigpackets=yes set in the config file. This will let the P4 grab QMD work, which requires 500 MB minimum ram. The point output difference is tremendous when a P4 can get QMD's. Just look at the points/day output of the P4 that Jon and Sally are hosting and then compare it to the one that dancer is hosting, which only has 256 MB ram. If you can get him another 256 MB stick of ram, that 1.6 P4 will be knocking on the heels of the fastest AMD systems the SMX project has running.

QCH

First off.... Thank you Muddocker. A little poke for info is a good thing. These are more your machines, as a team, than the staff's. It's the F@H Staff's responsibility to make sure they are being used for the purpose that the team has determined. We try the best we can but we are trying to make our list of machines that are folding 100%.

That being said...

SM2 - Redoulent
Well... we have tried, and I mean several of the staff have tried, to no avail. We have been ignored, blocked, and to my knowledge, he has not responded. If anyone has information about redoulent (i.e. real name), please IM one of the staff so we can try other means to get a hold of him.

GnomeWizardd

Hey mud do u mean deadlineless on the tinkers?

muddocktor

QCH2002 wrote:

First off.... Thank you Muddocker. A little poke for info is a good thing. These are more your machines, as a team, than the staff's. It's the F@H Staff's responsibility to make sure they are being used for the purpose that the team has determined. We try the best we can but we are trying to make our list of machines that are folding 100%.

That being said...

SM2 - Redoulent
Well... we have tried, and I mean several of the staff have tried, to no avail. We have been ignored, blocked, and to my knowledge, he has not responded. If anyone has information about redoulent (i.e. real name), please IM one of the staff so we can try other means to get a hold of him.

Thanks for the update on Redoulent, QCH. That info might be hard to come by, since he originally got the rig when we were Team Icrontic before the hack. All the info we had from there is lost, unfortunately. Maybe one of the old time committee members who were around when SM2 was originally assigned still have his info. Although that has probably changed too. I never would have guessed that Red would turn into a lying thieving SOB that deserves to be hung by his nads with a rusty nail.

GnomeWizardd wrote:

Hey mud do u mean deadlineless on the tinkers?

Yep, that's what I'm talking about, Gnome. They do great on my AXP machines, much better than a regular Gromacs and not too much slower even on the magic p147x series or the 600 point BigWu units, which are in such short supply.

Garg

muddocktor wrote:

For AMD systems with the present work being sent out, set the machine up to do timeless wu's, which are all Tinker units now...

Sounds like a good idea. At only 128mb, IC_11 can't handle the big WUs. How do I set it up to just grab Tinkers?

QCH

muddocktor wrote:

Thanks for the update on Redoulent, QCH. That info might be hard to come by, since he originally got the rig when we were Team Icrontic before the hack. All the info we had from there is lost, unfortunately. Maybe one of the old time committee members who were around when SM2 was originally assigned still have his info. Although that has probably changed too. I never would have guessed that Red would turn into a lying thieving SOB that deserves to be hung by his nads with a rusty nail.

I'm not saying that he falls into the same boat as some of the other guys that flat out told us "F*#& off, It's mine and you're not getting it back." but he is being abnormally hard to reach. No one on the staff has any info either. We've traced him down to a few other forums but he's not responding to us there either. IMing him doesn't work either... we are about out of ideas.

muddocktor

It's real easy to do Garg. Just start up with the -configonly flag, then go through the config setup until you get to the line asking about receipt of work assignments larger than 5 MB and answer no there, then answer yes to the change advanced options line. Keep going through that until you get to the line that asks if you want work without deadlines and answer yes, then on the following line answer no to the question if you want to download work in batches. I'm attaching an example of how to set it up for timeless work with the console.

Sledgehammer70

SM25 - 5 POINTS/DAY AVG

somthing is not right with that! who has this bad boy?

drasnor

I do, it's being unstable. I don't know what the problem is but it needs to be rebooted about once a week. I've been monitoring it to see if there's a pattern but it seems like just general instability. I really want to get a watchdog card for the machine but it has no ISA slots and I haven't been able to find a PCI version through my normal surplus channels. The BIOS on it sucks too, it won't POST without a keyboard hooked up and there's no "Halt on All but Key/Disk"

If you look at it's production log the reason the average is so low is because it's up for a week, down for a week, etc.

-drasnor

Sledgehammer70

what is that bad boys specs?

drasnor

From SMx Project Page:
Motherboard: MSI KT3 Ultra2
Processor: AMD Athlon XP 1900+
Heatsink: Generic Socket A
Ram: Infineon 512MB PC2100
Hard Drive(s): Fujitsu 4.1GB
Video Card: ATI Technologies, Inc. 3D RAGE PRO AGP 2X
Network Card: Linksys LNE100TX (v4) Fast Ethernet Adapter
Case: Rosewill R103A
Power Supply: Generic 350W
CD/DVD Rom: NEC CDR-3000a 32x
Fans & Cooling: Generic 80MM (2)

We just put in a new hard drive because the old one literally ran out of space after some big work units (we were cutting it a little close anyway with 2.1GB). The memory checks out okay on memtest so the problem isn't there.

