My critical issue list has stayed constant for a while:
Higher precision color formats.Solving the temporary problem.Flexible dependent texture reads.Virtualized graphics address space.
I am confident that we will be getting 64 bit color formats in some cards next year.
It doesn’t seem like there is a wonderful solution to the temporary problem, but next gen cards are going to have enough temporaries that dropping back to full screen buffers (possibly with software tiling or striping) for the overflow cases may be reasonable.
The dependent texture reads are there in GF3 if you degenerate to simple rendering. I expect next generation hardware to have completely flexible implementations.
Virtualizing the graphics address space seems to be gated on Microsoft’s unwillingness to properly fix the AGP interface and management. I think they should go ahead and do it even if they can only use 66mhz PCI reads for it, but that is clearly suboptimal.
Carmack made a large .plan update a year ago in which he explains the rationale for 64 color bits in detail.