F@H Maximising PPD - i7 2600 & GTX670

MGP

MGP

MGP

Soldato
Joined
24 Oct 2004
Posts
2,584
Location
Surrey
So as it says. what's the best way of making the most of this setup?

I'm Windows 7 with the F@H 7 client thingy. The i7 2600 is clocked at 4500 MHz.

Please don't suggest I should be clever and doing something with virtual machines as I have never got close to working that out, and have even less of a clue how you set anything up in linux.

I've tried all sorts of permutations. The predicted PPD seems to vary quite widely, in an unpredictable way and changes every few hours especially if a WU changes.

  • There was an incredible, but very unlikely 70k using a SMP limited to 6 clients (which when you see the logs says WUs could fail and in fact did) + 1 uniprocessor and the GPU.
  • With the SMP accessing all 8 CPU cores, plus the GPU, it was predicting some 45k, but then a few hours later has dropped to a a feeble 20K.
  • SMP only gets even worse :(

I know different WUs can create different returns but the range of PPD seems silly. I can see something changes with the lower PPD as the CPU temps also drop, currently down from 68 deg C to 62 deg, yet still 100% use. I know GPU needs a bit of a CPU core, but not the whole thing, but is it too harmful to the SMP for even part of the CPU core to be used? Are my PPD predictions even within a range of tolerance or might I have something else wrong :confused:
 
Last edited:
Go here and follow the instructions to download and install VMWare Player 3.0 and a pre-packaged Linux Folding appliance. It really is very simple this way and you'll probably get 15% more PPD than from client 7 for the CPU. Then you'll need to experiment to see if client 7 is better for the GPU than running the stand-alone console client (6.4?).

I find client 7's PPD predictions to be non-intuitive and prefer to watch everything from HFM.
 
Last edited:
The only way you will get consistant results is if you run the machine as a dedicated folder, plus you need to look more at what you get on a daily/weekly return from EOC rather than focusing on the client itself.

Another good tool is HFM.NET as advised if you want a more coherent overview of your machines performance.

Don't bother with Virtual machines, the only reason to use VMs was for bigadv, and those are now so out the window for 6/8/12 core machines that they just aren't worth bothering with anymore.

For my windows machines that are running mixed GPU/CPU projects I use the v7 client and on my [email protected] get anything from 17K (low performing WU) to 42K (high performing WU) but also on the same WU can fluctuate between 30 and 40K depending on if I am using it or not. Just browsing the web will affect the output by up to 10-15K

That machine is also running Milkyway@home on a 7970 with one HT dedicated to it.
 
Yep, as mentioned before, in order to get consistent PPD predictions is to have it as a dedicated folder. I suppose a more accurate way is to actually monitor PPD yourself by making notes each day at a certain time, and seeing how each day varies from the others. The results will obviously fluctuate for many reasons, but I guess it's more accurate than the client's rough on-the-fly estimates.

And just as an extra, my 3770k at 4.5 varies from around 13k to 30k in the v7 client, so it's pretty normal I guess.
 
You must log in or register to reply here.
Back
Top Bottom