Software rendering - any time soon?

[joshuaklint]

I am interested in hearing about whether multicore CPUs in the next 3-5 years will be capable of running software renderers that compete with the GPU. I don't mean integrated graphics running DX and OpenGL, I mean custom software rasterizers with textures stored in system memory and complete control by the developer. I suppose you would run a "vertex program" on one core, and a "fragment program" on another.

I believe the GPU was a fundamentally flawed (though necessary) approach to graphics. We're programming the GPU more than ever now, and it is frustrating to work with a piece of hardware that is at heart still a fixed function parallel processor. Then throw in all the driver and hardware incompatibilities, the frequent API revisions, and the recent debacle with DX9/10/10.1, and it is enough to make me wish for an alternative. Let's face it, the vendors have utterly failed in their attempts to provide programmable GPUs. We spend the majority of our time testing different cards and inventing workarounds rather than programming graphics.

The annoyances of the GPU and the potential for a return to software renderers in the near future makes me think it might be a good time to hold off on development and just do research for a couple of years. There is a driving force that will push developers back to CPU graphics, and I would like to know whether Intel is pursuing this angle.

[postaquestion]

I am interested in hearing about whether multicore CPUs in the next 3-5 years will be capable of running software renderers that compete with the GPU. I believe the GPU was a fundamentally flawed (though necessary) approach to graphics. We're programming the GPU more than ever now, and it is frustrating to work with a piece of hardware that is at heart still a fixed function parallel processor.

The annoyances of the GPU and the potential for a return to software renderers in the near future makes me think it might be a good time to hold off on development and just do research for a couple of years. There is a driving force that will push developers back to CPU graphics, and I would like to know whether Intel is pursuing this angle.

[Chuck_De_Sylva]

Long term, Intel's goals are 3 fold. We intend on keeping with the integrated graphics for the mid-tier to low end. And then with a discrete solution to follow next year (Larrabee). There is also a move to move some of this functionality closer to the CPU in 2-3 years time.

I would expect you to see GPGPU programming features to continue to evolve
over the next few years. There is already a push within the industry to come
up with programming models that treat CPUs and GPUs as "compute units". So you will see more of this in the years to come.

So, to answer your question, Intel is going to try to accommodate both programming on the GPU and CPU where it can.