• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

SMP Open Beta

M

marty9876

[H]ard|DCer of the Month - February 2006
Joined
Jun 11, 2003
Messages
4,906
http://folding.stanford.edu/download.html

Go get it!

Short: Only runs on that OSX crap and x64 Nix. I loaded up Fedora x64 on a box and was really easy (me=nix moron) to get going.

Designed for a true Quad. Dual Xeon boxes w/ HT run with HT on.

8way boxes run 2x instances

~ 200 meg RAM used or there abouts

PPD was good, new core came out yesterday which kinda slows things down. I think the benchmark PPD was adjusted for this (zim confirm?).

Careful on the save points (they don't save progress). Not sure if fixed yet. Close the client and your SOL.
 
hrmm... osx and linux.. that makes it nearly useless..

I've got one smp linux box but why bother reinstalling? its a pIII 750 x 2... not worth the effort, i suppose.
 
Well, it looks like you need a quad box anyway, I just don't have the time to take down a one of my 4/8/16 way boxes and load up linux just to test. Can't wait for a windows version though, Marty, pull some strings ;) or whatever you normally pull.
 
Will run on 2 way, decent PPD

Mage/Ben have been looking into qemu, lost cause I think.

Crap, I had another emulator to try and forgot the name of it.....
 
Qemu set up with 4 cores on my x2@2.5 at home took like 4 hours to compile a damn kernel, don't really think it's worth messing with any more.
 
Awesome timing! I've got a couple of Opteron 270s heading my way.... been looking forward to doing some SiMPy folding.
 
Too bad they haven't come out with a 32-bit version yet :( I have a Core Duo (not 2) that I could run this on, but they aren't 64 bit.

Yeah, qemu could work, but it's so slow that it's not really worth running it. In terms of getting any useful data out of it, it'd take a long, long time.
 
It's nice to see that all these clients are finally starting to be released.

 
Scorpionjwp said:
Hey where is the windows version, I sure would like to put it on my quad p3 xeon.
Click to expand...


its not out yet... they are working on it now.. I don't know how long it will take.
 
What's the advantage of a SMP client vs. running multiple clients? Is a SMP client able to do calculations on larger numbers of atoms in the protein? I realize there's incentive for Stanford to make a client that will (eventually) work on every platform, automatically scaling to the number of processors, but do they get more/better science done with a 2-CPU SMP client vs. 2 single-CPU clients?
 
Mohonri said:
What's the advantage of a SMP client vs. running multiple clients? Is a SMP client able to do calculations on larger numbers of atoms in the protein? I realize there's incentive for Stanford to make a client that will (eventually) work on every platform, automatically scaling to the number of processors, but do they get more/better science done with a 2-CPU SMP client vs. 2 single-CPU clients?
Click to expand...

An SMP client would mean you wouldn't have to run multiple clients anymore. This means lower RAM usage as you would only be working on one protein at a time. I haven't looked at any of it yet, but I would guess it would also be able to use Hyperthreading. This would put to rest the argument of running one or two clients on Hyperthreading boxen.

No, an SMP client would not likely be 100% efficient, but once all the bugs are worked out, it would probably give similar PPD as running two clients. Depending on the protein and stuff like that, there may be a chance it would be able to do better than two clients on a dually box.

Basically, there were a lot of problems with QMD's on multiprocessor boxen. The best PPD you would think would be one of them for each processor. However, the memory bus was oversaturated with running more than one instance. In this case, I believe it would take care of these problems. Also, it would allow Stanford to move to bigger proteins that take longer to crunch normally. Sometimes too long to be worth putting out due to time constraints. With more than one processor being able to run one protein, it would dramatically cut down on the time needed to process the protein.

Overall, this is something long overdue and something we have been asking for for a long time. It should simplify maintenance for some of us that are currently required to run more than one client on a machine due to multiple processors. It's a lot easier to manage one client rather than two or four or eight depending on the box.

By the time I end up with a quad core processor, I would probably have plenty of RAM to run multiple instances without too much trouble. This should take care of that problem altogether since I would only need to run one client instead of four.

 
The main thing is that they can get calculations back even faster. Up to 4 times faster if running on quad core. If they don't need the calculations back really fast they can give out work units that are much larger that will still complete in a good time length. The SMP page gives a pretty thorough overview of why they made this client.

 
SmokeRngs said:
I haven't looked at any of it yet, but I would guess it would also be able to use Hyperthreading. This would put to rest the argument of running one or two clients on Hyperthreading boxen.
Click to expand...

