4 ms·

It looks like they renamed the article to http://nvidianews.nvidia.com/news/gh200-grace-hopper-superchip-with-hbm3e-memory http://nvidianews.nvidia.com/news/gh2

It looks like they renamed the article to http://nvidianews.nvidia.com/news/gh200-grace-hopper-superchip-with-hbm3e-memory http://nvidianews.nvidia.com/news/gh200-grace-hopper-superch... As far as I was aware, the first gen was barely available yet?

htrp 3y ago
Not sure how many people want to rearchitect their entire ML architecture to use ARM
- ksec 3y ago
  Huh? They are not running on top of Raw ARM CPU but CUDA on Nvidia GPU. It makes little to no difference in what CPU they are using.
- pjmlp 3y ago
  GPGPU programming uses JIT compilation, either from source or bytecode, it doesn't matter what is the CPU being used on the cards.
  - hulitu 3y ago
    As an amateur, i will expect that some GPUs perform better than others.
    - pjmlp 3y ago
      Definitely, however that goes beyond it being Intel, ARM, or whatever.