How should I set this up?

[Alpha736]

Okay, here's the deal, I'm not a real hardcore folder, but I wanted to start running it again on my main box, which I use for everything, (Gaming, web browsing, media, etc.)

Here's my current hardware setup:

CPU: Q6600 OC to 3.2Ghz - 8x400FSB
RAM: 4GB DDR2 800, CAS4
GPU: Radeon HD 4870 - Stock
Mobo: Asus Rampage Formula
OS: Vista Ultimate 64-Bit

How should I configure Folding@Home to get the best PPD, but also without it interfering with my gaming/usability too much? I don't mind if I lose only a few FPS in games. I was thinking SMP MPICH 64bit, and GPU2. I also might need to lower the OC on my CPU since it could become unstable with Folding@Home running constantly.

I also have an old Geforce 8600 GTS laying around, would it be possible to install that in my box and run it alongside the 4870 without it screwing anything up? I don't really have another PCI-E box that I could stick it in.

Thanks for the help!

 

[styx0r]

The GPU client you should run for sure. I recommend the command line client, because it is simple and easy to turn on/off. The Systray client will work as well.

You will probably end up turning off the GPU client for gaming / 3d rendering, but otherwise it won't have a noticeable effect on system performance. I do believe that the ATI client will use 100% of 1 core's resources... I'm not sure that this is still the case, because I only fold on Nvidia hardware.

The windows SMP client is a bit of a PITA, but the guides that are posted in the forum are a great place to start. You might want to try running a VM with a linux diskless folding client. This will only use two of the four available cores, leaving you with one core completely unloaded.

As far as mixing the 4570 and a 8800gs - I think you're out of luck. Generally mixing shader counts does bad things to both GPUs output. You could of course test the nvidia card anyway, after you've been folding on the ATI card for a while, and have a good idea of what sort of PPD it will give you.

 

[sduneman3]

Run both SMP and GPU2 clients.

The GPU client will automatically suspend when you start a game up. ATI, with the newest drivers, use around 3-4% of my Q6600 @ 3.6Ghz

I also agree that the SMP client is kind of a PITA to setup, but I wouldn't recommend a VM, because you can only use 2 cores, and would possibly have to shut it off during gaming, otherwise it will have a very noticable impact on your games.
The windows SMP client is pretty easily setup if you look at some of the guides, and will benefit by using all 4 cores, or having 2 SMP clients running to maximize your point output. And would have little to no effects on your gaming/office work

And mixing your HD4870 and 8600GTS on the same motherboard would not be possible due to driver conflicts.

 

[Zero82z]

Run the GPU2 client and an SMP client. I have a similar system, and that's what I do to maximise PPD. Further than that, you have two choices. You can either run the SMP client in Windows, in which case you would need to use a program like Bill2's Process Manager to lock the GPU client to one core and the SMP client to the other three (otherwise you will have CPU conflicts and one of your clients will get stifled - probably the GPU), or you can set up a Linux VM. The first option is easier, but the VM option will give you significantly better PPD - probably along the lines of 1.5 times what you would get with the Windows client. Even on two cores, it will fold faster than the Windows SMP client on three cores.

If you need help configuring anything, just post here and we'll help you out .

I wouldn't recommend a VM, because you can only use 2 cores, and would possibly have to shut it off during gaming, otherwise it will have a very noticable impact on your games.

Not true. I run an SMP client in a VM 24/7 and it has no performance impact on anything I do, gaming included. The only thing that does have an impact in some cases is the GPU client, in which case I just pause it (I'm using the tray client, which makes it a lot more convenient to do so).

 

[sduneman3]

Not true. I run an SMP client in a VM 24/7 and it has no performance impact on anything I do, gaming included. The only thing that does have an impact in some cases is the GPU client, in which case I just pause it (I'm using the tray client, which makes it a lot more convenient to do so).

Well, that's good to know. When I tried it, my FPS suffered due to VM ware.

 

[Alpha736]

Great replies guys, I really appreciate it.

