Luke Kenneth Casson Leighton
lkcl at lkcl.net
Wed Feb 19 19:03:57 GMT 2020
On Wed, Feb 19, 2020 at 6:55 PM Scheming Pony
<scheming-pony at protonmail.com> wrote:
> > leaving the numbers aside: you're describing "single instruction,
> > multiple data" but gone mad. it's been recognised in the industry -
> > thanks to the billions spent - that SIMT is unmanageable at the
> > software level. Mitch Alsup was only a consultant on the Samsung GPU
> > project, and his warnings not to implement SIMT were not heeded.
> Research: Connection Machine (*Lisp and C*), SIMT done right--*Feyman* was behind that bad boy. If you scale this thing, and interconnect with hypercube topology, you might get a connection machine emerging.
> But, OK given your anti-SMT logic why not throw away Vulkan, the descendent of OpenGL, meant for SIMD GPU designs?
because Vulkan has strong industry support and is where the "adapters"
converge. there's an OpenGL "adapter", a DirectX "adapter". it's the
"focal point" of all of the APIs for parallel compute and 3D.
> For the prototype use OpenGL 3.1.
as jacob said, we've already made the decision, 18 months ago, to go
with Vulkan because it's the focal point/
> You have painted yourself back into the original design case for OpenGL anyway. It would run Curv and my Open Inventor 3D Lisp desktop.
... which will run successfully on top of the OpenGL-to-Vulkan adapter.
we simply do not have the engineering resources to do dozens of
different massively-complex software drivers suited and optimised to
every single corner of the 3D and Parallel Compute industry.
*later* we can *consider* that, once we have multi-million-dollar backing.
More information about the libre-riscv-dev