Open Source Graphic Card

[Nicram]

A company named <a href="http://www.techsource.com/" target="_blank">Tech Source</a> has steped up for creating an open source 2D/3D graphics card. The project has an open <a href="http://lists.duskglow.com/mailman/listinfo/open-graphics" target="_blank">maling list</a> and a spec proposal for the final card can be downloaded <a href="http://open-graphics.duskglow.com/openspec.pdf" target="_blank">here</a>. This project is a ground breaking effort for creating true open hardware, and is great chance for all with a "2D/3D programmer dreams" to see how a modern graphics card works.

[spadnos]

Cool!

I was hoping that this wouldn't be killed by a PHB attack.

I'm not sure they've decided to do it yet - it seems as though Timothy Miller (the guy at TechSource who first mentioned the idea on the Linux Kernel mailing list) is still in an information gathering stage.

Excellent that it hasn't died though.

- Steve

[rylan]

You don't want to know how much their current cards cost.

[Fluff]

Well with the words xilinx.. Don't spell cheap. They are FPGA's aren't they.

[borat]

So how does this work, open source community has some sort of input into the finalc product, and then does Tech Source sell the fruits of their efforts at its own profits?or does it charge cost for the final product?Or does it repatriate profits back to the open source people who put the work into it?

[Nicram]

borat... all benefits then will come into company, so the company will got better equipment for realasing new, better versions

About the price...
First version of articles about this sayd about 100$ per card...

[Fluff]

Or possibly NVidia or ATi (even Matrox) will try and poach the best developers and the project will never get completed.

[rylan]

I'll just say don't hold your breath waiting for anything like an 'open-source' graphics card, especially with any reasonable 3D capabilities. A FPGA based graphics card can't come close to competing with a custom asic in terms of performance. There are already industry standard APIs for 3D acceleration (look at the opengl command set and D3D for directx command sets).

[DGhost]

the FPGA idea is an interesting one. sadly, the open source community does not have the sort of resources to even dream of making it effective.

looking at their proposed "specs" are rather disappointing. they are not trying for anything high performance - seriously, 2 pixel pipelines @ 200mhz (although I am trying to figure that one out since the Spartan-III runs at 326Mhz), Bilinear interpolation, etc etc. it looks like something along the lines of a cracked out quasi-high performance Riva-128. quad-channel 400mhz DDR? better have a damn good memory controller if you are going that route. they keep mixing ""new"" technology with some really, really ass old things.

they say they want the user to be able to modify the graphics pipeline and add features by writing code for it. nice dream, but how are they going to get around the cost (and liscensing issues) of having to deliver chunk of the Xilinx toolkit with it?

i could see FPGA graphics cards meeting or beating custom asic's for performance *if* it was done correctly. honestly, you probably could do it with something similiar to the architecture that nvidia schemed up for the GeForceFX and newer. you basically have a dedicated asic to handle 2d operations, video decoding/rendering and memory management and then offload chunks of the pixel pipeline onto FPGA's - the less static functions, of course. it would still have to have a high performance ASIC behind it, but since the pixel shading on the GeForceFX and GeForce6 lines or cards is decoupled from the triangle setup and pixel rendering portions of the chip (source), it would be very easy to expand portions of the architecture on demand using FPGA(s).

aannywho. those guys are... well... i dunno wtf they are doing.

[rylan]

Don't forget that the Spartan-III speed spec is the max clock rate, which Xilinx basically determines from an ideal lightly loaded array. Once you start putting real logic in there your realistic max global clock speed drops quickly. sounds like they want to have the core running synch'd to the memory interface. Xilinx has some reference designs for 200MHz DDR interfaces with the Spartan 3. Would be more interesting with the Virtex4 chip, but thats not released yet.

[DGhost]

ahh, alright... i wasn't too positive on the Spartan-III spec's, i saw a spec sheet and was more or less guessing...

[G400SG16mb]

They seem to be serious now about going ahead and producing the thing. TechSource is interesting in that it basically does the same thing as Matrox, but is doing the opposite from Matrox in wanting to go into the consumer maket instead of away from it. This is their way of going about entering it.