AMD's MASTER PLAN COMPLETED? AMD Server Data Center Premiere Keynote Analysis

I analyze the recent AMD data centre premiere keynote, assessing it against the competition, AMD's past and their potential future.

#amd, #gpus, #server

[liaminwales]

Good video, thanks.
A nice example of cache use in games is how faster ram helps gain more FPS, when the CPU needs data and it's not in cache it has to wait for it from RAM. With bigger cache the CPU wont have to wait for data to come from ram (Ie 'stall').
Will be interesting to see if the new CPU's are less effected by RAM speed in benchmarks.
(I also wonder if bigger cache lets them do more aggressive pre fetching of data?)

I hope 'oneAPI' from intel can open up the CUDA space, for the home user AMD's consumer GPU's with large VRAM are super attractive. It's clear Nvidia are limiting VRAM on consumer cards to block there use for non gaming apps like ML/AI/video editing etc..
edit apple also is moving to 3nm in 2023

[VigneshBalasubramaniam]

The point you make about CUDA is very important. Its critical IMO that a proper open source alternative to CUDA is made available. And the day that happens, is the day that NVIDIA dies, since there will be no longer any reason to stick with NVIDIA.

[VigneshBalasubramaniam]

NVLink works on ARM and on the IBM OpenPower architectures AFAIK. NVIDIA is definitely never going to get Intel to play ball with them anymore now that Intel is getting serious in the datacenter GPU/accelerator space.

On a side note, I'm hoping that the NVIDIA-ARM deal fails, since that would be one of the worst deals ever for the entire computing industry.