By: wumpus (wumpus.delete.delete@this.this.lost.in.a.hole), March 21, 2021 12:24 pm
Room: Moderated Discussions
Linus Torvalds (torvalds.delete@this.linux-foundation.org) on March 20, 2021 1:20 pm wrote:
> Hugo Décharnes (hdecharn.delete@this.outlook.fr) on March 20, 2021 7:34 am wrote:
>
> If your new instruction set starts off with needing a new OS, new libraries, new drivers, and new applications,
> you better also have a billion "new users" you can provide. Because otherwise you're kind of stuck.
>
> Linus
Your best (only) bet would be to forget about a new architecture for general purpose programming and try to optimize some niche with a market big enough to support it. Best guess is machine learning, although if you managed to create a superior SQL architecture, that would also work. Not that there aren't other places, but they are probably small enough that making the jump to general purpose computing is likely impossible. Customers might prefer their current infrastructure, but I suspect AMD would do well to add Vega cores to FPGAs for their current (Xilinx).
On the other hand, also expect that your new market will drag you away from any chance to jump to general purpose land. I'm sure plenty of people at nvidia and AMD graphics wanted to build a GPUish processor that could handle graphics and general purpose code (AMD's ALU attempts seemed to be a first step), but the architecture you'd need for that wouldn't be good enough for graphics. And nvidia's GPGPU customers seem happy with nvidia's SIMT ideas (which doesn't seem to be what anybody wants in this thread).
> Hugo Décharnes (hdecharn.delete@this.outlook.fr) on March 20, 2021 7:34 am wrote:
>
> If your new instruction set starts off with needing a new OS, new libraries, new drivers, and new applications,
> you better also have a billion "new users" you can provide. Because otherwise you're kind of stuck.
>
> Linus
Your best (only) bet would be to forget about a new architecture for general purpose programming and try to optimize some niche with a market big enough to support it. Best guess is machine learning, although if you managed to create a superior SQL architecture, that would also work. Not that there aren't other places, but they are probably small enough that making the jump to general purpose computing is likely impossible. Customers might prefer their current infrastructure, but I suspect AMD would do well to add Vega cores to FPGAs for their current (Xilinx).
On the other hand, also expect that your new market will drag you away from any chance to jump to general purpose land. I'm sure plenty of people at nvidia and AMD graphics wanted to build a GPUish processor that could handle graphics and general purpose code (AMD's ALU attempts seemed to be a first step), but the architecture you'd need for that wouldn't be good enough for graphics. And nvidia's GPGPU customers seem happy with nvidia's SIMT ideas (which doesn't seem to be what anybody wants in this thread).