Building an Open Source GPU

14 points by BracketMaster ↗ HN
I'll take one last stab at this. Founder of Chip11 here, we're on a mission to disrupt the currently siloed Silicon industry by building an ecosystem of open source hardware components.

Our progress so far:

- we're funded

- functional/verified LPDDR4/5 PHY

- functional/verified SERDES PHY

- GPU RTL running some basic 3d application

- CPU RTL running some basic C code

Hoping to meet someone passionate about GPU and GPU driver design.

15 comments

[ 6.3 ms ] story [ 44.6 ms ] thread
Might be useful to mention what help you need. It sounds like you have all of the bases covered already!
We need somebody with GPU driver expertise as well as GPU architecture expertise. The guy we were considering going with for this summer wants to just wrap up his dissertation instead and graduate - which is totally understandable.

The GPU was largely a one man effort by him. He is very talented.

> we're on a mission to disrupt the currently siloed Silicon industry

What's your FLOPS target?

So the initial corporate customers we talked to just want something that plays 1080p videos and is passable at 3D graphics. Currently, the GPGPU hits at around 25GFLOPS peak performance on an ARIA FPGA. We're targeting GF12nm - some back of the envelope calculation put us at 125GFLOPS for actual silicon implementations.
I see, thanks for the info. In case you'll need to implement a Vulkan driver and pass the conformance tests, email me (iamvfx@gmail.com), I'm a long-time C, C++ GPU programmer. (https://github.com/procedural)
Just FYI, procedural has been insulting graphics experts left and right, impersonating other vendors ( they own the https://github.com/d3d12 account but are not affiliated with the D3D team ) and claiming conspiracy theories whenever they are called out on this behavior or asked to stop.
Do you have any sources, or is this just name calling?
See the tone of the sibling comment replying to me, but also, here's their blog:

https://redgpu.medium.com/valve-pays-to-harass-me-or-how-val...

https://redgpu.medium.com/shove-your-amplification-shader-a4...

Here's them falsely claiming they are an official Valve contractor (conversations with Valve contacts in my network denied this): https://redgpu.medium.com/redgpu-is-now-an-official-valve-co...

> Here's them falsely claiming they are an official Valve contractor (conversations with Valve contacts in my network denied this)

I pointed this out to show that you can't verify who is a Valve contractor and who is not, because that's how Valve can get away with stealing Microsoft's intellectual property.

So yes, I am a Valve contractor. (conversations with Valve contacts in my network confirm this)

I do insult most of graphics "experts" when they insult me first, but not the actual experts who actually know computer science. I don't impersonate anyone, d3d12 is a github group for my personal D3D12 code. I don't know what conspiracy theories you're talking about, it's probably your rich imagination.
Again - we're not going for 3070GTX performance levels here. We plan to disrupt industry by being the first production grade open source GPU implementation ever.
I'm a GPU dev (3D/VR) and I did FPGA acceleration in the past, for scientific computing. I can commit to hobby level time due to the other things I am doing. That said, if you need a beta tester and/or help about the GPU compiler, might be able to help ;-). BTW can your design run "simply enough" on F1 instances or U250 cloud instances?