By: David Kanter (dkanter.delete@this.realworldtech.com), August 10, 2011 6:01 pm
Room: Moderated Discussions
ltcommander.data (ltcommander.tuvok@gmail.com) on 8/10/11 wrote:
---------------------------
>While it is compatible with DX10.1 and OpenGL 3.0, it lacks the features necessary
>for OpenCL or DirectX Compute Shaders. To a large extent, this is because Intel
>designed the Sandy Bridge graphics before OpenCL standard was finalized. .
>
>Maybe I missed it, but it seems the article mentions several features in Sandy
>Bridge that will be useful for OpenCL, but doesn't actually mention what feature
>is missing that prevents OpenCL support in Sandy Bridge. >Is it possible for you
>to point out the big problems or is it under NDA?
I thought I explicitly mentioned it somewhere...but perhaps not. Here are several thoughts:
AFAIK, Intel has not exposed any shared memory to SW, which is required for OpenCL. They could use the L3 cache for shared memory, but the performance seems like it would be pretty awful due to high latency.
I also wonder about numerical accuracy.
It's also possible that OpenCL is feasible, but has such abhorrent performance that they judged it better to simply wait for the next generation.
>http://software.intel.com/file/34436
>
>And in regards to Compute Shader support, Sandy Bridge in fact supports CS4.x according
>to the Developer Guide above, pg 14 and 18. While not all DX10/10.1 GPUs support
>Compute Shaders and OpenCL (ie. ATI HD2000/HD3000), it previously seemed like GPUs
>that support at least Compute Shader 4.x also support OpenCL (ATI HD4000 and nVidia
>8000). Is there something distinctively different between CS4.1 and OpenCL 1.0 that
>can't be worked around, which Intel seemed to have >conveniently fallen into?
I'm probably not conversant enough with the inner details of OpenCL and CS to make that assessment. Honestly, there are a lot of subtle issues that could go wrong.
>http://downloadmirror.intel.com/20035/eng/Graphics%20driver%20release%20notes.pdf
>
>And Intel seems to have enabled OpenGL 3.1 support in their latest drivers for
>Sandy Bridge, up from OpenGL 3.0 at launch. I guess Ivy Bridge will need a new
>checkbox feature, perhaps they can jump directly to >OpenGL 3.3?
I would think that with IVB they should mostly catch up to NV/AMD in terms of APIs.
David
---------------------------
>While it is compatible with DX10.1 and OpenGL 3.0, it lacks the features necessary
>for OpenCL or DirectX Compute Shaders. To a large extent, this is because Intel
>designed the Sandy Bridge graphics before OpenCL standard was finalized. .
>
>Maybe I missed it, but it seems the article mentions several features in Sandy
>Bridge that will be useful for OpenCL, but doesn't actually mention what feature
>is missing that prevents OpenCL support in Sandy Bridge. >Is it possible for you
>to point out the big problems or is it under NDA?
I thought I explicitly mentioned it somewhere...but perhaps not. Here are several thoughts:
AFAIK, Intel has not exposed any shared memory to SW, which is required for OpenCL. They could use the L3 cache for shared memory, but the performance seems like it would be pretty awful due to high latency.
I also wonder about numerical accuracy.
It's also possible that OpenCL is feasible, but has such abhorrent performance that they judged it better to simply wait for the next generation.
>http://software.intel.com/file/34436
>
>And in regards to Compute Shader support, Sandy Bridge in fact supports CS4.x according
>to the Developer Guide above, pg 14 and 18. While not all DX10/10.1 GPUs support
>Compute Shaders and OpenCL (ie. ATI HD2000/HD3000), it previously seemed like GPUs
>that support at least Compute Shader 4.x also support OpenCL (ATI HD4000 and nVidia
>8000). Is there something distinctively different between CS4.1 and OpenCL 1.0 that
>can't be worked around, which Intel seemed to have >conveniently fallen into?
I'm probably not conversant enough with the inner details of OpenCL and CS to make that assessment. Honestly, there are a lot of subtle issues that could go wrong.
>http://downloadmirror.intel.com/20035/eng/Graphics%20driver%20release%20notes.pdf
>
>And Intel seems to have enabled OpenGL 3.1 support in their latest drivers for
>Sandy Bridge, up from OpenGL 3.0 at launch. I guess Ivy Bridge will need a new
>checkbox feature, perhaps they can jump directly to >OpenGL 3.3?
I would think that with IVB they should mostly catch up to NV/AMD in terms of APIs.
David
