View Single Post
  #44 (permalink)  
Old May 3, 2010, 08:22 PM
0o0 0o0 is offline
Banned
F@H
 
Join Date: Oct 2009
Location: Toronto ON
Posts: 853

My System Specs

Default

Quote:
Originally Posted by 10e View Post
Here ya go. I remember reading this last year, so I haven't read it lately but it should provide some info on this performance delta:

Folding Forum • View topic - Unuseful FLOPS in ATI GPU Client?


Of course not quite Zero82Z. There's always more:

Technically the -forcegpu flag does the following:

1) Bypasses the built-in GPU2 client "driver check" routines

2) Forces the client to use the architecture stipulated by the flag following the -forcegpu switch > ati_r600, ati_r700 or nvidia_g80.

There is no nvidia_g200 or nvidia_gt200 switch.

Note: Yes, you can use the -forcegpu ati_r700 flag on an nVidia client, and it will 100% fail. I've done it. Much fun to watch.

3) Tells the client which FAH"Cores" to download, which will use the libraries for CUDA for nVidia OR Brook/CAL for AMD (presently)

4) Depending on which core is downloaded in step 3), an appropriate work unit will be downloaded and processing commences.

In this case, 0o0 we would use the -forcegpu flag simply because the GPU2 client doesn't know what the heck an AMD 58x0 card is.

This is why you need to remove the amdcalcl.dll and amdcalrt.dll files. These are old and do not work on a 5xxx series ATI/AMD card. When you do this, the GPU2 client will auto-magically use the ATI/AMD (Brook)CAL DLLs that came with the Catalyst drivers, and these DO recognize the 5xxx series of cards.
ok thanks for the explanation :) I'm going to pull my 5870 and plug in a 4650 for this month after all I guess..

From what I understand of that link, ati cards are less efficient because their cache can't be used by fah clients and the number of flops are cut in half? Why would the client want to do it twice on ati but not on nv?
Reply With Quote