By the way, I heard somewhere a while back that in 64bit I didn't have to worry about CPU load distribution when running both GPU2 and the SMP client. Is this true?

 

[Alpha736]

Well, I went ahead and installed the SMP MPICH and GPU2 Systray clients. I've attached a screenshot, let me know if it looks like I'm doing something wrong.

Thanks!

Thanks again!

 

[APOLLO]

Not true. I run an SMP client in a VM 24/7 and it has no performance impact on anything I do, gaming included. The only thing that does have an impact in some cases is the GPU client, in which case I just pause it (I'm using the tray client, which makes it a lot more convenient to do so).

I have only experienced slowdowns with the GPU client and VMware when I didn't isolate the GPU processes to a specific core. You mentioned using a process manager and I noticed that you did lock the GPU process, but it also needs isolation from what I've observed in my systems. Mind you, this may have different results on different OSes. I stick to XP for the most part.

Well, I went ahead and installed the SMP MPICH and GPU2 Systray clients. I've attached a screenshot, let me know if it looks like I'm doing something wrong.

What is the PPD you're seeing on the GPU client? Look to my reply above if you believe you're not seeing the full potential of your video card.

 

[Zero82z]

By the way, I heard somewhere a while back that in 64bit I didn't have to worry about CPU load distribution when running both GPU2 and the SMP client. Is this true?

That isn't true. With the ATI client, you really need to use a process manager if you're running more than one type of client. I have my GPU2 client locked to a single core, with my VM running LinSMP locked to the other three. That gives me the greatest PPD.

Well, I went ahead and installed the SMP MPICH and GPU2 Systray clients. I've attached a screenshot, let me know if it looks like I'm doing something wrong.

Thanks!
http://img24.imageshack.us/img24/3748/foldinghomeconfig21309nl7.jpg

Like I said just above, you need to use a process manager if you're running the GPU and SMP clients at the same time. The ATI GPU2 client needs a full core to itself, whereas the Windows SMP client won't be affected by running on only three cores (it really only needs two). Give Bill2's Process Manager a shot. If you need help setting it up, just ask, and I'll run you through it .

I have only experienced slowdowns with the GPU client and VMware when I didn't isolate the GPU processes to a specific core. You mentioned using a process manager and I noticed that you did lock the GPU process, but it also needs isolation from what I've observed in my systems. Mind you, this may have different results on different OSes. I stick to XP for the most part.

Yes, I lock my VM to use only the three cores that the GPU2 client is not using. The problem is not the GPU2 client using too many resources, but rather the SMP client stifling the GPU2 client. The GPU2 client must be core-locked as well though, since otherwise it'll try to use the other cores as well and it'll get slowed down.

 

[APOLLO]

Yes, I lock my VM to use only the three cores that the GPU2 client is not using. The problem is not the GPU2 client using too many resources, but rather the SMP client stifling the GPU2 client. The GPU2 client must be core-locked as well though, since otherwise it'll try to use the other cores as well and it'll get slowed down.

If you isolate the GPU process with an utility like SMP Seesaw, no other process will move onto that core (except a few unmovable system processes). Isolation is not exactly the same thing as locking. This feature both locks and prevents other processes from accessing the core(s) in question with a single action, so no problems with affinity drift should occur (barring any rogue apps that misbehave).

Although I have seen it, I never actually installed and used Bill2's Process Manager. It might be capable of isolation in its feature set. There is also Process Lasso and some other managers. The list of process managers is growing and someone should conduct a comprehensive comparison of all of them so that folders will know the pros and cons of each one.

 

[Zero82z]

If you isolate the GPU process with an utility like SMP Seesaw, no other process will move onto that core (except a few unmovable system processes). Isolation is not exactly the same thing as locking. This feature both locks and prevents other processes from accessing the core(s) in question with a single action, so no problems with affinity drift should occur (barring any rogue apps that misbehave).

The problem with that is that sometimes it is advantageous to have certain smaller processes being able to move onto the core, since it won't have any significant effect on folding performance. Also, when the client is paused (during a game for example), that would also prevent other processes from taking advantage of the extra available resources.

