Parallella Community

Since the 64 core is not available, there is a competitive solution from NVIDIA with 192 cores and 365GFlops for $192.
It's programmable using CUDA (NVIDIA) and is available now....using the Tegra K1 processor.
http://nvidianews.nvidia.com/News/NVIDI ... s-ad8.aspx

....enjoy...

One very big difference is that a "CUDA core" is just a FPU pipeline. This means there isnt any branching in a CUDA core but rather in the parent SMX processor. the SMX processor controls all the CUDA cores and has 64KB of memory to share. why does that matter? well, do you like branching or reading/writing data? because you wont be doing it very quickly.

here is the whitepaper about the Kepler architecture: http://www.nvidia.com/content/PDF/keple ... epaper.pdf

here is the SMX:

SMX_processor.png (38.57 KiB) Viewed 20450 times

if you are primarily doing rendering then having tons of floating point pipelines is awesome. however, if you are making actual programs, you will need that branching, storing, loading and integer math which is not going to work well on this GPU.

You are free to suggest a different benchmark with which we could compare those two platforms.

HB