Will F@H take any advantage of the dual-cored 5970, or will it just use the first chip and idle the other? //Svein
The setup should be the same as if you have two seperate cards in crossfire, i.e. two instances of the f@h client.. follow the advanced setup guide for two cards here.. http://folding.stanford.edu/English/WinGPUGuide#ntoc4 Good luck, post again if you need more help..
If you are running multiple GPU cards with SLI or Crossfire, you need to extend the desktop for the client to recognize multiple GPUs. That's the part that's been holding me back.
That's the part that means folding@home won't be getting my second gpu if I upgrade card this autumn... //Svein
The reason why F@H will never support XFire or SLI is because it is too slow thus will have a negative effect on the Science. It is much faster to have two independent instances of F@H GPU Clients as it gets more science done. There are obviously set-up issues, especially if you are new to F@H, and that is going to be addressed in the v7 of the F@H Client. @brazen: If you have enabled XFire, and are using the latest version of the F@H GPU Client, it should work with the proper flags, First GPU: -gpu 0 -forcegpu ati_r800 -verbosity 9 Second GPU: -gpu 1 -forcegpu ati_r800 -verbosity 9 Is the F@H Client giving errors or something? Also you may want to use Environment variables to make GPU folding more efficient.
Haven't tried it since I don't want to run an extended desktop. If you're saying the documentation is outdated, however ...
The document isn't outdated, it's just that it contains the methods that work fine and is "fool-proof". However, there are some situations where methods A works fine for donor A but doesn't work fine for donor B. If you are using the GPU2 Client, I have read that it can work fine if you use the above flags and enable XFire (you may want to test it out). You may also want to try the GPU3 BETA Client as you can run without the -forcegpu flags. Same applies to Svein if are in the tweaking mood Do report your findings :nerd:
I will, if I get around to the 5970 part. Currently the temperatures are still high enough that I hemorrhage electricity pumping it out of the house, so upgrading GPU/room-heater now seems like a bad idea. But it'll come in handy for winter. //Svein
Success! I've downloaded Folding@home-Win32-GPU_Vista-631 to get the familiar console, but have to run it with -gpu 1 -forcegpu ati_r800 -verbosity 9; it wouldn't accept my graphics card without the forcegpu flag. It results in the following: Calling '.\FahCore_11.exe -dir work/ -suffix 01 -nice 19 -priority 96 -nocpulock -checkpoint 15 -verbose -lifeline 440 -version 630' Priority set to low as I also have the SMP client running and it's nicely doing its thing on the second GPU as it normally isn't used doing desktop stuff. HFM reports Core GROGPU2. Shouldn't that be 3? I want the GPU3 work! All in all, pretty painless.
Technically, the GPU3 BETA Client is NV only, however, the ATI GPUs will work using the normal WUs that were available with GPU2 Client. The GPU3 BETA Client shouldn't need the -forcegpu as the whitelist has been updated to include all latest GPUs. Remove the -forcegpu flag only and see what happens.
Like I posted: "it wouldn't accept my graphics card without the forcegpu flag" It's under 82% load and at 60 C. Pretty nice.
That's good. Have you set up Environment Variables to make your folding more efficient? It does take time and patience to find the right combination.
At the time of the first post, I didn't. But I've now set them according to the document you linked in my introductory post: FLUSH_INTERVAL = 256 BROOK_YIELD = 2 CAL_PRE_FLUSH = 1 CAL_NO_FLUSH = 1 This results in 97-100% load, so it definately helped. However, it seems that my SMP client isn't doing anything anymore. The console "hangs" at: [11:35:03] Completed 1291600 out of 2000000 steps (64%) Checking processes reveals that FahCore_a3.exe *32 is using around 80% of the processor though, so I wonder what it's doing. FahCore_11.exe *32 is using 16%, higher than I would have expected. The status column in HFM.NET tells me SMP is at RunningNoFrameTimes and GPU3 is at Running. How do I get the SMP client to do work too? EDIT: Oh, the SMP client is working. The percentage updates are just a lot slower than the GPU client. Disregard.:bang: Still, what do the status messages in HFM.NET mean?
Okay, new post, else my previous one gets messy. It does seem that my SMP client is very slow when i run the GPU3 client. [11:50:27] Completed 1300000 out of 2000000 steps (65%) [12:29:37] Completed 1320000 out of 2000000 steps (66%) Delta is 39 minutes! Is it possible to make it run at least close to its previous speed? Compare: [05:47:39] Completed 860000 out of 2000000 steps (43%) [06:00:00] Completed 880000 out of 2000000 steps (44%) Only 13 minutes ...
Okay, there are a few things you can try: 1) Further tweak the Environment Variables to further reduce the CPU Usage. This will involve a lot of trial and error. 2) You have to change the F@H Core Priority from within the F@H Clients. Use -configonly flag and do these: SMP2 Client = idle GPU3 BETA Client = low/slightly higher That way, the SMP2 will have lower priority than the GPU one. Hopefully, it will help you. I think that you have setup both the priorities to be same hence there is a lot of "fighting" going on between the F@H Clients. Mine is setup the same way. Also you have to remember that the FahCore_a3 doesn't like it when CPU Cycles are being "stolen" from it. If it still doesn't help you, you can use -smp 3 which will leave 1 core to feed the GPU. Do remember that the PPD gain from GPU should be greater than or at least equal to the loss of PPD from SMP2 Client. HFM.NET is telling you that the F@H Client is running but its TPF isn't in the record as it is taking too long. After a few frames, it will then become green once it has the new TPF. Nothing to worry about here.
1) I don't really know what the environment variables do as I haven't found an explanation that I understood, so I'll leave them as is for now. 2) Already doing that since a couple of posts up. GPU3 priority choices were idle/low and I chose low as per the docs. I've started using -smp 3 with the SMP client. We'll see how that goes. Thanks for your help.
The best article that I have come across is here, however, you can browse through this forum and see if you can find any useful tips.