Whether or not that actually matters depends on your particular situation though.

 

[APOLLO]

Whether or not that actually matters depends on your particular situation though.

Very true. Most of my systems including the dual quads are dedicated folders. Where there is no GPU client I don't bother with affinity assignments since there's nothing to be gained. On my primary system where I perform most of my work, I don't play any games. In my case, core isolation works very well where it is applicable. My GPU clients and VMs always see optimum performance per hardware architecture.

 

[[ocau]sreg0r]

I find that you get just as good results from mixed gpu/smp systems by exclusively setting the priority and not touching the affinity. On my quad core I have 2 vmware instance running at idle priority and then 3 gpu clients with above normal priority, all set using bill2's.

 

[Alpha736]

So guys, everything seemed to be running fine until I started getting this error on my GPU client.

[10:43:07] Folding@home Core Shutdown: UNSTABLE_MACHINE
[10:43:12] CoreStatus = 7A (122)
[10:43:12] Sending work to server
[10:43:12] Project: 4743 (Run 2, Clone 976, Gen 20)
[10:43:12] - Error: Could not get length of results file work/wuresults_06.dat
[10:43:12] - Error: Could not read unit 06 file. Removing from queue.
[10:43:12] EUE limit exceeded. Pausing 24 hours.
[16:21:04] + Working...

What could be causing this?

 

[APOLLO]

So guys, everything seemed to be running fine until I started getting this error on my GPU client.

[10:43:07] Folding@home Core Shutdown: UNSTABLE_MACHINE
[10:43:12] CoreStatus = 7A (122)
[10:43:12] Sending work to server
[10:43:12] Project: 4743 (Run 2, Clone 976, Gen 20)
[10:43:12] - Error: Could not get length of results file work/wuresults_06.dat
[10:43:12] - Error: Could not read unit 06 file. Removing from queue.
[10:43:12] EUE limit exceeded. Pausing 24 hours.
[16:21:04] + Working...

What could be causing this?

That happens on my systems when the OC is too high or there's something wrong with card's cooling.

 

[Alpha736]

I'm trying to use one of these applications to set the core affinity, and I can't tell if it's working. I went to this page and tried using PriFinitty.

Also, why can't I just go into the task manager to set the core affinity?

Edit: It seems like PriFinitty is 32bit only, but it seems like it's might still working (I've gotten no errors, and the task manager displays the same affinity). Is there something else that I should be running on a 64bit system?

 

[Zero82z]

I'm trying to use one of these applications to set the core affinity, and I can't tell if it's working. I went to this page and tried using PriFinitty.

Go into Task Manager, right-click on your processes, select Set Affinity, and see which cores are checked off. If everything is the way you set it up to be, then it is working.

Also, why can't I just go into the task manager to set the core affinity?

You can, except it isn't persistent, so it will reset every time a unit is completed and the core restarts.

Is there something else that I should be running on a 64bit system?

Yes, try Bill2's Process Manager: http://www.bill2-software.com/processmanager/download.shtml

 

[Alpha736]

Go into Task Manager, right-click on your processes, select Set Affinity, and see which cores are checked off. If everything is the way you set it up to be, then it is working.

You can, except it isn't persistent, so it will reset every time a unit is completed and the core restarts.

Yes, try Bill2's Process Manager: http://www.bill2-software.com/processmanager/download.shtml

It appears that PriFinitty actually is working, would I get any benefit if I were to switch to Bill2's?

 

[Zero82z]

It appears that PriFinitty actually is working, would I get any benefit if I were to switch to Bill2's?

If what you're using now is doing the trick, I don't see a reason for you to switch unless there's a particular feature that Bill2's has and PriFinitty doesn't.

 

[Alpha736]

It appears that PriFinitty actually is working, would I get any benefit if I were to switch to Bill2's?

If what you're using now is doing the trick, I don't see a reason for you to switch unless there's a particular feature that Bill2's has and PriFinitty doesn't.

Thanks, I think I'll just keep it the way I have it.