Trying to use ACML-GPU fails on low-end hardware
It is possible to use the ACML on a Radeon 2400 XT.
The website
http://ati.amd.com/technology/streamcomputing/requirements.html
says: yes.
I installed the latest driver 9.1 (8.57.2) and the latest acmlg v0.7on an 7.1.0 XOrg on debian/lenny.
In the examples directory, there is acmlinfo.exe and a "info" binary.
The info binary fails to detect a GPU, if you are not
- running an XServer, e.g. DISPLAY=:0.0
- have the authentication do connect to the main Display, (see xhost,xauth and "export DISPLAY=:0.0)
Then:
./info
CAL RT version: 1.3.158
CAL CL version: 1.3.158
GPUs found: 0
On the other hand, if you are working locally, or have remotely access to the local XServer, it works well:
LD_LIBRARY_PATH=../opt/acmlg0.7/gfortran64/lib/ ./info
CAL RT version: 1.3.158
CAL CL version: 1.3.158
gpu0:
Type: RV610 GPU ISA
Maximum resource 1D width: 8192
Maximum resource 2D width: 8192
Maximum resource 2D height: 8192
Local GPU RAM: 256 megabytes
Uncached remote GPU memory: 1967 megabytes
Cached remote GPU memory: 60 megabytes
GPU device clock rate: 0 megahertz
GPU memory clock rate: 0 megahertz
Wavefront size: 64
Number of SIMDs: 2
double precision: Not supported
local data share: Not supported
global data share: Not supported
global GPR: Not supported
compute shader: Not supported
memexport: Not supported
calResCreate pitch alignment: 64 data elements
calResCreate address alignment: 256 bytes
GPUs found: 1
Also, the pre-compiled binary work, but recompiling the test programs under debian/lenny fails on first sight (using gfortran-4.2 or 4.3) ...
More experience on the "low-end" hardware (2400, 2600, 2900) is welcome.