• 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.

Multi-VM + Multi+GPU issue

mavalpha

[H]F Junkie
Joined
Jan 3, 2005
Messages
10,448
Just got back from a friend's house, having some issues with his setup.

i7
3x GTX295
12GB
Vista x64
Nothing overclocked, no temps over 80C.

What I got working:
NotFred x3 in VMWare Server, Console monitoring not working. Says plugin not installed, even when it is. CPU utilization is where it should be, though, so I'm 98% sure it's good. Didn't want to set up a fourth, so as to leave one core dedicated to the GPU clients.
GPU2 x3, between 6500 and 9000 ppd.

What's not working:
The other GPU2 x3 clients. Dummy plugs are there, but all three EUE as soon as as they start. Interestingly, I think it is actually instances 3-5 (using 0-5 for naming) that are crashing. Coincidence?
 
Did you do a barrel roll?

I am really interested in knowing the solution etc, cause my set up to be built will be basically multi VMs / Multi GPUs.
 
Whenever I've had problems with GPU clients EUE'ing at the start it was usually because I had forgotten to extend the desktops. Also, everyone says Vista will support a max of 5 GPU clients.
 
Have you tried running the GPU clients without the VMs running to see if they will work by themselves? Also, try running each of the problematic GPU clients one at a time to see if they'll fold alright when nothing else is running.
 
Whenever I've had problems with GPU clients EUE'ing at the start it was usually because I had forgotten to extend the desktops. Also, everyone says Vista will support a max of 5 GPU clients.
Nope, definitely extended across all. And I'm stuck at 3 out of 6, I warned him about the 5-GPU limit and he said it was okay. We'll be happy to get two more.
Have you tried running the GPU clients without the VMs running to see if they will work by themselves? Also, try running each of the problematic GPU clients one at a time to see if they'll fold alright when nothing else is running.
Same problem. Actually focused on GPU's first, then gave up and switched over to VM's. I'm not 100% sure right now, but I think they had the same problem when run individually as well.
Did you do a barrel roll?

I am really interested in knowing the solution etc, cause my set up to be built will be basically multi VMs / Multi GPUs.
You may want to do that soon, because stats project that my retired account will overtake you in less than two months- and that's just with stragglers from my glory days! :p
 
Last edited:
Are these sandwich or single PCB 295s?
What Brand?
Which drivers?
What mobo?
 
You probably won't get 3x GTX 295s to work in Vista. I worked on it for a long time (very similar setup) and was able to get at most 5 of the 6 GPUs to fold. It's an issue with WDDM and the NVidia drivers. The old tweaks of dummy plugs and extending desktops don't help with the 295, because CUDA still has trouble seeing the second GPU on each card (except maybe the first).

The workaround is to install the 190.xx drivers and enable quad-SLI. That will allow you to use all four GPUs on two of the cards. You should then be able to use a single GPU on the remaining card (with dummy plug attached). I got that working, but didn't like wasting a GPU. Also, three cards packed tightly together ran too hot for my configuration. I pulled one 295 and put it in another machine, and I'm just using two cards in quad-SLI.

The other workaround it to install Windows XP. That is known to use all GPUs with no problems.
 
Try adding the -forcegpu nvidia_g80 switch to the 4th client, which is, as you say the first one not working out of 3,4,5.

If it works, add it to the other two. If they are the single PCB GTX 295s that's likely the issue, as the second port is not enabled unless there is a monitor or dummy plug on it, or you use that flag/switch. If they are the dual-PCB version then you can still try it. No harm in it.

Good luck!
 
All three are dual-PCB eVGA's of various OC releases, in a Classified board. On a whim, I clocked all of the OC'd boards down to reference speeds, but nothing changed so they're back at original frequencies. All drivers are the latest, AFAIK.

Next time I'm over there I'll try the -forcegpu flag, as well as Quad-SLI (although I'll admit, this is the first time I've heard about SLI being supported, let alone beneficial, for FAH).
 
Ok well since they are dual PCB only 5 GPUs are going to be recognized by Vista.

So next time you go over there try it with just one card in and try to get both GPUs folding on that single card.
Then install the next card,
Then put the other on in a different system.
 
Try adding the -forcegpu nvidia_g80 switch to the 4th client, which is, as you say the first one not working out of 3,4,5.

Try installing the latest BOINC client. It will tell you how many CUDA devices are available on the Messages tab when it starts up. I'm guessing you'll only see three CUDA devices, so adding command-line flags in F@H won't make a difference.

With dual-PCB cards, you can try adding dummy plugs to the HDMI ports. People have had mixed results doing this, but it's better than absolutely not working. I have the single-PCB cards, and I haven't seen any improvement as far as getting things working in Vista.
 
What monitor(s) do you have?

Instead of dummy plugs on my one rig I just ran HDMI -> DVI cable to the extra input on the monitor and it works fine.

You cant use VGA dummy plugs with an adaptor for HDMI ports because of the different signal.
 
Back
Top