A little assistance for a semi-noob, please!
it's been a while since I folded (or at least set up a rig to fold properly..) last, and with such active competiton on these here boards I couldn't help but try and dive into some folding. Searching through the forums has deemed useless, but maybe I'm not searching for the right keyword.. anyhow.
I've never used the enhanced or GPU clients before, and I'm trying to set up both (enhanced SMP and GPU clients) to run simultaneously on my main rig. The problem is I don't believe they like running together at the same time. the CPU client takes anywhere from 30 mins - 1 hour to fold 1% (big workunits) when the GPU client is running, about 7 minutes elsewise. Should I be tying down affinities to specific CPU cores? or not even trying to run both at the same time and just sticking to one or the other?
-smp flag is used (not -advmethods though) big workunits set in the config. A3 core.
thanks for any help! anddd... Way To Fold WOOT!
Looks fine to me, having a screenshot of Fahmon would be great to see the actual ppd. If you think the affinities are not correct, see this: http://www.hardwarecanucks.com/forum...ty-winafc.html
well apparently fahmon doesnt correctly predict PPD, but what it is predicting is multiple times lower when the GPU client is enabled.
I'm going to try the latest version of fahspy in a moment!
further investigation yields that the GPU client is taking up ~ 25% cpu power. I might try dedicating that to one core, and giving the cpu client the other 3.
Indeed, with an ATI card you need to have 1 CPU core dedicated to GPU folding.
Fahmon is fine. Fahspy too, same for HFM.net
What operating system are you folding with?
On XP I have been using the affinities to seperate the cpu folding from the gpu folding. Set priorities for GPU at one level higher the then the cpu processes.
If using VMware or Linux with Freds diskless setup you can add a couple of lines to your virtual machine's VMX file and add the following lines make each VM use only specific cores.
As an example:
First vm add these lines to the vmx file
processor0.use = "TRUE"
processor1.use = "TRUE"
processor2.use = "FALSE"
processor3.use = "FALSE"
Second vm add
processor0.use = "FALSE"
processor1.use = "FALSE"
processor2.use = "TRUE"
processor3.use = "FALSE"
Then use your favorite affinity program like prifinitty2 to force the gpu to use your final core.
Hope it helps.
win 7, no VM (yet?)
First, follow the instructions in this thread to reduce the CPU usage of your GPU client: GUIDE: ATI Client - Lower CPU % + Raise PPD - [H]ard|Forum
Set the FLUSH_INTERVAL to 128 since you have a 4890. Also, you don't have to use the 9.4 drivers specifically; anything newer will work as well.
Next, download and install Bill2's Process Manager. Configure the program to run on startup by going into Options>Main>Automatically start with Windows (you'll probably want to check off the "Start minimized in the system tray" option as well). Run both the CPU and GPU F@H clients. You should see Fahcore_11.exe and either Fahcore_a3.exe or four instances of Fahcore_a1.exe both listed in the main Bill2's window. If not, check off the "Show hidden processes" box at the top.
Right-click on Fahcore_a3.exe or one of the Fahcore_a1.exe instances, and select "Create a rule for this process". A window will pop up giving you options. Uncheck the "Use full path" box. In both the "Normal window" and "Minimized window or hidden process" tabs, check off the "Set priority" box and pick the "Idle" radio button. Click Ok.
Now, you are running an A1 or A3 core workunit, but at some point you will most likely receive one using whichever core you aren't running right now. When that happens, open Bill2's and do the same thing for that process as well.
alright, thanks. I've followed these instructions. the GPU core does seem alot lower.
I'll let it run for a while and see how the PPD works out.
Zero, I owe you a beer next time I'm in Montreal.
curiously, would I have had such problems in a VM anyways? from what I'm reading on here VMs look like an easy way to significantly boost PPD, so I might try it out sometime...
but not yet! and I have another (noobish?) question! (as I search for the answer via other forums..)
I'm having a dual core crunch numbers, and it's downloaded the A1 client I noticed, but it's running 4 of them, not 2 as I would think would be the correct amount to run. Is this correct or did I make a configuration mistake somewhere?
thanks again in advance.
Using a VM wouldn't really be any easier, and you'd end up with more resource usage and less system responsiveness. VMs were a way to boost PPD before the A3 core was made public, but now they're just about obsolete unless you have 8 available cores or more that you can run -bigadv workunits on.
|All times are GMT -7. The time now is 01:32 PM.|