Re: [Algorithms] General purpose task parallel threading approach

[Ola O. <ola...@gm...>]

> I can't speak on ultraspec, but neither larabee or GPUs use
> hyperthreading as the main form of threading (or afaik at all) They
> are both data parallel architectures primarily. Additionally,
> hyperthreading is build to fill bubbles in the pipeline and do such
> rather well because they don't have to worry about any sort of context
> switch, as the threading is implemented in hardware and multiple
> states can actually be active at the same time.
>

Well, NVidias current GPUs are fundamentally built on using 'hyperthreading' 
to hide latency, as opposed to keeping large caches. Each core is capable of 
keeping 1024 threads going at a time, which are processed SIMD fashoin 32 at 
a time. So I don't see what you mean by saying they do not use 
hyperthreading?

Larabee is different, and less known, but from their presentations they have 
4 'hyperthreads' per core.
However from their presentations, they intend to use cooperative 
threads/fibers to run their shaders, switching when hitting a texture fetch 
or suchlike to achieve the same latency hiding as other GPUs.

cheers
.ola