The Open Graphics Project is dedicated to producing open-architecture graphics hardware that is friendly to free and open source operating systems like Linux and BSD. Yesterday morning, they released schematics for OGD1 for public review and critique. OGD1 is an FPGA-based development and prototyping platform that they decided to turn into a commercial product to raise funds. Check out an article on KernelTrap. The release of these schematics was accompanied by a discussion about how to price the OGD1 to maximize fund-raising while keeping it accessible to hobbyists; KernelTrap has another article about that as well.
no matter what it is.
a graphic card with full specs is one of the things that is stopping opensource and particularly x.org from developing xgl-like things. Is not about filosofical ideas, there’re lot of new things to be designed and you need full specs and openness to work, you can’t run being blind.
If I can afford it, I will buy one, even if I don’t use it. It’s the first time someone decides to do things like this (really open hardware!) and I’m going to do what I can to make it succesful.
You’ve freedom to do nothing and let nvidia and ati sell you closed hardware with no hope of open source drivers or even drm-free hardware. But don’t forget to never bitch again about the lack of open hardware.
Edited 2006-03-02 14:08
no matter what it is.
I don’t know how this falembait is supposed to be comical, but I have to agree with your decision anyway. Unfortunately the lack of PCI Express is just too painful for this otherwise awsome product
Edit: Ignore the above; it looks like they have clear plans for using PCIe in the final product
Edited 2006-03-02 14:41
I will buy it. I need open 3D acceleration for FreeBSD.
Even if it’s just to help them get them going.
The Open Graphics Project link to http://www.opengraphics.com is broken (missing http:// ).
the first link should be http://opengraphics.org/ but it seems to be unavailable right now anyway.
If you read the proposal on their site it says that PCI-Express will be forthcoming after the project is shipping and supporting itself. Presumably because there is more AGP systems out there at the moment.
I saw a comment about buying it for the 3D acceleration. Don’t do that yet. This board is the OGD, a FPGA based board with a PCI controller, some RAM, a programmable gate array, and video output.
It won’t actually _do_ anything.
Not until someone designs some circuits for the FPGA, and some software to drive it. If you’d like to get in on that, then buying an OGD would be a good idea.
😀
I suppose if you told us WHY you wouldn’t buy it, that would help us. If you reason were “I’m not interested in this sort of thing” or “it’s out of my price range”, then you wouldn’t bother to comment. So it’s got to be something else.
The correct link is “http://www.opengraphics.org“. Don’t forget the www at the beginning.
The product that is being designed now is the Open Graphics *Development* board (“OGD”), based on reprogrammable circuits, FPGAs. (You prototype the circuits in FPGA and when you’re satisfied you turn it into regular chips, ASICs.)
The final graphics card (“OGC”?) will be populated by standard ASICs. They’re cheap (for large series), more power efficient, and may run at higher frequencies. ASIC drawbacks are the high initial production costs, and that you can’t reprogram the ASIC, so you’re stuck with what the ASIC does/does not. (including bad design and bugs) Hence the initial prototyping stage.
The *development* board will not be cheap, when compared to common graphics card. That’s not its purpose. It’s a tool, primarily being the prototyping platform for the current and future developers of the Open Graphics project, but also for people with an interest in FPGA: electrical engineers, people in computer science, etc.
If you’re not one of these, or you don’t know what an FPGA is, you should probably not get the development board (as it will be pretty much useless to you), but instead consider simply donating. (If/when the project opens up for donations. I don’t know if they will. I think they should.)
Alot is going to depend on the performance of these boards. As much as I want a card that has open sourced drivers for all of the obvious reasons, if this thing performs like a TNT1 it simply won’t be worth it.
Apparently there’s a rumor going around that ATi is going to buy XGI. Which kind of bothers me. Even though XGI said they’d release full specs for 3d(though I hope you’re not holding your breath, I’m not) they haven’t done so yet and if ATi does actually purchase them up it simply won’t happen at that point.
Now if these cards had similar performance to my Radeon 9000 and Firegl 8800s on the DRI drivers, then I’d be yet another buying customer. Based on partial specs or not, performance is an important thing to consider.
As long as it’d be fast enough to handle a full 3d desktop, I’d put one in my grandparent’s linux box but mine wouldn’t see one.
> As long as it’d be fast enough to handle a full 3d
> desktop, I’d put one in my grandparent’s linux box but
> mine wouldn’t see one.
But isn’t that enough, for the beginning? A lot of people are not interested in gaming, so performance in the desktop area is enough. A good example is office boxes, these need little more than standard office tools which are already available for linux, so there are few obstacles for linux there. But even these could benefit from an accelerated desktop which can bring advantages in usability.
I have only read a part of this so far but my 1st impression is a of very poor project in both senses, poorly thought out and poor in $. First it seems they want the OS HW community to act as a VC to develop these boards with no guarantee they will work. Typically when you buy a board from 1 of 10 or so well established vendors you are getting an incredible bargain in how little engineering you have to do other than fill the FPGA and understand the board capabilities. Trying to rummage $1M to do an ASIC is frankly rediculous and I wouldn’t consider it seriously. There are reasons why ATI/nVidia own the graphics market, deep deep pockets, good performance and unfortunately no open drivers.
Some of the features on this board are non trivial to develop, PCI at 32/33 is a pain in the neck & too slow anyway with few open cores for it. PCI 64/66 is completely out of the question with out alot of lab resources, you can buy cores for (guessing) $20K or so. This has been a bugbear of mine for the last decade, PCI even PCI 64/66 is ancient, dating back to early 90s and yet there are few low cost cores around for the avg EE to experiment with. The only practical way to get PCI to work on FPGA at lower development cost is to use a PCI ASIC like PCX and leave the FPGA for the algorithm grunt work.
