SoylentNews is people
SoylentNews
After a failed Kickstarter back in 2013, the GPLGPU project has delivered a first version of its free (GPL v3) GPU.
As explicated by Hackaday, this FPGA based GPU does 3D over VGA and works with a standard PCI interface. This is not exactly a powerhouse, but it's a starting point for truly free GPU solutions, and an exciting news for everyone in the FLOSS community since truly free GPU are quite rare.
You can read the official announcement here.
This discussion has been archived.
No new comments can be posted.
First Release of the GPLGPU - A Truly Free GPU
|
Log In/Create an Account
| Top
| 35 comments
| Search Discussion
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
reintarnation:
Coming back to life as a hillbilly.
(Score: 1) by tftp on Wednesday August 20 2014, @06:02AM
I understood cafebabe as saying that the only thing you'd have to do to make this a viable GPU is to put the design with minimal changes on an ASIC.
I didn't get that impression. And perhaps it's not even possible/practical. Today's PCI interface allows you to make an ASIC, but it has to be on an add-on card for an old desktop computer, as parallel PCI busses are dead as a doornail.
The PCIe is more interesting, and it opens some ways to use this ASIC as a component in single-board computers... but the project does not have PCIe. Not that it's difficult to add, since PCI/32 and PCIe cores are (IIRC) compatible from the application end, but they require MGTs on the FPGA - and that costs dollars and watts. I don't know what PCIe cores are available for ASICs, as LogiCore netlists are built with Xilinx primitives.
Another bus that the developer is going to add is AXI [xilinx.com]. However this bus is specifically intended for use within an FPGA, in a soft SoC. I am not sure how would one use it as a standalone ASIC. Altera's Avalon is used with their soft CPUs (Nios.)
This explains why so many modern chipsets are integrating a reasonably good GPU. You do not want to mess with yet another bus in a low power design, as all this handholding and bit-shuffling costs you in size and in power. A GPU that is connected to the ARM via local AXI, and all that is implemented in one ASIC, along with GPIO and whatever else you need, is the ideal solution. Make it in millions, and it allows someone else to sell budget tablets for less than $100.
The GPL GPU as a standalone GPU chip is not very likely for these reasons. At very least it ought to have PCIe, otherwise you'd have to route full width busses, and that's a task that not even every PCB layout tool can do well. However that chip would have to compete with integrated solutions that are already working in every smartphone and every tablet on this planet.
As a chip it could be handy under control of Atmel, PIC and TI microcontrollers. Most of them do not have a GPU inside, and it is all but impossible to run a large LCD with bit-banging. However those MCUs do not have PCI to talk to this GPU design. One would need high speed SPI or something similar. Arduino runs the Ethernet interface over SPI just fine. Perhaps that's one of good niches for this GPU, as adding SPI and a local DRAM connection is trivial. An accelerated GPU does not require a huge bus bandwidth, and there are many industrial applications that need to draw colored boxes and lines.
An integrated CPU, GPU and GPIO, all under GPL, and manufactured as an ASIC, could be an interesting idea... but patents (?) on CPU command sets or some implementation methods may make that difficult. If successful, that would be an equivalent of the chipset (TI's OMAP, etc.) that drives a smartphone. It would be free, but it would require extreme investments ($10M) to implement.