[gsrcrxsi]

ok so i just started folding. it seems to be running fine, been running since last night.

system:
q6600 @ 3.0 GHz
2 GB DDR3 1066
ubuntu 7.10 32 bit

currently i just run 4 instances of fah to use my full system capabilities, since i cant run SMP on 32-bit linux. would it be any faster to run 2x 64- bit under a VM?

also, is there anything i can do to get smaller WU's? (so i can process more of them) when i started last night i was getting ones needing about 50000 steps, letting me do about 1 WU/hr (4 WUs/4hrs) but when i check it this morning 2 of them are running on 50000 step WUs and and the other 2 are working on some massive 4000000 step WUs. which based on my calcs will take like 2 days to complete

also where can i look at my progress? like ppd, # WUs completed, etc? my stuff just runs in the 4 terminals

[harlam357]

Originally Posted by gsrcrxsi

currently i just run 4 instances of fah to use my full system capabilities, since i cant run SMP on 32-bit linux. would it be any faster to run 2x 64- bit under a VM?

also, is there anything i can do to get smaller WU's? (so i can process more of them) when i started last night i was getting ones needing about 50000 steps, letting me do about 1 WU/hr (4 WUs/4hrs) but when i check it this morning 2 of them are running on 50000 step WUs and and the other 2 are working on some massive 4000000 step WUs. which based on my calcs will take like 2 days to complete

also where can i look at my progress? like ppd, # WUs completed, etc? my stuff just runs in the 4 terminals

2x64-bit under VMWare?... yes, that's the ppd goldmine, you can make 4000+ppd from that quad. There's a sticky here on how to do that sort of setup, it's based on WinXP but you should be able to adapt the process to Linux.

Smaller units?... probably not, the only thing you can do is set your client to receive normal sized WUs. You can set this option when running fah -configonly (need to answer "yes" when it asks for setting "advanced options" so you can get to that setting). Typically a standard WU can take anywhere from a few hours to weeks depending on the speed of the machine. I have a PIII 1.2GHz box here at work that I do release builds on and it takes about a week to crunch the current breed of WU.

Stats?... here's your EOC stats page: http://folding.extremeoverclocking.c...hp?s=&u=327246

[gsrcrxsi]

when i run fah it says command not found. ?? ive been starting FAH by running ./FAH504-Linux.exe

and what is a "normal" sized WU? > or <5MB? ive been doing "yes" to >5MB because thats what everyone says you have to do with team 32

but what i was asking is if running it in VMware rather than the 4x FAH that im doing now will run any faster?

[WarriorII]

Ok, I'll ask since you didn't say & we all don't want to "ass_u_me".

What Quad are you running? An Intel Quad?

Can you change your FF 7.10 over to the 64 bit version easy enuff?
I'm nt sure how wrapped up in it you are?

Then DL the Linux SMP client, upon advanced set up, choose no big packets I think it is.
or you can just run with the non-smp client, but that won't make use of ALL your cores.
-only 1.

[WarriorII]

Command should look something like this:
typethis:

CD folding/FAH
./FAH504-Linux.exe -forceasm -verbosity 9

-I think it should start running after that if you installed it into its default location

[gsrcrxsi]

i did say. its the q6600. intel quad, stock @ 2.4 GHz, i OCed it to 3.0 GHz (333x9). im not sure what would be involved in stepping up to the 64 bit version. probably have to wipe everything and start over, which i really dont want to do. this machine also doubles as my mythtv backend. it would take a good while to reset it back up how i want it.

i know the non SMP client only stresses 1 core, thats why i run 4 of them simultaneously. i was just wondering if 2x64-bit under VMware with SMP clients will work any faster.

but whats weird is that i seem to get bigger packets when i choose no big packets than with choosing yes to big packets.

[harlam357]

Originally Posted by gsrcrxsi

when i run fah it says command not found. ?? ive been starting FAH by running ./FAH504-Linux.exe

and what is a "normal" sized WU? > or <5MB? ive been doing "yes" to >5MB because thats what everyone says you have to do with team 32

but what i was asking is if running it in VMware rather than the 4x FAH that im doing now will run any faster?

Yeah, that's it... ./FAH504-Linux. Sorry I just use the command fah as a way to say, the F@H client executable.

Basically if you set the client to allow it to accept > 5MB WUs then you have a chance to get bonus WUs which will be worth a lot more points and make better ppd. You may run into some memory bandwidth issues though if you happened to get all bonus WUs as they are more memory intensive. Best option is to set two clients to accept > 5MB WUs and two clients as "normal".

Of course, this is all in lieu of running 2xVM 64-bit Linux with the SMP client... again, that's the ppd goldmine. And the reason you need to do 2xVMs is because (as you may be aware) the free VMWare server version only allows the guest OS 2 cores, so you need 2xVMs to max out the quad.

[gsrcrxsi]

yea i know about the VM only getting 2 cores thing. really points arent my concern. WUs/day is. so that why i was wondering if 4xnon-SMP under 32-bit linux will be faster/slower/same as 2xSMP under 64-bit linux in VM. im just trying to see if its worth it to download the 64-bit ISO and do it.

