Linux and desktop graphics [LWN.net]

Linux and desktop graphics

Posted Sep 4, 2005 0:31 UTC (Sun) by njhurst (guest, #6022)
Parent article: Linux and desktop graphics

"run on graphical adaptors which, in terms of processor performance, far outclass the central processor they serve. "

Is this really true? GPUs are certainly faster at doing certain classes of operations (such as drawing textured polygons), but are they in general superior to CPUs? A simple test would be to implement a CPU emulator on top of a GPU and a GPU emulator on top of a CPU. I suspect that for general purpose computing a CPU is a better compromise.

Not that this is really relevant to the thrust of the argument :)

(Log in to post comments)

Linux and desktop graphics

Posted Sep 8, 2005 12:38 UTC (Thu) by csamuel (✭ supporter ✭, #2624) [Link]

There's a group of folks who are starting to use graphics cards for
general purpose (HPC) programming because of their massive memory
bandwidth and floating point performance. Of course you need to be able
to frame your problem in terms of polygons, etc, that the cards
understand, but there are people doing it for all sorts of things
(including databases).

Check out http://www.gpgpu.org/ for a quick intro!

Linux and desktop graphics

Posted Sep 10, 2005 9:32 UTC (Sat) by renox (guest, #23785) [Link]

And you need also a problem where you don't need high precision FP operations: AFAIK currently the GPU don't provide 64 floating point numbers: when NVidia talk about 64bit, it means 4 component times 16bit FP per component..

So not only you need algorithms which are well suited to the architecture of the GPU, you need also to be very careful about the precision (as a reminder Intel provides upto 80bit FP operations), so color me not very impressed by the hype about doing everything on GPUs.

GPUs, precision and general purpose computing

Posted Sep 11, 2005 10:05 UTC (Sun) by csamuel (✭ supporter ✭, #2624) [Link]

People are working on it though..

Accelerating Double-Precision FEM Simulations with GPUs

This paper by Dominik Göddeke, Robert Strzodka and Stefan Turek describes a preliminary algorithm to achieve double precision results by adding a CPU-based defect correction to iterative linear system solvers on the GPU. We demonstrate that identical accuracy as compared to a full CPU double precision solver is possible while still gaining a factor of 2 in speedup compared to a highly tuned cache-aware CPU reference implementation in double precision.

(Accelerating Double Precision FEM Simulations with GPUs. Dominik Göddeke, Robert Strzodka and Stefan Turek. To appear in Proceedings of ASIM 2005 - 18th Symposium on Simulation Technique.)