GPU or PPUs on AMD hyper-transport bus?


Jul 7, 2006

Put a GPU or PPU in the second socket of a 2socket AMD board?

Via, nVidia, IBM, motorolla able to make a CPU for AMD socket?

Finally an open socket again, and a much more useful one at that. The potential for more options for consumers would be great! It would be nice to choose between CPUs and GPUs from more then 2 companies instead of choosing between the 10billion different crippled versions of the same product that Intel, AMD, ATI, and nVidia have been putting out for their main product lines since they have so little competition (I hate marketing jerks that do stuff like that).

A GPU in the second socket could probably even be configured to use it's own stick of "system" ram. Probably won't replace high-end GFX cards anytime soon, but mid-range GFX cards using a socket should have comperable performance to their card-based brothers while costing a lot less (even factoring in additional system ram), and drawing less power (creating less heat) and being easier to put beefy HSFs on them, control their voltage and frequency, and OC them.

With this open socket it's possible that Via could make a chipset that could run AMD or Via CPUs while using a graphics "co-processor" from ATI or nVidia. Custom heatsyncs could be designed for one socket and used for CPUs or GPUs. If some kind of "default" location, or releative location, for the sockets was adopted both chips could be cooled by the same heatsync and ducted cases would be easy to design.

It's a stop-over step for trying to integrate the GPU into the CPU... maybe not a such a good thing, but it would give people a reason to buy a new CPU every 8 months again, something I'm sure AMD and Intel both miss. But I hope open socket designs stay around, and specialized co-processor chips become more mainstream. Specialized co-processors are the only place we're going to see major performance increases for the forseable future (for example: a 500mhz GPU can out-render a 3ghz CPU). If you co a lot of audio/video editing you could get a specialized chip that can do it in a fraction of the time. Even an FPGA at 100mhz can accomplish real-time tasks a piece of software running in an operating system could never dream of (i've got a 53mhz one sitting next to me that can route 2ghz FC traffic on 5 seperate channels simultaneously, no PC can do that). You could then load "programs" into your fpga whenever you need them. Need a special Physics processor for a simulation? Video encoder for editing? AI for a game? All of these are possibilities. An FPGA in a second CPU slot with a fast link (like hyper-transport) directly to the system... you could do anything with it.

Open standards are our friend!


Aug 29, 2009
Are these shipping (as of 2009) yet? I see the FPGAs from XtremeData, Inc. and DRC - where are the GPUs?