Forum > Discussion Forum

Intel OpenCL

<< < (2/2)

cristipurdel:

--- Quote from: Frizz on 20 Nov 2010, 04:13:01 pm ---
--- Quote from: cristipurdel on 20 Nov 2010, 02:21:27 pm ---
--- End quote ---
Dreaming of a 17W Sandy Bridge running 2 + 2 + 12 seti tasks :P

--- Quote ---
--- End quote ---

do you know if the integrated GPU of sandy/-ivy bridge will support OpenCL - or only the CPU?

--- End quote ---
http://en.wikipedia.org/wiki/GMA_HD#HD_Graphics_.28GMA_HD.29 ;)

cristipurdel:
... Number of CL Platforms : 3 on an intel + nvidia system :P
Apparently OpenCL is the ...future :)

Frizz:
1) I would like to try AMDs OpenCL CPU implementation (because of printf(), etc. *g*). Are there any limitations when using the CPU device compared to the GPU device? I think local work groups was limited to 1 for CPU, no?

2) Another question: When I change local work group size, does it have an impact on the algorithm itself (besides performance)?

Raistmer:
Not right thread for such questions IMO, but :) :
1) max workgroup size remains the same.
Look into stderr of AP task, it liss device parameters.
2) yes, algorithm can change indeed. For example, look my post about why NV MB build should not work with workgroup size of 32.
[more technically, kernel aware of its launch parameters, in particular, about workgroup size, via get_local_size() call.
So, surely it can and sometimes will behave differently for different workgroup sizes]

cristipurdel:
So now OpenCL on Intel EUs....grrrr

Navigation

[0] Message Index

[*] Previous page