Thursday, November 12, 2009

AMD's Audacious But Risky Fusion

AMD's Fusion strategy -- merging a central processor(s) (CPU) and graphics processor (GPU) onto one silicon chip -- aims, like nuclear fusion, to generate more computing "energy" than heat. Considering that today's graphics cards have more than a gigabyte of dedicated memory, the architectural issues of marrying CPU and GPU on one chip are significant. AMD, losing ground to Intel's Core i5 and i7 processors in performance systems, is making an audacious move to leapfrog Intel in the mainstream notebook and desktop market.

But this move is risky: the plan could hit the rocks with schedule delays or early defects (the first Opteron's were recalled with an erroneous computing defect). The current plan delivers Fusion products in 2011. To succeed with this schedule, AMD has to manage on schedule a raft of complex technology:
  1. Architectures for two new CPU cores and a GPU core
  2. Fusing the CPU cores with the GPU in a single, high-performance chip
  3. Migrating from 45 nm to 32 nm silicon fabrication technology at high yield
  4. Getting the CPU-GPU Fusion processor to work error-free.
When Intel tried to do an architecture change with a silicon process change when introducing the Pentium 4, the results were poor. Intel lost several years of competitive advantage before catching back up with Core 2. That's why Intel, much chagrined, now alternates new architectures with new fabrication processes.

You have to give AMD the tip of the hat for audacity; they really are putting all the chips on the table. But don't be surprised in a year if all this technology change proves too difficult to pull off on schedule.

No comments:

Post a Comment

All comments are moderated.

Note: Only a member of this blog may post a comment.