4 ms·

It's crazy to me that no other hardware company has sought to compete for the deep learning training/inference market yet ... The existing ecosystems (cuda, py

It's crazy to me that no other hardware company has sought to compete for the deep learning training/inference market yet ... The existing ecosystems (cuda, pytorch etc) are all pretty garbage anyway -- aside from the massive number of tutorials it doesn't seem like it would actually be hard to build a vertically integrated competitor ecosystem ... it feels a little like the rise of rails to me -- is a million articles about how to build a blog engine really that deep a moat ..?

wmf 4y ago
There are tons of companies trying; they just aren't succeeding.
KeplerBoy 4y ago
How could their moat possibly be deeper? First of all you need hardware with cutting-edge chips. Chips which can only be supplied by TSMC and Samsung. Then you need the software ranging all the way from the firmware and driver over something analogous to CUDA with libraries like cuDNN, cuBLAS and many others to integrations into pytorch and tensorflow. And none of that will come for free, like it came to Nvidia. Nvidia built CUDA and people built their DL frameworks around it in the last decade, but nobody will invest their time into doing the same for a competitor, when they could just do their research on Nvidia hardware instead. Realistically it's up to AMD or Intel.
- rcme 4y ago
  There will probably be Chinese options as well. China has an incentive to provide a domestic competitor due to deteriorating relations with the U.S.
  - KeplerBoy 4y ago
    They certainly will have to try, since nvidia is banned from exporting A100 and H100 chips.
    - HTTP418 4y ago
      They do ship A800 and H800 to China. H800 is the H100 with a much slower memory bandwidth. A800 is also a tiered down version of the A100
runnerup 4y ago
No other company has sought this? https://www.cerebras.net/ https://www.cerebras.net/ Has innovative technology, has actual customers, and is gaining a foothold in software-system stacks by integrating their platform into the OpenXLA GPU compiler.

wmf 4y ago

There are tons of companies trying; they just aren't succeeding.

KeplerBoy 4y ago

How could their moat possibly be deeper? First of all you need hardware with cutting-edge chips. Chips which can only be supplied by TSMC and Samsung. Then you need the software ranging all the way from the firmware and driver over something analogous to CUDA with libraries like cuDNN, cuBLAS and many others to integrations into pytorch and tensorflow. And none of that will come for free, like it came to Nvidia. Nvidia built CUDA and people built their DL frameworks around it in the last decade, but nobody will invest their time into doing the same for a competitor, when they could just do their research on Nvidia hardware instead. Realistically it's up to AMD or Intel.

rcme 4y ago
There will probably be Chinese options as well. China has an incentive to provide a domestic competitor due to deteriorating relations with the U.S.
- KeplerBoy 4y ago
  They certainly will have to try, since nvidia is banned from exporting A100 and H100 chips.
  - HTTP418 4y ago
    They do ship A800 and H800 to China. H800 is the H100 with a much slower memory bandwidth. A800 is also a tiered down version of the A100

runnerup 4y ago

No other company has sought this? https://www.cerebras.net/ https://www.cerebras.net/ Has innovative technology, has actual customers, and is gaining a foothold in software-system stacks by integrating their platform into the OpenXLA GPU compiler.