3 ms·

The GPUs seemed smart. Too late, and timid, but smart. What continually blew my mind was Intel's chiplet strategy. While AMD was making scads of a single chi

The GPUs seemed smart. Too late, and timid, but smart. What continually blew my mind was Intel's chiplet strategy. While AMD was making scads of a single chiplet and binning the best for Epyc and recovering cost at the low end with Ryzen, Intel designed and fabricated a dizzying number of single-purpose chiplets. In some cases, just the mirror image of another otherwise identical chiplet. The mind boggles. What phenomenal inattention to opportunity.

washadjeffmad 2y ago
Same with the general preparedness for the AGI Cambrian explosion. Give their position, they should have been able to keep pace with Nvidia beyond the data center, and they fumbled it. Their purported VRAM offerings for Battlemage are also lower than hoped for, which is going to be a turnoff to many buying for local inference. I think we have too many decision-makers gunshy from crypto mining that haven't yet realized that compute isn't just a phase.
HelloNurse 2y ago
What you describe is mad inattention to costs, not only to opportunities: maybe a symptom of widespread misaligned incentives (e.g. delivering a design with mirrored chiplets quickly could be more "useful" for an engineer than saving a few millions for the company by taking one more week to design a more complex assembly of identical chiplets) and toxic priorities (e.g. theoretical quality over market value and profits, risk aversion, risk/cost externalization towards departments you want to be axed instead of your own).
- timschmidt 2y ago
  Couldn't agree more. To me, AMD's approach demonstrates planning and buy-in across multiple teams (CPU die, Ryzen IO die, Epyc IO die, etc), and that suggests a high degree of management involvement in these engineering decisions. Intel's activities seem comparatively chaotic. Which I agree smells of disincentives and disinterested middle management hyperfixated on KPIs.

washadjeffmad 2y ago

Same with the general preparedness for the AGI Cambrian explosion. Give their position, they should have been able to keep pace with Nvidia beyond the data center, and they fumbled it. Their purported VRAM offerings for Battlemage are also lower than hoped for, which is going to be a turnoff to many buying for local inference. I think we have too many decision-makers gunshy from crypto mining that haven't yet realized that compute isn't just a phase.

HelloNurse 2y ago

What you describe is mad inattention to costs, not only to opportunities: maybe a symptom of widespread misaligned incentives (e.g. delivering a design with mirrored chiplets quickly could be more "useful" for an engineer than saving a few millions for the company by taking one more week to design a more complex assembly of identical chiplets) and toxic priorities (e.g. theoretical quality over market value and profits, risk aversion, risk/cost externalization towards departments you want to be axed instead of your own).

timschmidt 2y ago
Couldn't agree more. To me, AMD's approach demonstrates planning and buy-in across multiple teams (CPU die, Ryzen IO die, Epyc IO die, etc), and that suggests a high degree of management involvement in these engineering decisions. Intel's activities seem comparatively chaotic. Which I agree smells of disincentives and disinterested middle management hyperfixated on KPIs.