Right, each need their own cores though. Use Process Lasso to make sure everyone sticks to their own cores. Make sure you have the latest drivers loaded up it helps if you forget and start playing movies or something visual while still folding on the GPU, helps avoids restarts from system unresponsivenessish(you'll understand if you do it).
0
LeonardoWake up and smell the glaciersEagle River, AlaskaIcrontian
edited November 2008
Question, I can run the GPU2 and SMP clients at the same time without one hindering the other right?
Yes you can! (not a political answer )
If you are running a quad core CPU, you can run two GPU clients as well in addition to an SMP client. Here's how I do that: WinSMP Folding on Cores 0, 1; GPU Folding one each on cores 2 and 3. You can also put two GPU clients on one CPU core, letting SMP have the three other cores. I get better, more consistent results with SMP limited to cores 0 and 1, GPUs on 2 and 3.
If you will be using a dual core CPU, you can run one standard CPU Folding client and one GPU. Each GPU client needs a dedicated core of the CPU to feed data to the GPU (or something like that).
OK, i've been out of the folding scene for a long time now.
I just have a C2D E7200 with the 9800GT mentioned about.
What's the most efficient setup for this in your opinion?
Thanks. Cheers
0
LeonardoWake up and smell the glaciersEagle River, AlaskaIcrontian
edited November 2008
One GPU2 client, using one of the cores.
One standard F@H client using the other CPU core.
You should be good for 5000+PPD.
Use Process Lasso to lock core affinities, one client process for each core.
Overclocking - both CPU and GPU - brings very good results as long as your machine is stable.
The GPU clients have a feature to lock the process (affinity) to a single core. This does not always work optimally on a machine with multiple Folding clients. That's why K and I recommended Process Lasso, which is easy to install and configure.
I've noticed on my rigs that the standard client can take nearly as long to complete a WU as the SMP client. But, my rigs are getting dated. What I'm wondering though is if it might be more points productive to run the SMP client with the GPU client even though there are only two cores, so long as the SMP client finishes the WUs within the deadline. Maybe this would require Vista to get the lower CPU usage for the GPU client.
I can't test this myself the theory myself - my desktop's card is too old and the newest Dell/Nvidia drivers broke the GPU client on my laptop.
0
LeonardoWake up and smell the glaciersEagle River, AlaskaIcrontian
edited November 2008
Well, you've got a good point concerning testing. Many users have found that letting SMP share a CPU core with the GPU client (necessary with a dual core) robs the GPU of more production than it makes up for through SMP efficiency. Some say that on the balance, they get slightly more production without cores set for specific affinity. There are several factors at play, especially video card model and operating system. For instance, Vista requires less of a CPU core's power for GPU Folding than does XP.
My quads, folding one each GPU and SMP, I set affinities for the GPU client to have one core to itself and for the the SMP client to have cores 0,1,and 2. I experimented with SMP running on all four cores. The SMP client did produce much better, but it robbed about 2000PPD production from the GPU client.
I'm also running a quad core (Q6600) with one SMP and two GPU clients. SMP is assigned cores 0,1, and the GPUs are each exclusively assigned one of the remaining cores. That's on XP SP3.
Ya thats what I found you really need to block cores off for the GPU client, I even dropped down to just a single core on a CPU client and it bumped my GPU clients up a small amount. I think that was just because of the system resources needed for xp-64. Thats what I have been running for the past few months and it seems to work really well 11k-12k when left alone.
Comments
Question, I can run the GPU2 and SMP clients at the same time without one hindering the other right?
I'm gonna setup my 9800GT tonight
FOLD ON :cheers2:
If you are running a quad core CPU, you can run two GPU clients as well in addition to an SMP client. Here's how I do that: WinSMP Folding on Cores 0, 1; GPU Folding one each on cores 2 and 3. You can also put two GPU clients on one CPU core, letting SMP have the three other cores. I get better, more consistent results with SMP limited to cores 0 and 1, GPUs on 2 and 3.
If you will be using a dual core CPU, you can run one standard CPU Folding client and one GPU. Each GPU client needs a dedicated core of the CPU to feed data to the GPU (or something like that).
Here's a good guide for setting up GPU Folding - multi or single GPU.
If you run into any problems, please post! GPUs can be amazing processors.
I just have a C2D E7200 with the 9800GT mentioned about.
What's the most efficient setup for this in your opinion?
Thanks. Cheers
One standard F@H client using the other CPU core.
You should be good for 5000+PPD.
Use Process Lasso to lock core affinities, one client process for each core.
Overclocking - both CPU and GPU - brings very good results as long as your machine is stable.
The GPU clients have a feature to lock the process (affinity) to a single core. This does not always work optimally on a machine with multiple Folding clients. That's why K and I recommended Process Lasso, which is easy to install and configure.
I can't test this myself the theory myself - my desktop's card is too old and the newest Dell/Nvidia drivers broke the GPU client on my laptop.
My quads, folding one each GPU and SMP, I set affinities for the GPU client to have one core to itself and for the the SMP client to have cores 0,1,and 2. I experimented with SMP running on all four cores. The SMP client did produce much better, but it robbed about 2000PPD production from the GPU client.
I'm also running a quad core (Q6600) with one SMP and two GPU clients. SMP is assigned cores 0,1, and the GPUs are each exclusively assigned one of the remaining cores. That's on XP SP3.