This should not be used on a single core CPU with HT. That is still just one core and defeats the purpose of being able to get more complicated caculations completed and returned quickly. If it is a dual CPU box with HT or even a dual core with HT, then that is fine. I'm hoping that they have systems with single core HT CPUs blocked from being able to get SMP units.

Also, this is a quote regarding 2 and 4 CPU boxes with HT.
Vijay Pande said:
On a system with 2 physical processors, HT(which makes the OS thinks the machine has 4 processors) will help a bit and we recommend it. For something with 4 physical processors, adding HT (making it look like 8 processors) will hurt as half of theose processors will be really slow (HT "processors" are not really processors) and we don't recommend it.
Click to expand...
 
Originally Posted by Vijay Pande
On a system with 2 physical processors, HT(which makes the OS thinks the machine has 4 processors) will help a bit and we recommend it. For something with 4 physical processors, adding HT (making it look like 8 processors) will hurt as half of theose processors will be really slow (HT "processors" are not really processors) and we don't recommend it.
Click to expand...

Now that doesn't makes any sense, on one hand he's saying it's good and on the other he is saying it is not good.

WTF? make up your damn brain.
 
Jon855 said:
Now that doesn't makes any sense, on one hand he's saying it's good and on the other he is saying it is not good.

WTF? make up your damn brain.
Click to expand...

Makes perfect sense to me. What part don't you understand and I'll try to explain it.
 
Jon855 said:
Now that doesn't makes any sense, on one hand he's saying it's good and on the other he is saying it is not good.

WTF? make up your damn brain.
Click to expand...
Could it be related to memory bandwidth?
 
drizzt81 said:
Could it be related to memory bandwidth?
Click to expand...

No, it's related to the fact that the "HT" cpu is not a real core.

The SMP client is set to scale to 4 CPUs right now....this is a beta client. With 2 CPUs and HT on, that makes the FAH run slightly better than if if were running on just the 2 actual cores. If you have 4 actual cores, you are better off to run the smp fah client on the 4 actual cores to get the work done 2X faster than running on 2 real cores and 2 "fake" HT cores.
 
Killer[MoB] said:
Makes perfect sense to me. What part don't you understand and I'll try to explain it.
Click to expand...
Alrighty, here I go...

On a system with 2 physical processors, HT(which makes the OS thinks the machine has 4 processors) will help a bit and we recommend it. For something with 4 physical processors, adding HT (making it look like 8 processors) will hurt as half of theose processors will be really slow (HT "processors" are not really processors) and we don't recommend it.

In both case, they are using HT, one has his recommendation and the other doesn't.
 
Jon, check out the post before yours, it hits it right on the head, you two were probably posting at the same time.
 
drizzt81 said:
Could it be related to memory bandwidth?
Click to expand...
Boy, I hope not... my quad core box is a p3 xeon. Memory bandwidth is like 450 MB/s :(

I guess I'll just have to buy another box :D

 
Jon855 said:
In both case, they are using HT, one has his recommendation and the other doesn't.
Click to expand...

It's all about getting the work done quickly. Running on actual cores is always faster than running on virtual cores. See my previous post. Hope this helps.
 
I didn't know that the SMP client were scaled to only 4 "CPUs". Now that I know, it now makes sense.
 
Not being a Mac user, but probably able to borg some new Macs,
I am curious about what hardware/OS a Mac needs to run the SMP client.

Anyone out there running the SMP on Macs? Tips?

Fold on!

 
How many cores do I need to run this? What types of CPUs?
In the beta test, we are strongly recommending that this code be run on 4-core boxes, although it can be run on 2-core boxes with reasonable performance. The code does best on Core 2 Duo/Woodcrest class chips and we recommend these systems (new iMacs, Mac Pro's, etc).
Click to expand...
http://folding.stanford.edu/FAQ-SMP.html
The FAQ has a lot of good info. The MACs are doing great with the Intel CPUs. OSX for Intel is the current supported OS for MAC.
 
Thanks. I'll check out the platforms. Since these are production machines, I may wait until after the Beta.

Fold on!

 
Celerator said:
Thanks. I'll check out the platforms. Since these are production machines, I may wait until after the Beta.

Fold on!
Click to expand...
Good idea since it is beta and anything can happen. Plus the current version will quit working in 2 months. There will be a new version out before then to ensure we can keep right on SMPing. ;)
 
Killer[MoB] said:
Good idea since it is beta and anything can happen. Plus the current version will quit working in 2 months. There will be a new version out before then to ensure we can keep right on SMPing. ;)
Click to expand...

yea, in case you get ran over by a green Volvo we can be sure the box is still folding two months form now.

;)
 
You must log in or register to reply here.
Back
Top