Unknowns on market shares, Larrabee
What's harder to predict is whether the new chips will drive shifts in graphics market share for AMD and Nvidia. Both are designed in 40nm TSMC processes which are production constrained, said Peddie.
Whenever TSMC's capacity constraints lift, Nvidia is likely to gain unit share over AMD—at least from the supercomputer and workstation sector, Peddie said. That's due to Fermi's unique built-in features and the relative sophistication of its software for that market, he added.
Nvidia's Cuda programming environment for running general purpose programs on its graphics processors includes a suite of tools for programming in C, Fortran and specialty environments. AMD has focused on supporting open source tools such as OpenCL which are less mature, Peddie said.
Intel is still standing on the sidelines in the high-end graphics market, having cancelled in December plans to market its initial Larrabee processor. It reportedly used 16 x86 cores and other specialty hardware to handle both graphics and supercomputing jobs.
Intel said it will provide more details on its plans for the Larrabee architecture later this year. Given the size of the x86 cores, the chip would not have been competitive with the latest AMD and Nvidia Fermi chips, Peddie said.
"When Intel gets to 22nm, the rules could change but they will not roll out anything in [high-end graphics] in 32nm," Peddie said. "That's a given because they can't get the core density to be competitive," he added.