How to use filters that require OpenCL without a graphics card

redGSRguy · Messages: 99 Registration: 02.10.2002

I have server with Xeon 1275 v1 processor and whenever i try to filters like dehalo or AA or QTGMC it fail because i don't have opencl drivers installed. anyway to bypass this?

Hollywood Hamilton · Messages: 3,015 Registration: 10.25.2002

OpenCL is designed as versatile API. It provides functions which can take advantage of massive parallelization in a GPGPU if available. But the same functions may also be implemented with SIMD parallelization on a CPU, just with lower efficiency due to limited parallelism. It all depends only on the drivers, whether they provide code for both cases or only for the specific computing hardware they are designed for. It's well possible that if you don't have a GPGPU useful for OpenCL, the graphic drivers may not offer OpenCL as a whole. Intel provides "legacy" CPU-only OpenCL drivers for Core and Xeon 64-bit processors. AMD seems not to support CPU-only OpenCL.

sluggo · Messages: 134 Registration: 06.16.2002

Deprecated. Check release notes for supported hardware. May require registration or similar, not sure.

///RBR · Messages: 35 Registration: 01.29.2002

ACKR, can you post your script?

Luke · Messages: 2,065 Registration: 08.15.2003

Real.Finder's interest may be moslty in QTGMC usage. Give Staxrip's args to QTGMC if staxrip exposes it. also, what version QTGMC does staxrip use.

ramboz · Messages: 31 Registration: 08.28.2003

I do believe you are going to have to get a video card that works with OpenCL, as every CPU implementation is bad (from everything that I can find). It is driver based, but if the drivers suck, then it might as well not exist.

\\\WsO816 · Messages: 18 Registration: 08.14.2003

SIMD CPU instructions can parallelize a handful of calculations; GPGPU instructions possibly hundreds (as calculated in shader units). So OpenCL on a GPU can speed up calculations when you have algorithms that need to calculate rather simple instructions on a large number of values which don't depend much on each other. Rather small filter kernels on a whole line or stripe of pixels in a video frame are such a case: With hundreds of shader units you can filter the whole width in a few steps, whereas a CPU needs several hundred steps despite using SIMD instructions because they can still handle only e.g. 4 or 8 values at once.