New to Folding - Need Advice (Optimal Setup for Dual Core + GPU Folding)

[tgxiii]

Hi everyone,

I'm newcomer to Folding with a lot of questions, and I was hoping you guys could help me out. I'll start with one for now. :)

I have a computer with an Intel E6550 and a 8600 GTS video card. What would be the optimal setup for this kind of rig? I've read enough about Folding to know I should run the GPU client, so I'm thinking one of the following options:

1. 1 GPU client + 2 CPU clients
2. 1 GPU client + 1 CPU client
3. 1 GPU client (no CPU clients)

Also, if you could provide a short explanation as to why it's the optimal setup just so I could gain a better understanding, I would greatly appreciate it. :)

[Adak]

Hi everyone,

I'm newcomer to Folding with a lot of questions, and I was hoping you guys could help me out. I'll start with one for now. :)

I have a computer with an Intel E6550 and a 8600 GTS video card. What would be the optimal setup for this kind of rig? I've read enough about Folding to know I should run the GPU client, so I'm thinking one of the following options:

1. 1 GPU client + 2 CPU clients
2. 1 GPU client + 1 CPU client
3. 1 GPU client (no CPU clients)

Also, if you could provide a short explanation as to why it's the optimal setup just so I could gain a better understanding, I would greatly appreciate it. :)

1 GPU client + 2 cpu clients allows the GPU to romp, and the cpu to use all the spare cycles. Note that this will make your system somewhat less responsive than 1GPU + 1 cpu client, would be.

But you should test this yourself and see which combo you prefer. Get FahMon to help:

http://fahmon.net/

(it's free)

show you what your current points per day are. Set up the preferences so only the last three "frames" are used to make the calculations.

You never know, you might find that 1 GPU + 1 SMP client, is actually the best combination for your C2D! :)

The reason is that the GPU client doesn't use much of the cpu core it is working with.

Oh! Before I forget - :welcome: to the team, and our team's forum! :clap: :clap:

[Jolly-Swagman]

What OS are you using tgxiii,,,, and welcome to the team

[Jason Kitchens]

welcome to folding tg

[dfonda]

I would do #3...run that client for awhile. Its simple and fast.Then move ahead with what ADAK talks about.

Once you become addicted....:)

:welcome:to the Fold

[tgxiii]

Thanks everyone.

To respond with some of your comments:

I ran a 2 CPU / 1 GPU setup for a couple of days and saw that my GPU ppd was around 400 (forgot the actual number and unfortunately no screenshot). This is with "Calculate PPD based on All frames" setting when I checked. This sounds low... is this typical from this type of setup with an 8600 GTS?

As for my OS, I'm using Windows XP SP2.

[dfonda]

Very low, concentrate on the GPU client first since it does so much more.

It will just be a matter of prioritizing your cores , so the GPU can get its fair share of CPU.

There are at least 2 threads on the first page here on this subject.

http://www.ocforums.com/showthread.php?t=569466

[Adak]

Tgxiii, I agree with Dfonda, the GPU client should be maximized, and should be the client you start with. Get a frame of reference for it's production, all by itself.

Then introduce your other clients, one at a time, and let it fold for awhile. Check your ppd from FahMon and DO set it for only the last 3 frames or it's no good for this test.

When your ppd begins to get lower, or your system responsiveness becomes too low for comfort, then remove one (non-GPU) client.

That should be your ideal FAH client mix. No one can give you better info on this than your test, with your system, and FahMon.

FahMon MUST be set for ONLY the last 3 frames, or your test will not be valid, and you must let your test clients all run for at least 3 frames worth of time.

[tgxiii]

Thanks for the advice. I'll be sure to keep that in mind.

[LandShark]

with WinXP, I would say 1 GPU + 1 CPU (NOT SMP!) client. as in WinXP, the GPU will use a full core by itself, and that left one 1 core to fold something else.

[tgxiii]

Good to know. Thanks!

[harlam357]

:welcome: to Team 32 tgxiii! New guy stocks the fridge. :beer:

I see you're already off and running with the excellent advice of our membership... so I'll just say, we're glad to have you here. :)

[pejsaboy]

One thing I just started, and seems to be working well- give the GPU the 'slightly higher' priority, set the SMP to idle, and config it to only use 40% cpu. That 40% is about what was left after the GPU took its share, and FAHMon is estimating the full ~5k from my 8800gt, and ~980 ppd from the SMP client. I think the final numbers are going to be more like 5-600 from the SMP client, but as long as the GPU is getting its full share, it's just icing on the cake :)

[LandShark]

One thing I just started, and seems to be working well- give the GPU the 'slightly higher' priority, set the SMP to idle, and config it to only use 40% cpu. That 40% is about what was left after the GPU took its share, and FAHMon is estimating the full ~5k from my 8800gt, and ~980 ppd from the SMP client. I think the final numbers are going to be more like 5-600 from the SMP client, but as long as the GPU is getting its full share, it's just icing on the cake :)

but do watch out on the SMP client to make sure it'll meet the deadline!! at 500-600ppd on SMP WU (depends on which project), it might be very close!!

[pejsaboy]

That's good advice... I'm actually still tinkering with the SMP client. After leveling out for a while, even 40% seems to be a bit high. Trying 30% now, but I'm leaning toward the idea that I'll just be shutting the SMP back off altogether. Both are just too much for a dual-core, at least a low-end one.

[Shiggity]

Hope you have fun with T32 ;)

:welcome:

[pejsaboy]

Just as I suspected, the SMP client has been retired altogether :(