Yes, that should be possible.
Xeon Phi (formerly knights corner and knights ferry) is a PCIe 2.0 accelerator card (it is not a graphic card!) with 60+ Pentium-like CPUs (modified P54C, i.e. the old Pentium). It operates at 1GHz, provides 1+ TFLOPs (peak) double precision, and has 320 GB/s memory bandwidth to the on-board memory:
http://www.intel.com/content/www/us/en/processors/xeon/xeon-phi-detail.html
The main conceptual differences to GPUs is the programming model. Intel claims it is simpler, because the cores are x86 cores. So, it is kind of certain, that you will be able to implement your image and video processing algorithm on Xeon Phi. But to really accelerate your application, you will have to parallelize it. And typically, that also means to consider the details of your target architecture to achieve a good performance (same is true for GPUs). How Xeon Phi competes against GPUs is still an open question at the time being. At least I'm not aware of any neutral comparisons.
Thank you for your clear answer Lars Bauer. I'll start my tests on Xeon Phi. I'll take you informed by results
An other point to consider is that some hardwired functions from the graphics pipeline are available in GPU like the bilinear interpolation in the texture cache. Also in a near futur, all the graphics pipeline will be exposed in compute mode making available some functions usable for Image Processing. This can make the difference with Xeon Phi!
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs