One thing to keep in mind is that AMD is making acquisitions and Intel is selling off parts of the business. So, it's possible that in a year or two, Altera and MobilEye will come off completely from Intel revenue.
As far as Gaudi line of accelerators, It may just never amount to anything. IIRC, Intel just downgraded its forecast for full year revenue from 500m to 300m(?). Apparently, the programming model / tools are quite non-existent. So the same gap that exists between CUDA and ROCm, exists between ROCm and Gaudi programming.
There is another programming framework that Intel has worked on, for Ponte Vecchio, which also may continue to Falcon Shores, but it does not work on Gaudi. So in the end, this Gaudi and the acquisition of the company that made it may turn out to be complete write-off. Only keeping some staff and moving them over to Falcon Shores.
Falcon Shores, in a year, at the end of 2025, will be where AMD was at the end of 2023 from market penetration. Or maybe even worse, because AMD had only 1 competitor (alternative) and Falcon Shores will be facing 2.
While it is possible that the actual product - Falcon Shores GPU - will be a competent product, gaining ground even with a good product will be challenging. From the POV of shortage of these datacenter GPU cards, which helped AMD, there will be less of that a year from now, the supply / demand will be more balanced - we will see. |