Visit comp.arch.fpga and lurk, read the faqs and find the listing of current FPGA boards out there, Broaddown and Digilent post alot. The commercial boards are almost all Xilinx, then Altera with Lattice far behind. These boards can be had for $50 to $50K, but there are alot of really really good boards out there for <$1K with many features suitable for doing algorithm development. The EDA SW from all these vendors is free to use for FPGAs far in excess of amatour work. The tools only go to charges for the higher end parts, so guess what, most of the low cost boards aim at the middle and lower parts for budget prices and leave the high end boards with high end prices to more specialised board vendors like Nallatech.
Almost all the top selling boards are Xilinx based and the hobyist ones include VGA output at modest resolution only since they have cheap video output section. The PCI boards automatically cost $500-$5K to pay back the PCI core license fee built in from Xilinx or Altera.
I would consider some FPGA vendors that have a PCI core built in as ASIC hardware right on the FPGA to be a good alternative to using PCX ASIC chip. Still I really only want to deal with Xilinx, maybe Altera for computing work.
I really don’t know what to advise the community to do since the market is so tiny, dev costs cannot be recovered with out guaging the early buyers. The edu vendors are doing a great job already, just set your expectations right.
I think you probably didn’t look very hard into the project.
1) They’re not trying to rummage $1m right now for the ASIC. They’re planning on selling FPGAs until they can decide whether an ASIC is a profitable proposition.
2) They already have a functional PCI core running on a Lattice FPGA.
3) The people behind the project work at a company that makes graphics chips. I’d wager they at least know what they’re doing regarding graphics.
Rayiner
I read enough to see that the $1M ASIC is a far off proposition and that they did seem to know something about graphics. I didn’t get the impression that the PCI core was complete at all.
Getting functional PCI core running is non trivial, you can get them up and running so so on your own system with incomplete features but the tricky part comes from having the board work in everyone else’s system with other boards plugged in to mess up the timing. Search for PCI discussion on comp.arch.fpga yourself. I have also worked with folks that did PCI cores going back to the very first one for DEC.
The Lattice situation maybe different from Xilinx, Lattice maybe easier since they came in later, might already be supported onboard the FPGA. Which still brings me back to choosing a vendor that is tiny compared to Xilinx and that leaves most of my previous comments intact.
If I were interested in exploring graphics development I would look for a board that has already shipped & developed commercially with tools and support. But for graphics I haven’t really seen anything that looks right, if its affordable, the output is too low rez and no PCI etc.
Also these boards don’t seem to stay around very long either which makes long term development a problem. The FPGAs are really moving along very quickly these days hence the dependancy on commercial tools.
There was an interesting FPGA graphics board done in Germany (U Mannheim?) some time back, even got mentioned her. They showed that FPGA graphics done in hardware can be competetive to nVidia for some things by being smarter about the algorithm, lost the link though.
I’m an ASIC designer, and I have developed 64-bit PCI cores before. For OGP, I have developed a 32-bit PCI core. I had it working well in simulation as a target. At the moment, it’s somewhat broken in that I haven’t quite finished adding the bus mastering (DMA) feature.
Getting the core to run at 133MHz won’t be all that difficult. In fact, I did some early synthesis on it and figured out where the long combinatorial paths were, mostly in the address incrementer, so I played some tricks to break that up into steps.
If we decide we really want 64-bit, we can add that capability later. It’s not THAT hard. (Like I say, I’ve done it before.)
I found a PCI-X core available for sale for $16000. The OGP one will be free. I think that’ll attract some attention too.
Glad to hear you have done ASIC PCI work, but FPGAs are so much slower internally than ASICs. From what I know about Xilinx PCI cores, I have only seen 64/66 and lesser, I have not seen anything at 133MHz.
I suggest in your address incrementer maybe use gray counters to help timing, they get used lots in FIFO buffers, makes async timing so much easier. Even long counters are not that difficult, its the combination of too much logic within 1 pipeline that breaks timing.
BTW if you want to raise serious $ for your project and if your PCI skills are good enough to ship working cores on a high end graphics card, why not sell those cores as IP product to other FPGA developers. Better still do this in the Xilinx market and sell your board in a more general format. Don’t give away something that might be worth $16K then beg for money elsewhere, just seems odd.
Good luck with project anyway.
I have considered making the PCI core GPL rather than LGPL. This way, if you use it under GPL, you have to open source your entire design. If you don’t want that, you can buy a commercial license. I’m pondering it, and I have mentioned it on OGML.
I did a PCI 64/66 design on an ASIC back in 2000. This ASIC was slower than the FPGA we’re using here. On top of that, I’ve gotten better as a chip designer. I think 133MHz is probably doable. If not, it’ll make it just fine in the ASIC, where that speed is important for the PCIe-to-PCIX bridge.
It’s important to minimize your internal combinatorial delays. Often a harder problem is keeping your external I/O’s fast. If it all possible, always use the register in the I/O buffer. The Opencores PCI design separates the PCI target and master logic and then multiplexes that together at the I/O buffers (it looks like to me from the diagrams). I’ve decided to merge the master and target state machines so that they implicitly share the I/O’s, so we can always use the registers in the I/O buffers for output. I’ve played other tricks to ensure that inputs are handled well also.
http://nature.atari.org/supervidel.html
http://nature.atari.org/news.html
Look at : http://nature.atari.org/Pics/SuperVidel_Lite_2005-12-11.gif
Kochise