Dual Core Folding - Maximum Performance

Leonardo

This thread is an edited copy of the thread here. As it started centering on Folding efficiency, I though I'd start a new thread on dual core and hyperthreading for Folding.



Finally got all the parts for my new budget Folding box. Specifications:

Intel Pentium D 820 (Dual Core, 2.8GHz/core, 800MHz FSB, 2MB L2 Cache)
MSI 945P Neo-F motherboard
1 GB (2X512MB) Gigaram DDR2 667 (value RAM)
ATI Rage 8MB PCI video card (it does 1280X1024, 32bit color!)
Robanton 600W PSU, modified
Chieftec (Antec 1030) case, modified
Zalman CNPS 7000-AlCu HSF
WD800JB HDD

The processor came from an eBay auction, the motherboard was Newegg "refurbished", the RAM was Newegg new. Video card was on sale at CompUSA. All the other components came from my parts bin.

I was a bit concerned about the CPU, in that the seller wasn't open that it was an engineering sample. To his credit though, he posted the etched ID in his add. Had I researched better I would have known. Not to worry, the CPU and MSI BIOS played happily together the first time I fired up the partially built system - memory, CPU w/HSF, and mobo outside the case. I was also concerned about the Robantan PSU. It's a decent unit, but it had given my poor 12V lines on my overclocked P4 systems (No. 1 and 3 in signature). I don't understand it, but it has a much higher 12v line than ever before. I haven't played around in the BIOS yet. I don't know if this engineering sample CPU has the multipliers locked or not.

lemonlime

Nice work Leo! Interesting to see that CPU-Z recognizes it as an ES chip as well. That should make a monster folding rig .. I'm not too familiar with intel dual cores, do they support HyperThreading on each core independantly for four logical cpus? I have a single 2.8 pressy at work that really chews through WUs with HT on.

Bud

not the 820's dont feature hyperthreading. mine folds okay but those ht p4 keep up though on the folding side. I do think if ur do more intense thinks and the same time its easier though

muddocktor

It's looking good, Leo!

The 820 doesn't support HT, but that's not really a problem with QMD's anyways. The QMD wu's use so much memory bandwidth that HT doesn't actually give you much, if any, boost in production. However, with 2 cores, that 820 should supply a much more substantial boost in production over a HT enabled single core P4. They are still somewhat memory bandwidth handicapped, but 2 separate cores will help out on production.

Another scenario to explore Leo, is to try folding 1 QMD and 1 normal Gro wu and see if your points production is actually higher than with 2 QMD's. A guy over at the overclockers team says that this is the hot ticket for a dual proc Xeon setup, which should be kind of comparable to a DC P4 setup like yours.

Leonardo

Another scenario to explore Leo, is to try folding 1 QMD and 1 normal Gro

How do I ask for assignments? I thought the "-advmetheds" was no longer viable. Please, do instruct me on this.

Out of curiosity, is there a reason besides maybe noise you use foil tape on all the screws, etc?

The tape merely secures the fan grills. The case came stock with the punched-out steel plate grill, which is highly airflow restrictive. I've just never bothered drilling holes for screws. The only time I see the back of the case anyway, is when I'm working on it.

Gonna being playing cards with the family this afternoon. If time permits later, I may get on with FSB and mulitplier tweaking. Well see if/how much this is possible and if the Zalman can keep up!

muddocktor

For QMD's, set "bigpackets=yes" in the client.cfg file and then start the client with the -advmethods switch and you are almost guarranteed to draw a QMD wu. To set up a client to draw regular Gromacs wu's (non-bigWU), set the client.cfg file to "bigpackets=no" and start it with no flag. It also might draw a Tinker too, configured like the second example.

Donut

It will on occasion draw a tinker wu, I've got mine set up similar to what muddocktor posted. (2 for QMD's, 2 for reg. units. on a dual Xeon HT) 3 QMD's kill this rig as far as memory bandwith.

Leonardo

muddocktor wrote:

Another scenario to explore Leo, is to try folding 1 QMD and 1 normal Gro wu and see if your points production is actually higher than with 2 QMD's. A guy over at the overclockers team says that this is the hot ticket for a dual proc Xeon setup, which should be kind of comparable to a DC P4 setup like yours.

Alright then, I set it up the two Folding instances just that way. Thanks for the advice. Now, another question: How do you monitor individual CPU/core/box totals when you have several? Sure, I've got Electron Microscope III running, but where can I look at point tallies, other than just keeping a manual log?

Leonardo

Mudd, you were right! With the two cores cranking now, one on a 450 point QMD, and the other on a 600 point Gromacs, Electron Microscope estimates 624 points per day! I should try this with my HT machines.

Donut

AFAIK Jon and Sally are running 2 QMD on at least 1 P4. (SM-27?)

Leonardo

This morning before leaving for work I checked production of the D820 rig. After a night of no activity on the machine other than Folding, production improved to a rate of about 725points/day!

Should I try the same combination with my hyperthreaded P4s: "bigpackets=yes" in the client.cfg file and then start the client with the -advmethods switch and you are almost guarranteed to draw a QMD wu. To set up a client to draw regular Gromacs wu's (non-bigWU), set the client.cfg file to "bigpackets=no" and start it with no flag.? One configuration for the physical CPU and a different configuration for the virtual CPU?