-drasnor

Sledgehammer70

I think I found the problem!

Processor: AMD Athlon XP 1900+


That is a pretty old CPU... maybe they could update you out fo the new Proc pile.....

But how is the coputer running in general. if you turn off Folding what is it running at? how is the ram..maybe some memory leakage is going on...

drasnor

1900+ is not a bad processor; I used to run a dual 1900+ system and it put me in the top 50 by itself. The computer is pretty well behaved, I haven't actually caught it crashing and it hasn't had any trouble with the massive compiler jobs it takes to build a Gentoo system and it goes a week at full CPU usage. It get points for what it turns in, so it's producing good data. Memory leak is possible but unlikely since it only has a Gentoo base system that's kept up to date and a 1GB swapfile. I can check that though first I need to find a console system status monitor (Gentoo base system has no GUI). I'm trying to catch a kernel error but it seems like the machine just locks up for no apparent reason.

-drasnor

Sledgehammer70

It isn't a bad proc but a 2800+ put up allot more fight I thinki still have a few 1900+ in a box somewhere! It seemed at the time 4 years ago they were easy to collect! they were all over the place

Garg

Is that 1900+ overclocked at all? I've managed to OC the 1600+ in IC_11 to 2000+ speed.

Hey Sledge, if you've got a box of 1900+s laying around, I'm sure we could make good use of them (if anybody mentions spare parts around me, they better be prepared for me to try to assimilate them )

Sledgehammer70

I have actually been going through boxes trying to find parts to build new Towers to fold. I know I have an 18 x 24 x 18 box full of CPU's in plastic covers.... Once I find that box. and than shift thought the 1995 chips that are just to old to think of I will see what I got and whatever I can make a system of I would be happy to send them to people who can make good use of them.

If you’re wondering why I have all this. My old friend owns a cop repair store and ought recycled comps. Well when he went out of business he had about 280 comps that were going to be chucked. I gave him a couple hundy bucks and took every cpu, memory chip, CD drive Video Card and PSU I could get my hands on. I used and sold allot of the items and made good use of it all but I still have that box of CPU's somewhere

drasnor

No, it's not overclocked at all and I doubt it would overclock very well being a locked Palomino. I'm holding off on the overclocking until I figure out a way to get the machine stable. I don't know why, but my intuition says it's a motherboard issue.

I just had a thought about trying to build a serial port watchdog; set up a cron job that generates serial port activity every hour resetting a solid state timer or purging a microcontroller register. If the timer reaches some voltage or counter value the circuit closes the reset switch. Doing it with a microcontroller wouldn't be so bad but my basic stamp is a bit on the large side and my HC05's require support circuits (clock generator et al) and except for one EPROM are all OTP. Solid state would be a lot cheaper but it would almost certainly involve some sort of RC timing circuit and the capacitor would have to be fully discharged after the machine pops off to prevent going into a reset cycle.

Ideas?

-drasnor

MrBill

SM10 - 314 POINTS/DAY AVG

Now that we're somewhat settled since the move in April, I hope to get the real SM10 out of the box and get it folding again. I've had 3 comps folding in its place until I can get it set up. The PPD average for SM10 will go way down when it's back to fending for itself.

Garg

drasnor wrote:

No, it's not overclocked at all and I doubt it would overclock very well being a locked Palomino. I'm holding off on the overclocking until I figure out a way to get the machine stable. I don't know why, but my intuition says it's a motherboard issue.

Yeah, I suppose it'd only overclock another 66-100 mhz or so then. 1720mhz is the highest I've ever had a Palomino stable, and then they're pretty sensitive to temperature. 1650 is a safe spot for the Palomino in IC_11 (which is unlocked). Another 50mhz or so isn't worth stability issues.

dragonV8

Just to add some info to this thread........

When we built SM27, it came with 2x256M sticks. However after Stanford changed the minimum ram for QMD's to 1 Gig, Sally added another 2x256M sticks she had stashed away.

SM27 has been running with 4x256MB since, working it's butt off.

muddocktor

Hey Jon, Stanford has backed the memory requirements back down to 512 MB for QMD work again. So you and Sally can build 1 more Intel QMD cruncher with the extra ram.

