Comment by JonChesterfield
12 days ago
AMD was close to ruin when Lisa took over. It had completely lost the graphics and x64 wars and was limping by on low margin games consoles.
Since then Epyc has broken Intel. Anyone buying Xeon's today is at credible risk of being fired over it when someone notices the power bill relative to their competition.
The graphics ramp is behind Nvidia in mind share and market share. The ROCm software stack gets a lot of grief on here. Nevertheless, Nvidia lost the frontier and el cap bids and now Microsoft is running GPT4 on AMD hardware. Sounds pretty good to me given it's one product line of many.
If turning a multinational semiconductor firm away from the edge of bankruptcy and into the profitable conglomerate it is today doesn't qualify as "solving a hard problem" I don't know what would. It's way beyond what I'd be capable of doing.
Major clouds have been investing in ARM alternatives. x86 is still king compatibility matters a lot. But it is not as simple as Su paints for teams of x86 chip designers to switch to arm chip designers and reach top place, specifically because fabs also matter and AMD is the player in the market with less money to pay the fabs.
GPU market would be hard to recover, and reason to use AMD (for the money printing AI) is just budget. Software is not good enough, it’s understandably not in AMD’s DNA, it was simply lacking in budget as it was close to bankruptcy when CUDA started taking off.
emphasis on top, ofc great designers would still design great no matter the ISA, but best and great is different