[Adak]

SMP folding has averaged about twice as many WU's per month, compared to non SMP folding. All this depends very much on the specific WU's you are folding, of course.

but whats weird is that i seem to get bigger packets when i choose no big packets than with choosing yes to big packets.

Yes, well - we have a server called "Wrong Way Corrigan", that occasionally works backwards when the human tells it to do so, by mistake.

[gsrcrxsi]

well assuming the same WU's, youre saying that running 2 virtual machines (with 2 cores each) with SMP clients will run faster than my native linux running 4 simultaneous non-SMP clients?

[Adak]

Originally Posted by gsrcrxsi

well assuming the same WU's, youre saying that running 2 virtual machines (with 2 cores each) with SMP clients will run faster than my native linux running 4 simultaneous non-SMP clients?

The WU's are not the same. The data type they use is not the same. SMP is always faster, by something slightly more than twice as much, in my experience.

So, yes. Your Quad on 2 cores, should run similarly to my E6600 at stock speed of 2.4GHz iirc. It puts out 1 SMP WU every 22 - 24 hours, for 1760 points, day in and day out. (2653 project WU).

I'm not that sure what native Linux did on non-SMP WU's. What I heard repeated was that it was slower than WindowsXP on those WU's. On SMP WU's, Linux is significantly faster, because of a shortcoming in the Windows architecture.

That last part I can personally swear to, since I've run both on the same rig.

Folding should be comfortable though. That's the first rule. If you don't feel like that's what you want to do, then please, don't do that.

We want to see you smiling, buddy!

[gsrcrxsi]

maybe im wrong on what counts as a WU? i started last night and running 4 simultaneous FAH instances (using all 4 cores 100%) i did about 1 WU/hour, really 4WUs/4hrs cause it takes 1 about 4 hours to complete but it does 4 of them at a time (im counting 1 WU as 0-100% of the 50000 steps. it said "work unit complete" and started over at 0%, thats a whole WU right? running all night, my stats page says i did like 10 work units.

maybe im not clear? my computer is running 4 non SMP clients at the same time. its stressing all 4 cores to 100%. i fully understand that SMP vs, a single non-SMP, SMP would be much faster on a c2d or c2q which has more processing resources available, but running 4 at the same time in using every bit of available processing power that i have. and thats where im wondering if SMP will still have an advantage over what im currently doing.

[harlam357]

One WU is finishing a simulation from 0-100%. Since you are running 4 clients and 4 simulations simultaneously, you are in the middle of working on 4 WUs (work units).

Make sense?

[gsrcrxsi]

yea i know, i just wanted to make it clear to everyone. it didnt seem to be getting through that im doing 4 nonSMPs at the same time. im trying to figure out if it works faster running 2x SMP in VM. i guess no one has tried both? im downloading the 64 bit ubuntu now and ill load them up in the VM and see which is faster

screen shot of what im currently doing

[WarriorII]

I think ChaR said at one time that you should only run 1- client of SMP not 2 on a Quad.

I don't remember the circumstances though. He plods new territory a lot.

[gsrcrxsi]

you need to run 2 (1 in each virtual machine) because the free version of VMware server only lets you give a max of 2 cores per virtual machine

[ChasR]

Run two (one in each VM) in VMware, but only one in native Linux, unless you have a rig that's not assigned to the "quad only" server. I don't have any C2Qs running native Windows SMP, but others have reported a good ppd boost running two SMP instances when using Affinity Changer. Without AC, two won't help.

[gsrcrxsi]

Originally Posted by ChasR

Run two (one in each VM) in VMware, but only one in native Linux, unless you have a rig that's not assigned to the "quad only" server. I don't have any C2Qs running native Windows SMP, but others have reported a good ppd boost running two SMP instances when using Affinity Changer. Without AC, two won't help.

i would if i had 64-bit linux. but i have 32. see my dilemma? 4xnonSMP runs great. im just trying to figure out if the VMware option will run faster, or do more WUs/day or /hr however you want to break it down

[ChasR]

I don't see your dilemma. Run two 64 bit VMs with 32 bit Linux as host. 4000+ ppd (SMP) vs 1000 ppd or so (uni cpu). BTW, nobody really counts WUs. The competition is about the points.

[gsrcrxsi]

but when i run 4 at the same time, it uses all cores. not just one. thats what im saying. so it will do 4 WUs simultaneously using one core EACH. so instead of running 1 WU using 4 cores, i run 4 WU's at the same time. what im asking is if i will get better performance with the VM method than my current method. my brain tells me that in a perfect world they should be identical.

ideal scenario (which i cant do)

native 64 bit linux with SMP. should run 4 times as fast, 1 WU at a time.

next:

native 32 bit linux. 2x 64bit linux (2 cores each) with SMP. should run twice as fast, 2 WU's at a time

last:

native 32 bit linux. 4x nonSMP clients. should run at 1x speed, 4 WU's at a time

make sense now? i feel like im banging my head into a wall here, lol. can someone explain to me why either of these scenarios would run faster than the other?

edit: oh and i upped my OC to 3.2 GHz (355x9). when i get all settled in ill be even better