Support for new GPUs (such as Kepler) in the v7 FAH client

hmmblah

Administrator
Staff member
A post from the PG blog:

In the past, support for specific GPUs was built into the client. We are working on ways to automatically update this information more easily within the v7 client to support new GPUs, such as the Kepler GPUs which have just came out. While the automatic update isn't ready yet, here is how one can manually do this:
  1. Download the GPUs.txt file from https://fah-web.stan...public/GPUs.txt
  2. Copy the downloaded GPUs.txt file to the client's run directory. The run directory is also called the data directory. In Windows there is a link to this directory in the start menu.
  3. After installing the file you must restart your client.

The client has a built-in GPUs.txt which it will use if it does not find one on disk. The client will print a message to the log, very early on, when it reads GPUs.txt from the run directory.

In a future version of the v7 client, this will happen automatically, but for now, we are updating this file on our web site and donors can do this update manually for new hardware.

http://folding.typep...fah-client.html

I can't get this to work. I never see the message that GPUs.txt was read. Anyone else have better luck?
 
I saw this post earlier but I didn't post it because I was confused with what PG was actually saying. To me, it seems like the v7 client has support to recognize kepler GPUs, but there is not yet support for an actual core/work units to allow kepler to fold. And since I don't have one I couldn't test it either way.
unsure.png
 
Yeah, that's how I read it too. They don't want to release new clients to detect new GPUs. By updating this file and building automatic downloads into the v7 client you wouldn't have to reinstall the client. I tried the file, but my client doesn't read it.

They have kepler listed wrong anyway still. 7im has filed a bug probably 14 days ago, but they haven't responded yet. Currently, a 680 is detected as Fermi GF114 (GTX 680). It should be Kepler GK104 (GTX 680). Getting the client to see the GPU properly is only half the battle though. Cores need to make use of new features and architecture.
 
Back
Top