Open Source Graphic Card Project Seeks Experts

An anonymous reader writes "Could this dream of many open source developers and users finally happen? A 100% open sourced graphic card with 3D support? Proper 3D card support for OpenBSD, NetBSD and other minority operating systems? A company named Tech Source will try to make it happen. You can download the preliminary specs for the card here (pdf). The project, though a commercial one, wants to become a true community project and encourages experts and everyone who have good ideas to add to the development process to join the mailing list. You can also sign a petition and tell how much you would be willing to pay for the final product."

Comments

Great!!

[Anonymous Coward]

I've kind of waited for this for years.

In theory other companies might steal the design and build and sell the card on their own, but if the design is community-owned, then that actually works to lower prices...

Anonymous Cow

Waste of time

[jarich]

This has come up before.

Building a good open 2D card? Mabye... I doubt it's really feasible, but have at it. Chase that dream.

But a 3D card? You are going to make a card to run the latest Quake and Doom? Or even release back of the games? Do you realize how much time, how many thousands of man hours go into these cards? The dollar amount for the simulators, the fabs to make the prototypes, etc

This could however, make a great teaching tool.

I take it back... if the card can target elementary 3D and stellar 2D, it could (in a few years) be THE card to own for a commodity Linux box. Target your audience carefully and don't get caught up in the IdSoftware upgrade cycle! :)

Re:Waste of time

[ArbitraryConstant]

Well, because they use an FPGA the card can potentially be reprogrammed to support just enough of OpenGL to do what Quartz Extreme does.

When you think about it, Quartz Extreme only needs to handle a relatively small number of parallel polygons at basically a constant distance away. That's a much simpler job than millions of triangles at arbitrary angles to each other at varying distances and whatnot.

The job the video card does can potentially be as simple as figuring out which window is exposed in a given area and grabbing pixels from the appropriate frame buffer. OpenGL is a good deal more complicated than that, but since both the driver and the FPGA are under our control, I would think it would be possible.

Re:Waste of time

[Sentry21]

The issue I see is that these interfaces don't need to be 3D accelerated - because they're not 3D. Why couldn't 2D acceleration accomplish the same thing? Store the windows in memory as textures and move them around in hardware. This doesn't require 3D itself, it just requires hardware compositing and alpha-blending, hardware accelerated windowing, offscreen rendering, z-buffering, scaling & rotation, and so on.

That being said, 3D provides a lot more possibilities - you could make windows be actual objects that could be moved forward or backwards, stacked up, leaned against each other, and so on. Implement HAVOC physics so I can grab an icon and smash it into my other icons and watch them scatter all over my desktop, or throw it and watch it bounce off the edge of the screen and land in my network drive.

Eventually, all we'll need to do to solve the spyware problem is to use a wallhack and noclip and go bounce that crap to the curb. Sure, we'll have to endure the cries of spyware makers shouting 'lamer!' or 'wallhack' or 'aimbot', but we can just kick them off the network if it comes to that, or /ignore them.

Great Idea

[mhaisley]

This is a really great idea, but it will probably never work, a mailing list will bring way to many points of view.

Really what a project like this needs is the developer to shut out the open source community, until the project is done. If linus had made a large project out of the original kernel, I seriously doubt if it would have ever been completed. This should be kept simple, and then open sourced, only once there is a good code base to build from.

Re:Dupe!

[log2.0]

They are trying to get devolper interest, not announcing the open card. Its a different story. Although I may have missed the story you are referring to? :)

ati & nvidia release old specs?

[Goalie_Ca]

I can understand that this card will never compete with ATI and nvidia which raises the question, is there any reason why ATI can't open source their old graphic cards, such as their 7000 series. Surely that technology is no longer critical to their lead. Sure many of those cards aren't being sold any more, but there are still plenty around and this may open up a niche market so they can produce some as a low-cost device.

Re:ati & nvidia release old specs?

[DaHat]

I'd wager that even the older cards still bare some similarity to the newer ones, enough so that such designs could give a competitor a major head start in designing future cards. Opening up their plans is nice in theory, but in practice... would almost certainly come back to bite them in the rear.

Re:RTFA/RTFWS/RTFE!

[Tough Love]

If you'd read-up on this subject, you'd have seen that these folk *do* know their hardware. They are also not being overly ambitious. While they expect to be able to develop a card which has 3D accelleration for desktop applications, they make no bold claims about gaming.

Falling anywhere short of, say, OpenGL 1.4 support would make it pretty much useless. In other words, it doesn't have to have pixel shaders, but it has to have good, filtered texture mapping, lighting, alpha, quite a bag of stuff. The Spartan 3 (not III as the tech spec suggests) has 1.5 million gates and 384 MHz, which ought to be enough for a decent 3D core, with one catch: it's got 32 18x18 multipliers, no dividers. Don't even think about floating point, obviously, but without dividers, perspective interpolation is going to be pretty tough. Without perspective interpolation... well, think "1970's".

I just hope there's a standard way of getting around this. Any hardware hacks out there?

get it out quickly and create a framework

[jeif1k]

Some advice:

• Get the hardware out quickly; if you wait too long, it will be obsolete before you ship.
  
• Create a basic development platform (gcc, loader, etc.) and a basic framework with at least a little bit of useful functionality (2D acceleration, minimal 3D); it can be quite incomplete, but it should make it easy for contributors to add functionality one small piece at a time.
  
• You can charge a little more than a comparable regular graphics card, but not a lot more. If this becomes a premium custom hardware product, it's dead on arrival.
  

Re:Yay! It has an FPGA on it.

[eigendude]

Well, it seems that Xilinx and other big FPGA vendors have already thought of making cards on which you may try to create your own GPU.

Boards such as the Multimedia Board http://www.xilinx.com/products/boards/multimedia/ contain everything you would need. Not cheap though...

They have not put the whole thing on a PCI card, probably because it's even more fun to integrate a CPU core and build the whole system-on-chip on the FPGA while at it.

Cheers!

Re:RTFA/RTFWS/RTFE!

[SSpade]

Odds are that your CPU doesn't have a divider on it either.

Google for Newton-Raphson.

Fast hardware dividers are big and expensive - somewhat more expensive than a multiplier. But if you have a multiplier and you're not too concerned about performance, or are happy to tradeoff precision for performance, then you can do division using your multiplier, a small seed ROM and a microcode engine.

Trend: Graphics Cards become General Purpose Cards

[macmurph]

I've heard that 3D cards of today are exponentially increasing in number of transistors. It's been said that the problem of displaying 3D is "embarrassingly parallel". Hence, the performance of these cards far outstrip the CPU for parallel processing.

Some of the thoughts expressed by experts are that 3D cards may become general purpose parallel computing cards.

If it weren't for bottlenecks in the AGP bus, it would be possible to use 3D cards of today for more general purpose computing (I'm fuzzy on what the actual hold ups are here...timing issues?).

There have been Slashdot discussions about using the graphics card for audio processing, because audio is usually less than a 32 bit stream. The problem is that audio and often general purpose computing have "real time" requirements.

Also, make sure your open source card supports ARB_fragment!

What about this???

[g_braad]

http://www.icculus.org/manticore/

Manticore already exists for some time and it is also what they call Open Hardware. If they could work together, this could result into a good implementation for a Linux/Un*x hardware design.