Skip to contentSkip to conversation

HARDWARE: Open-Source-Friendly Graphics Cards -- Viable?

DISCUSSION Greg Buchholz 2 participants 2 messages Oct 21, 2004 – Oct 22, 2004

lore
Share & tools
lore
#0 lore raw
Quoted 10 lines · Stephen Wille Padnos

I would think that a chip that has a lot of simple functions, but requires the OS to put them together to actually do something, would be great. This would be the UNIX mentality brought to hardware: lots of small components that get strung together in ways their creator(s) never imagined. If there can be a programmable side as well (other than re-burning the FPGA), that would be great.

I guess I would look at this as an opportunity to make a "visual coprocessor", that also has the hardware necessary to output to a monitor (preferably multiple monitors).

This idea is a step in the right direction. To make the project
viable, you might be better off trying to court a slightly different audience (instead of the cost-sensitive/3D-performant market). What if instead, you were selling a highly parallel reprogrammable computing core, which also happened to do graphics? I could see a potentially much bigger and higher profit margin market for a standardized interface from Linux to an FPGA. Image people buying them for headless servers as crypto accellerators. Or as DSP/FFT accellerators (for speech recognition , MPEG compression, or whatever). I'm sure you'd sell a few to grad students writing theses on data flow machines, parallel languages, prime factorization etc. Heck, I'd buy one just because it'd be cool to try and write a 1000 element merge sort in hardware that completed in one or two clock cycles. It's not hard to imaging people using it to speed up emulators like QEMU. Maybe the distributed computing folks (Folding@Home, SETI) would also be interested, since their work is already highly parallelizable. You get the idea.

In my mind, this could be a much better "hook" than the promise of
openess alone.

Here's some people trying to do general purpose computing with current graphics cards.

http://www.gpgpu.org/
http://graphics.stanford.edu/projects/brookgpu/

And be sure to look into existing open hardware projects to see if they have anything to offer.

http://opencores.org/browse.cgi/by_category

Greg Buchholz

#1 lore raw

Greg Buchholz <[email protected]> said on Thu, 21 Oct 2004 10:08:08 -0700:

Quoted 5 lines
Quoted 3 lines · Stephen Wille Padnos

I guess I would look at this as an opportunity to make a "visual coprocessor", that also has the hardware necessary to output to a monitor (preferably multiple monitors).

Quoted 16 lines

This idea is a step in the right direction. To make the project
viable, you might be better off trying to court a slightly different audience (instead of the cost-sensitive/3D-performant market). What if instead, you were selling a highly parallel reprogrammable computing core, which also happened to do graphics? I could see a potentially much bigger and higher profit margin market for a standardized interface from Linux to an FPGA. Image people buying them for headless servers as crypto accellerators. Or as DSP/FFT accellerators (for speech recognition , MPEG compression, or whatever). I'm sure you'd sell a few to grad students writing theses on data flow machines, parallel languages, prime factorization etc. Heck, I'd buy one just because it'd be cool to try and write a 1000 element merge sort in hardware that completed in one or two clock cycles. It's not hard to imaging people using it to speed up emulators like QEMU. Maybe the distributed computing folks (Folding@Home, SETI) would also be interested, since their work is already highly parallelizable. You get the idea.

People would happily pay millions for this.

Tim has probably heard of Grape?

I don't know whether grape uses FPGAs or not, but take a look at the photo down the bottom of this:

http://grape.c.u-tokyo.ac.jp/~fukushig/paper/sc96/sc96.html

I reckon if we could put an accelarator card in each of our 200 new machines, that could be programmed on the fly to do N-body calculations, or then changed to do SPH, or whatever, and we only had to pay $500 per card, and it doubled performace, we'd do it in a second. Half the number of computers, drop energy consumption (and cooling), etc. It'd be great.

Signature · 2 lines
TimC -- http://astronomy.swin.edu.au/staff/tconnors/
A mouse is a device used to focus xterms.