<html>
<head>
<meta content="text/html; charset=UTF-8" http-equiv="Content-Type">
</head>
<body bgcolor="#FFFFFF" text="#000000">
Since AMD just published it's Linux FGLX Catalyst driver with their
renamed Accelerated Parallel Processing (APP) in conjunction with
their AMD 7990 GPGPU<br>
<br>
<a
href="http://blogs.amd.com/play/2011/12/21/764-days-as-graphics-performance-champion-%E2%80%93-and-counting/">http://blogs.amd.com/play/2011/12/21/764-days-as-graphics-performance-champion-%E2%80%93-and-counting/</a><br>
<br>
Developers: <a
href="http://developer.amd.com/sdks/AMDAPPSDK/Pages/default.aspx">http://developer.amd.com/sdks/AMDAPPSDK/Pages/default.aspx</a><br>
<br>
In the release notes: <a
href="http://developer.amd.com/sdks/AMDAPPSDK/assets/AMD_APP_SDK_Release_Notes_Developer.pdf">http://developer.amd.com/sdks/AMDAPPSDK/assets/AMD_APP_SDK_Release_Notes_Developer.pdf</a><br>
<br>
<blockquote type="cite">1 What’s New in AMD APP SDK v2.6<br>
Key features supported in SDK 2.6 and the Catalyst 11.12 drivers
include:<br>
<b>• OpenCLTM runtime integration into Linux and Windows® Catalyst
drivers.</b><br>
• Inclusion of the Khronos C++ wrapper API.<br>
• Multi-GPU support on Linux platforms.<br>
• PX5 support.<br>
• Preview: Support for AVX instructions on CPUs that support AVX.<br>
• Support for FMA4 instructions in OpenCL built-in function
libraries on CPUs that support<br>
FMA4.<br>
• Kernel reflection, query kernel parameters, and enable use of
OpenCL kernels in data-driven<br>
applications.<br>
• Support for atomic counters on APUs.<br>
• Redesign of OpenCL run-time on CPU, significantly improving
performance.<br>
• Support for the cl_amd_media_ops2 extension, exposing hardware
capabilities for<br>
accelerating image-related processing.<br>
• Async copies preview (set environment variable
GPU_ASYNC_MEM_COPY=2 to enable).<br>
The OpenCLTM 1.2 preview includes the following capabilities
(requires 8.93.10 preview drivers):<br>
• Host access flags for memory objects enables more efficient
buffer handling.<br>
• Pattern-based GPU buffer and image initialization eliminates
need for certain buffer/image<br>
transfers.<br>
• Memory objects migration supports early transfer of buffers in
preparation for when they are<br>
needed.<br>
• New generalized image creation API.<br>
• Enhanced image/buffer map operations.<br>
• OpenCL 1.2 CPU device partition, including partition of a CPU
after addition to a context.<br>
• Generalized 1D and 2D images, image arrays, and
image<-->buffer interop.<br>
</blockquote>
<br>
It seems like AMD woke up and followed NVidia's move for it's
proprietary [non-free] driver to make my wishes come true.<br>
<br>
I'm not interested in server environments without top end GPUs that
can utlize the AMD driver. Nor is anyone else going to use any of
the streams on such an entry level card from AMD [though it's rather
hard to find such a weak card today] for GPGPU cycles.<br>
<br>
In my fields of Computer Science and Mechanical Engineering it's
rather obvious any blade servers and workstations will want cycles
for Finite Element Analysis and Computational Fluid Dynamics first
for HPC over database and app server deployment.<br>
<br>
If Debian can't manage to package what now is required I'm sure I
can build my own custom debs.<br>
<br>
- Marc<br>
<br>
On 12/22/2011 12:07 AM, Bastien ROUCARIES wrote:
<blockquote
cite="mid:CAE2SPAaLKhAQfTiSe4k6Kz=-bNFqmoZ_G9APs5ox5yvcwn8BBg@mail.gmail.com"
type="cite">
<p>I will not do this without a free generic software fallback. </p>
<p>Think about server environnement without top edge graphic card.
Imagemagick should work in this environment without modification</p>
<p>Open an rfp or an itp, test it and package it and will do
something</p>
<p>Bastien</p>
<blockquote type="cite">Le 22 déc. 2011 05:51, "Marc J.
Driftmeyer" <<a moz-do-not-send="true"
href="mailto:mjd@reanimality.com">mjd@reanimality.com</a>> a
écrit :<br>
<br>
<div bgcolor="#FFFFFF" text="#000000"> I'm not too concerned
with OpenCL support in a free version, nor have I ever been.<br>
<br>
Seeing as AMD and Nvidia, not to mention Intel are releasing
their implementations directly for BSD or similar license use
my needs have been fulfilled.<br>
<br>
More to the point, all three have to adhere to Khronos
specification created by Apple and now in their committee for
all to leverage.<br>
<br>
That's all that really needs to happen.<br>
<br>
I would suggest when AMD releases their opencl sdk that will
be working with LLVM/Clang that the Clang package guys
coordinate with the AMD non-free driver packagers and make
sure the AMD SDK for OpenCL is included in the AMD fglx driver
files ala what Nvidia package maintainers do for the nvidia
non-free driver stack.<br>
<br>
- Marc
<p><font color="#500050"><br>
<br>
On 12/21/2011 07:04 AM, Bastien ROUCARIES wrote:<br>
><br>
> tags 630321 + wontfix<br>
> severity 630321 wishl...</font></p>
<font color="#888888">
<div>-- <br>
Marc J. Driftmeyer<br>
Email :: <a moz-do-not-send="true"
href="mailto:mjd@reanimality.com" target="_blank">mjd@reanimality.com</a><br>
Web :: <a moz-do-not-send="true"
href="http://www.reanimality.com" target="_blank">http://www.reanimality.com</a><br>
Cell :: <a moz-do-not-send="true"
href="tel:%28509%29%20435-5212" value="+15094355212"
target="_blank">(509) 435-5212</a> </div>
</font></div>
</blockquote>
</blockquote>
<br>
<div class="moz-signature">-- <br>
Marc J. Driftmeyer<br>
Email :: <a href="mailto:mjd@reanimality.com">mjd@reanimality.com</a><br>
Web :: <a href="http://www.reanimality.com">http://www.reanimality.com</a><br>
Cell :: (509) 435-5212
</div>
</body>
</html>