The goal was to publish source code to a GPU that is register compatible with the late 90’s era Number Nine “Ticket To Ride IV” GPU. Although the project didn’t meet its funding goal, the person behind it later published the code on github.
Despite the fact that this is an older design, it has lots of stuff that is worth studying. It’s interesting to compare this design to the VideoCore GPU that I walked through in a previous post. While there are some fundamental differences, there are surprising number of functions that are similar, which shows how modern GPUs evolved from earlier ones.
A walkthrough of the GPLGPU as well as some history and backstory of the Number Nine “Ticket To Ride IV” GPU.
Wow. Reading through Anandtech’s review of that chip was an interesting trip down memory lane and the state of graphics technology in the late 90’s
http://www.anandtech.com/show/192
Thanks for sharing that review. Interesting card, and you can see Anand’s writing has come a long way since.
Has anyone seized upon the GPLGPU for an open source graphics accelerator? Now its available as open source, is it useful to the open hardware guys?
Dirge,
While that’s a good question, I would think the answer is a definite no.
While NVidia and AMD are both hardware vendors, they are both fabless, they personally are only responsible for the “IP” in the chips. If they adopted GPL for their hardware designs then every buyer would be legally entitled to a copy. Eventually an enterprising soul would take advantage of their ability to create a perfect clone with very little R&D of their own, order a few million units at a fab (maybe even the same fab), and then start to resell them, undercutting the original and possibly being sold as fakes.
A BSD license might have been a bit more compelling for them IMHO, but even then I don’t think there’s that much the big guys can take from this legacy design that they don’t already have anyways. Where’s the selling point?
Edited 2016-07-31 11:33 UTC
I didn’t know that Nvidia and AMD were into open hardware…
However I doubt even the real open hardware types are too interested, there are more modern GPU designs that are open. While not complete in any way I’d think it would be easier to build a real open GPU from those systems than trying to modify something really old.
Megol,
Theoretically open GPUs might be an interesting area for independent innovation. I actually think indy GPU designs could fair pretty well – if only we could build them! However if it costs a million bucks to build the first prototype in silicon, then it’s kind of out of reach for an indy developer.
A programmable FPGA is probably the only viable option for indy GPUs. The marginal unit costs will be more expensive and it’s going to perform worse than etched transistors directly in silicon, but at least the upfront fabrication costs are already paid.
If they were to be done in any quantity eAsic might be an option but other wise yeah… FPGA all the way.
Some of Altera’s next generation high end FPGAs have integrated HBM2 memory. Which would make sense for use in GPU designs… I mean it might acutally make it semi-practical. I doubt it would scratch even the lowest end current GPU but it might be enough to say render COMPIZ on Linux with recent extensions and/or play basic OpenGL/Vulkan games if someone were to implement it.
The Stratics 10 MX is supposed to have a quad core A53 arm built in also… which I could imagine being used as a soft command scheduler, or failing that a management engine for a big array of simple graphics cores. And if people didn’t like that… the RISC-V could be used which is what Nvidia is using anyway.
cb88,
I’d be interested to try realtime raytracing. It’s been done on clusters with low resolution but a $5k FPGA cluster today ought to handle it nicely in real time HD.
On the other hand polygon driven GPUs have come such a long way (ie using environment maps to fool our brain with fake reflections) that the additional accuracy provided by raytracing might not even get noticed in most scenes except those that are contrived to show off ray tracing.
No serious company will ever touch anything with GPL v3 mark on it, especially hardware blocks. Also none of “DIY” “hackers” working on projects made by slapping 10 raspberry pi boards together with glue and cheap paper tape, because their PHP and javascript GPIO bit banging code can’t keep up with hardware block – they have no knowledge to even touch that kind of project. It’s similar to making open source hardware space shuttle – too large scale project and if someone knows how to develop such project, they will work for one of the big companies in the industry instead.
agentj,
Interesting example, but I don’t think there is anything they couldn’t do assuming they had the resources. Elon Musk is a perfect example of someone who was able to work on space innovation as an outsider just because of his wealth. If you look at the economic factors, that’s really the biggest impediment holding the independent groups back.
Edited 2016-08-02 13:55 UTC