Why The Open-Source Graphics Card Failed
Michael Meeuwisse of Project VGA fired off an email to me providing a lengthy email as to why Project VGA is dead.
Among the reasons:
- Only seven people ever tried to order one of the hundreds of thousands of people that visited the web-site. They needed at least 100 orders for a single bath order.
- The graphics card prototype was flawed. It didn't even work! The PCI interface wasn't functioning and there were other problems too.
- There were not enough skilled people that knew about hardware design to design a truly open graphics card.
- AMD and NVIDIA open-source efforts got off.
- Loss of interest in the project.
Below is the email in full from Michael Meeuwisse.
Hi!
I still find it amazing that projectvga comes up from time to time, I get the occasional email asking for more details, source code, etc, and now it found it's way once again on Phoronix. Let me first make an end to any remaining doubters; yes it's dead. The prototype is still laying around somewhere in my room, but I haven't touched it in years.
There's a bunch of reasons for this.
First;
> ...making it really not viable for anything but enthusiasts/developers wishing to tinker...
The website hit around the time of fosdem '08 and the general '15 minutes of fame' a few hundred thousand visitors, the guy in charge of the webserver had to upgrade due to traffic, etc. To keep things going, we needed some bootstrap money (a problem not unknown to OGP either, but I'll get to that) so we jotted down a target date and basically said "whoever wants one, speak up, if there's enough people, we can get things rolling". There wasn't. In total of those hundreds of thousands people, 7 (!) actually wanted one, and a handful more wanted just the PCB to hand solder. With 'enough', the target was around 100. Only 93-ish people short. Not encouraging. These days I would toss it up on kickstarter and might be more lucky, especially after seeing the Raspberry Pi craze lately. But in 2008 it proved to be a too alien concept.
Second;
The prototype was flawed. Badly. The PCI wasn't working at all. The memory interface was too noisy. The data loader for the FPGA was Xilinx's proprietary interface was, combined with open source software, unreliable as hell. Working with the official tools did the trick on that front, but that would mean Windows etc, not very inviting for the few devvers who would actually want to use the card. I did reverse engineer it to the point that I got it working reliably, but it didn't interface decently with other existing open source tools etc. Basically the chips weren't properly supported. These days they are (and support more friendly non-proprietary interfaces) so again I'd say it was a little too much ahead of time. But note that the FPGA 'compiler' of sorts still is mainly closed source windows material last time I checked.
On the hardware buggyness front, that wasn't really surprising. I was a student (still am, but again I'll get to that later), and getting it right in one shot would've been nothing short of a miracle (even if I hadn't been a student). I still think I could've handled this better by recruiting more people from the community somehow, but I was new to this management-angle and failed. Lesson learned.
It's not uncommon to do a few prototype iterations, but without proper funding (see earlier) I had to invest from my own wallet, and being a student and all, enough was enough.
Third;
OGP. The problem of money-bootstrapping and getting people onboard who really knew what they were doing when it comes to hardware design had been partially solved, but everybody was kept in the dark about the how. Turns out the company a few main contributors work for invested time & money, designed the hardware card for (iirc) flight control, and OGP could 'piggyback' on this. All they had to do was write their own firmware. They was also some input in the hw design naturally, and for marketing reasons I'm told there are a few subtle differences between the two end results, but that hardly matters. We weren't told. To this day this pisses me off. Essentially all discussions on the OGP mailing list about design etc were used as input to produce a proprietary card for a company, and once the card was done OGP still had to find the 100 or so 'first buyers' before they could produce their 'own' batch. Worse, the big shots in the project knew about this, agreed on this, and kept it from the rest of us. I for one felt used. A real "Luke, I am your father" moment. Here we were, trying to get open source hardware out in this world, only to find it was a corporate product all along. Maybe I'm overreacting, and I knew I didn't manage particularly well community-wise for projectvga, but this was simply too hard to grok. Too evil. There's plenty of projects with developers supported by companies (just to name a random one; linux) but doing it secretly? No. Just no. I abandoned the idea of OGP entirely after getting to know this.
If memory serves me right this was after I spun off the projectvga project to drive the price down, because I couldn't get my head around the insanely high requirements and equally high cost of the design that was pushed. Hindsight 20/20. It really put a dent in my spirit.
Fourth;
amd/nvidia open source efforts really got off. I remember being floored by the Nouveau team that day at Fosdem '08, and amd (ati back then? doesn't matter) stepped up their game as well. I like to think the open graphics 'movement', and OGP in particular, really did help push for more open source support from them. I don't think projectvga had much influence on it, but these better drivers did put another nail in its coffin. If there ever was a window to get it to market, it was back then, not now.
Fifth;
Interests. As I mentioned earlier, I'm still a student. I got a little fed up with sitting behind my desk all day staring at hardware designs, so I've been doing a cognitive science master as well as a cognitive neuropsychology master. Two masters simultaneously really murder all your spare time, but I love it. In short; I moved from how hardware computers work to how wetware computers (i.e. your brain) work, and ideally I'll someday translate one to the other; true artificial intelligence. And this might need some specialized hardware, who knows... I'll let you know when I get there. ;)
Hope this clarifies things, I really outta update the projectvga website with this rant some day. Regarding OGP (I'm sending this also to phoronix, so sorry for you mailinglist guys who already know this), there are cards out there and as far as I know, they are in use. Which is fantastic! Alas not by me.
For projectvga, I abandoned it entirely after I switched to my masters a few years back (time flies, other things happened in between, not going to bore you with that). I get the occasional email from enthousiasts who want to give the source a fresh look, see if they can use bits and pieces (it did control a screen through the USB debug interface in the end after all) for their own projects, and I send them in the right direction. It's still all available as GPLv3 but god knows in what kind of buggy state. But other than that it's completely dead. I've learned TONS doing it and absolutely loved most of it, but times change.
I've moved from what happens in front of the eye to what happens behind it.
Cheers,
Michael Meeuwisse
PS. The website still works, to my knowledge. http://projectvga.org or http://wacco.mveas.com
PPS. Apologies to anyone offended / not happy me bringing up the 'taboo' of the third point. This isn't personal, but the 'secret' is out and I'm not going to play along doing ssssshhhh.
17 Comments
