Why is each Catalyst version getting worse than the previous one?
Is it because AMD already redesigns their drivers to support the new GCN architecture? And breaks one (old) feature after the other?
Owners of those very expensive dual-GPU cards like the 5970 and the 6990 must be really angry - since those cards are STILL not supported properly.
Why in the world would anyone trust AMD and buy their new high-end products - when history has proven that they f**k their old customers over and over again.
Comments like these are not too helpful to any parties.
Drivers getting worse than previous versions is not the worst thing, it means they are working on it. Otherwise, you would have bought a N*I**A card to develop your OpenCL stuff.
Until there are killer applications out there using OpenCL, posts like yours won't do a dent on them. Anyone trying to release applications implemented using OpenCL in next few months is probably a bad idea.
Yeah that's a great idea! I just wait some month - or years - till AMD get's it sh*t together and fixes it's buggy drivers.
Or: I switch to CUDA / NVIDIA.
I'm trying to write a cross-platform 2D vector graphics library accelerated with OpenCL. Bugs like write_imagef writing wrong color values (http://forums.amd.com/forum/messageview.cfm?catid=390&threadid=156117&enterthread=y) mean I can't work on my project. The only option to me right now is to work with an nvidia card, hoping there are no such bugs in nvidia's OpenCL and waiting for AMD to fix the bugs.
If that's the way you feel, you should report _specific_ problems. Let me give you an example:
My system integrator recently installed a FireStream 93x0 next to an NVIDIA Tesla C2075. In order to get the NVIDIA CUDA tools working for customers, his standard installation changes the default symlink of GCC to 4.3 instead of 4.5. Most packages which require a specific GCC version to build will include the appropriate check against the version. AMD's did not.
This required time to diagnose.
Now I have the machine and am preparing some code on it, and the installation of the drivers didn't include any OpenCL... So I have to manually prepare the machine... I once wrote an install script for the AMD APP SDK, but that was for 2.5. I shouldn't need to install the whole APP SDK to run OpenCL programs on a Linux box. At the moment, my code runs fine on the OpenCL installed by NVIDIA.
Please AMD, if you want people to take AMD Fusion seriously, make it easy to install OpenCL support on Linux.