Crypto

Two points from me:

I'm still waiting for IC7 to be delivered. I guess it's in the post somewhere. I have a nice new case and PSU all waiting.

overclocking: I seem to remember for folding that it's better to increase the multiplyer rather than trying to get the frequency up? This is not the best form of overclocking for general work but good for folding?
Advice please.

Cheers

Crypto

Linc

Crypto wrote:

I'm still waiting for IC7 to be delivered. I guess it's in the post somewhere.

I just got the contract via mail within the last day or two. I'm sure it will be out the door soon. Takes a little while to move stuff across the pond it seems.

mmonnin

A 1900+ is one of the better CPUs compared to a lot of the other systems that were handed out over the past several years. Most of the CPUs are just old, almost forgotten extras people have laying around so they are hardly ever top of the line. We use what we can get.:)

shwaip

drasnor wrote:

I just had a thought about trying to build a serial port watchdog; set up a cron job that generates serial port activity every hour resetting a solid state timer or purging a microcontroller register. If the timer reaches some voltage or counter value the circuit closes the reset switch. Doing it with a microcontroller wouldn't be so bad but my basic stamp is a bit on the large side and my HC05's require support circuits (clock generator et al) and except for one EPROM are all OTP. Solid state would be a lot cheaper but it would almost certainly involve some sort of RC timing circuit and the capacitor would have to be fully discharged after the machine pops off to prevent going into a reset cycle.

Ideas?

-drasnor

You should be able to do it with a PIC. They have an internal R-C oscillator that you can use if you don't want to hook up some sort of crystal for generating the clock. You can use one of the (16-bit) timers to count up. When it recieves a signal from the serial port, you can reset the timer. If the timer ever wraps from (2^16 - 1) to zero, you could have it toggle a relay or something. PICs are very cheap, as well. The only disadvantage is that you'll have to have a circuit to convert from serial voltage levels to 0-5v levels. Not too hard though.

Crypto

General Keebler wrote:

I just got the contract via mail within the last day or two. I'm sure it will be out the door soon. Takes a little while to move stuff across the pond it seems.

Can't wait

drasnor

shwaip wrote:

You should be able to do it with a PIC. They have an internal R-C oscillator that you can use if you don't want to hook up some sort of crystal for generating the clock. You can use one of the (16-bit) timers to count up. When it recieves a signal from the serial port, you can reset the timer. If the timer ever wraps from (2^16 - 1) to zero, you could have it toggle a relay or something. PICs are very cheap, as well. The only disadvantage is that you'll have to have a circuit to convert from serial voltage levels to 0-5v levels. Not too hard though.

The tough part is I don't have any PICs, hardware to program PICs, or software to program PICs with. I could probably build the hardware, download the software (iirc Microchip offers free C compiler?) but getting PICs means ordering from Digikey and they like to ream on unit price and shipping. I think I can do this with a couple op amps, I just need to find my big book o' op amp circuits to see where I can start.

-drasnor

muddocktor

Crypto, the mobo they are sending you has no provision to overclock with either multiplier or fsb speed stock. There is a modded bios out there that will let you overclock the fsb speed a bit though, but you are out of luck on overclocking by multiplier unless you pinmod a higher multi.

Crypto

muddocktor wrote:

Crypto, the mobo they are sending you has no provision to overclock with either multiplier or fsb speed stock. There is a modded bios out there that will let you overclock the fsb speed a bit though, but you are out of luck on overclocking by multiplier unless you pinmod a higher multi.

Thanks for that info Mud, to be honest, that's somewhat of a relief!

Overclocking my existing rig was fraught with problems, and overall, I'm not convinced that the small gains made paid for the downtime due to the frequent crashes.

I guess I'm not too adventurous, steady as you go is my motto.

I'll be a steady contributor to the cause, I won't let you folks down.

Cheers

Crypto

drasnor

I think I found a good solution: Dallas Semiconductor DS275 line transceiver connected to PC serial port, 555 timer reset switch. The 555 is calibrated to count to a few hours whereupon if it hasn't been reset by traffic on PC serial port it drags the motherboard's reset pin low, rebooting the computer. The reset pin on the 555 restarts the timing cycle so I don't have to worry about discharging the timing capacitor manually. I'm going to try to build the whole thing inside a 9-pin D-sub hood. If it works, I'll post the schematic and building instructions and maybe make another $50 .

-drasnor

drasnor

Ok, I finished designing the circuit; I think this will work. I decided not to use the reset pin since there was such a nice event failure alarm schematic in my 555 circuits book that would be easy to adapt for this purpose . I'm waiting for my IC's to arrive in the mail to actually build it but I'll let y'all know how it goes.

On the other hand, SM25 is getting more and more unstable. It won't even stay up for a day now. Short of pulling it apart and testing each part individually I don't know what I can do for it.

-drasnor