Adding Another Rig Next Week

[Nemo]

Well, I finally pulled the trigger, and ordered some new folding kit.  I've decided to give Intel a try on a dedicated folding sled.

 

From what I could tell, folding is hungry for cache.  Loves the stuff, in fact.  So I got a 3GHz P4 Prescott with 2MB L2 cache.  No plans to overclock initially, but of course we have all promised that at one time or another. 

 

My informal benchmarks so far put my 1200MHz Pentium M (1MB L2 cache) at almost exactly 1/2 the speed of my Athlon 64 3000+ @ 2.4GHz (512K L2).  So I'm expecting this new P4 system to really scream along nicely.

 

Anyone have good (or bad...or any...) experiences with the P4 Prescott in S775?

 

Sqopt

577960[/snapback]

For the most part, processor speed trumps cache size in folding. The only projects I've seen the perform better with a larger cache are the 1477-1481 work units. For some reason, those projects do well on larger cache processors. With that P4 you have the best of both worlds. I'd sugeest configuring your folding setup to accept big work units > 5 MB in size and use the -advmethods flag. As long as you have a good bit of RAM you should get QMD work units which give me around 400 ppd on my P4c.

[BabyBalrog]

For QMD's you need MORE then 512 RAM, nto exactly a budget box. But if you turn it into a join file server which only gets acsessed periodically then you may get better useage.

 

I remember that the absolute best ppd/$ CPU was a dual Xeon. OC-AMD did the caculations over a the official Foldign web forum.

[Sqopt]

For the most part, processor speed trumps cache size in folding.  The only projects I've seen the perform better with a larger cache are the 1477-1481 work units.  For some reason, those projects do well on larger cache processors.

 

My experience on the 1477-1481 WUs is, I believe, consistent with that observation. My A64 @2.4GHz/512 cache pumps out a few ppd less than a 1.8GHz Pentium M/2MB cache on these WUs.

 

How much of the difference is an amd-vs-intel-on-gromacs thing, and how much is a faster-on-larger-cache thing is anybody's guess. I suspect a little of both.

 

 

  With that P4 you have the best of both worlds.  I'd sugeest configuring your folding setup to accept big work units > 5 MB in size and use the -advmethods flag.  As long as you have a good bit of RAM you should get QMD work units which give me around 400 ppd on my P4c.

580730[/snapback]

 

Interestingly enough, I've never seen a QMD WU on my Athlon system with a gig of RAM. Theoretically I should have been able to get one, since I run Linux, right?

 

I'll probably run 512 RAM on the P4 at least initially. I'll worry about whether I get QMDs after I get the thing to a stable condition.

 

Can't wait to run some head-to-head benchmarks to see how it all plays out. That's part of the fun!

 

Sqopt

[oblivescence]

qmd's are strictly intel w/ windows. atleast for the moment. atleast thats what their website says.

 

http://folding.stanford.edu/QMD.html

 

[alown]

i say, the moblie XPs are reallly good. they got big cache, use less power, and overclock good.