Intel Xeon Phi

[komyali]

Can somebody explain this?

 

http://ark.intel.com/products/family/71840

 

http://software.intel.com/sites/default/files/article/383067/is-xeon-phi-right-for-me.pdf

Edited July 12, 2013 by komyali

[Dimitris Tolios]

It is Intel's take on GPGPU/co-processing unit using the PCIe bus.

Think of it as a GPU accelerator, that can do OpenCL, but can also do x86 instructions, almost identically with your regular x86 intel CPU, so companies/researchers etc clients that would turn into a supercomputer solution, won't have to recompile their x86 oriented software to CUDA or OpenCL.

[komyali]

or turn your multi xeon processors to supercomputer Big question is over my head, will it be capable to render together with processors and do you need to fit inside memory on PHI or you can use computer memory? This is something like boost for your processor? They put together CPU and "GPU" on x86. Price is not big if you speed up your process 3 times? and cores on PHI are not same as CPU? I really want to see production render with this!

 

go to 2:22

[Dimitris Tolios]

I believe the Xeon Phi has far less complex processing pipelines than a full-blown Xeon Sandy/Ivy/Haswell CPU. They could not fit that in the Phi's die anyways.

 

But as GPU accelerated rendering has proven, we don't need complex pipelines for efficient and fast raytracing - the contrary.

 

Afaik there is no support in VRay or similar products for Phi - yet.

[komyali]

I am lazy to register on chaos group forum but they are talking about Phi... I hope they will test soon

[komyali]

Yes, Chaos Group is testing Phi, you need to fit in memory but it will be different from classic GPU so you could use vraymesh (you need to fit in GPU)

cores on Xeon and on Phi are not same,60 cores 4x hyper threading, 240 threads, If they do this, process will be 2-3 times faster rendering... good bye farms

Edited July 13, 2013 by komyali