A New Open-Source GPU Comes About

[phoronix]

Phoronix: A New Open-Source GPU Comes About

After writing last month the open-source graphics card is dead and why the open-source graphics card failed, this weekend I received an email that begins with "Open Graphics! Here we go again! As our master thesis work we have implemented a open source graphics accelerator."..

http://www.phoronix.com/vr.php?view=MTExMzE

[spiritofreason]

Aha, I don't think this will really do much for open-source graphics. It appears to be a basic fixed-point, non-programmable pipeline, if I'm not missing a lot by skimming the code. An actually good set of floating point units would probably help out far more.

Kudos for finishing the master's thesis and making something neat, though.

[maiden]

"An actually good set of floating point units would probably help out far more."
Very true, but that will be a lot of more work.. you do what you can with the time you got. This is still better then nothing

[droidhacker]

So they've implemented a GPU using a CPU.... isn't that essentially what you get by mixing LLVMpipe + GMA500?

[maiden]

So they've implemented a GPU using a CPU.... isn't that essentially what you get by mixing LLVMpipe + GMA500?

The pipeline is implemented in hardware, also on in the hardware there is a CPU, the OpenRISC processor that can send instructions to the graphics accelerator. If its still unclear please read this: http://en.wikipedia.org/wiki/Field-p...ble_gate_array

[ssvb]

They've implemented a GPU using a FPGA...

[hoohoo]

Hmmm, I think this GPU will not be getting me 100 FPS in Crysis any time soon.

[smitty3268]

Hmmm, I think this GPU will not be getting me 100 FPS in Crysis any time soon.

They just need some good driver optimizations.

case "crysis.exe":
//display screenshots
break;

[spiritofreason]

They just need some good driver optimizations.

case "crysis.exe":
//display screenshots
break;

That's it: prerendered scenes! FPGAs are just a sea of lookup tables, right? This one will only be a little bigger...

[artivision]

This is close to my proposal but has mistakes. 1)We have a good software-rasterizer(llvm-pipe), just add some 3d-instructions in OpenRisc (like Mips-3d), in order to accelerate the rasterizer and write an llvm-backend. Don't create asic-circuits, they are difficult even for companies like Nvidia, they want to end them by adding more 3d-instructions to the shaders(general cores). 2)Use little big processing, 2-4cores for general computing with 7dmips(20m transistors), and 32-64-128cores for graphics 2,5dmips(1m transistor). Each mini core has 512bit-fmac, that is 64gflops@2ghz per core, also that gives many tflops/watt on latest lithography. 3)Add emulation-instructions like godson-mips, then you will be able to execute "qemu wine" in live speed.