Linux and Folding.

[mookydooky]

I gleaned the following from IdeaMagnate at oc.com:

"The client's core is written in FORTRAN. The reason why the Linux client is slower (which it is by about 15%) is that the FORTRAN compilers for Windows are much better than the ones for Linux. The good people at Stanford are dilligently looking for ways to improve the Linux client, and the version we have now is significantly more effecient than most previous versions were..."

For those of you considering building pure folding rigs, it seems Windows is the way to go. Furthermore if the machine is also a fairly old machine with very little RAM and HD space (compared to today's machines), it seems Win98SE is the choice of champions.

[Allan]

I've used Linux for folding, it works smoothly but it is indeed a bit slower.

You are right win98se is probably the best way to go about folding, but it seems XP is about the same when it comes to speed. (But it takes up a lot more HD space!)

[r0l0e]

hey all, my post refers to the conversation we were having at the alternative os forum.

i don't have a basis for my argument but i'm folding on both my xp and linux box both are workstations with minimal services running, i fealt that linux was folding faster. is there a way to know which is the faster of the two?

-r

[mookydooky]

Originally posted by r0l0e
hey all, my post refers to the conversation we were having at the alternative os forum.

i don't have a basis for my argument but i'm folding on both my xp and linux box both are workstations with minimal services running, i fealt that linux was folding faster. is there a way to know which is the faster of the two?

-r

Which OS are you running on the 1576 AXP? Mine's at 1540 and it finishes a frame of the ProteinA series in a little less than 9 (8.82 according to Fold Monitor, not the log which gives faster times) minutes when I'm not on the machine. I use WinXP, I also run a lot of background services. If you use Linux on the AXP, and if your per frame time is significantly faster than mine, then it would be reasonable to assume that Linux is faster. I'm not 100% sure myself, but the person who originally posted the above quote seemed to be pretty knowledgable about these kinds of matters.

[r0l0e]

Originally posted by mookydooky


Which OS are you running on the 1576 AXP? Mine's at 1540 and it finishes a frame of the ProteinA series in a little less than 9 (8.82 according to Fold Monitor, not the log which gives faster times) minutes when I'm not on the machine. I use WinXP, I also run a lot of background services. If you use Linux on the AXP, and if your per frame time is significantly faster than mine, then it would be reasonable to assume that Linux is faster. I'm not 100% sure myself, but the person who originally posted the above quote seemed to be pretty knowledgable about these kinds of matters.

i'm running winxp and linux on the 1576 A4 (not AXP), is there a rule of thumb on how to bench the frames? in linux, i finish a frame in 4-5 minute, but in winxp, 5-6 min. but i couldn't tell you which protein though.

if significant meant 3x faster then your frame time, there was no way for me to tell. i just notice it working from one os to the other.

-r

[mookydooky]

There is barely a difference in folding speed between a A4 and an AXP, something like less than 1%. Folding doesn't use any of the SSE instructions and the data prefetch apparently doesn't do much for folding computations. The times for frames depends on the particular protein you are working on. As far as I know, the ProteinA series takes the longest per frame of the 100 frame proteins. Some of the new Peptides take an hour, but they have only 10 frames. As I said above, if I don't use my machine, it takes about 8.82 minutes (according to Fold Monitor which is differnent than the fahlog) for a frame of ProteinA. I don't have Linux on any of my machines so I can't be sure. To benchmark the speed of your folding, you have to benchmark with the same protein series because different series take different amounts of time. Try to see if you get a ProteinA and see how long it takes for a frame in Linux.

[r0l0e]

if that's the case... when i first installed the console on linux, the first protein i was working on was the one with 100 frames. i just can't remember what protein it was. in any case, i completed a new frame every 4-6 minutes, but mostly completed each frame in 5 minutes while browsing, note that this in linux.

so i can say at 5min./frame, to do 100 woould take 500min. or 8.333hrs. in linux.

when working on a protein in linux, will the console notify me as to which protein i'm working on? or do i just consider that if it takes 100 frames to finish, it's automaticaly a ProteinA?

[mookydooky]

There's a lot of different proteins. Most of them have 100 frames. It's only recently with the Peptides that we're seeing proteins with 10. Here's a link to a list of proteins we've seen so far.

Protein Scoring.

If you look, there is a great variety in scoring. Basically, proteins with the same number of points will take about the same amount of time. Unless they're of the same protein series, ala ProteinA (ProteinA, ProteinAnat, ProteinAg29aNat, ProteinAg29a, and a few others), even proteins with the same number of points will give different times. So far, the ProteinA series is the largest, and also the only 5 pointer, protein we've seen. To see which protein you are working on when using the console, you can scroll up to where the work begins, you'll see something like this:

Protein: NativeBBA5-300K
- Project: 3, (Clone 88, Gen 19)
- Frames Completed: 0, Remaining: 100
- Dynamic steps required: 500000

This is for a NativeBBA5-300K. On an AXP 1.6Ghz, it takes about 1.46 minutes a frame. If you look at the NativeBBA protein series on the list, you'll see that it's worth only .6 points, which makes it a very small protein. A ProteinA is worth 5 points which makes it about 8 times larger.

[r0l0e]

i'm starting to see a the bigger picture now. thanks mooky. i'm migrating my webmail server at the moment but when i get some down time, i'll do some benching on my A4 1576/linux and the A4 1.44/winxp. there both a little differnet but i don't expect a big difference in frame time.