Warning: Some posts on this platform may contain adult material intended for mature audiences only. Viewer discretion is advised. By clicking ‘Continue’, you confirm that you are 18 years or older and consent to viewing explicit content.
They could make their own RISC-V implementations, what annoys them is investing resources on it while they could just milk their already existing propietary ISAs.
I’d wager that both AMD and Intel very worried about RISC because of what Apple managed to do with M1 architecture. This is a great technical write up about it. The TLDR is that there are two major benefits.
First is that system on a chip approach allows sharing memory between different chips such as CPU, GPU, and so on, without needing a bus which removes a major bottleneck. You no longer have to copy data from CPU memory to GPU memory, do some computation, and then copy stuff back. This bit is compatible with CISC architecture, and AMD has already been building SoC style chips for a little while now.
The second benefit is the one that I suspect is making AMD and Intel nervous. Since RISC instructions have a fixed length, it’s possible to load a batch of instructions, figure out which ones have to be run in order and which ones don’t, and then execute all the independent instructions in parallel. This makes it possible to just keep adding more cores and parallelize the work across them. And this scales to hundreds of cores working together.
Turns out that this is basically impossible to do at scale with CISC chips because the instructions are variable length, and calculating dependencies between them simply doesn’t scale. AMD found that the overhead of figuring out dependencies starts negating the benefits you get from parallel execution around 3-4 instructions.
This is why Apple M series chips are now in a class of their own right now. They run way faster than x86 chips and they use a lot less power.
Now that Apple showed the benefits of such architecture, I think it’s only a matter of time before we’ll see similar approach taken with RISC-V chips which will basically make x86 architecture obsolete.
I remember my dad saying much the same thing back in the mid 90s. He was certain that RISC would destroy CISC because of the obvious benefits inherent in the design. Turns out markets don’t always go for the best or most efficient solution, at least not right away.
This is a regressive development. Do Intel, AMD, ARM and the likes feel threatened by RISC-V?
They could make their own RISC-V implementations, what annoys them is investing resources on it while they could just milk their already existing propietary ISAs.
That would be competing in the market without government intervention. That is against the American way!
I’d wager that both AMD and Intel very worried about RISC because of what Apple managed to do with M1 architecture. This is a great technical write up about it. The TLDR is that there are two major benefits.
First is that system on a chip approach allows sharing memory between different chips such as CPU, GPU, and so on, without needing a bus which removes a major bottleneck. You no longer have to copy data from CPU memory to GPU memory, do some computation, and then copy stuff back. This bit is compatible with CISC architecture, and AMD has already been building SoC style chips for a little while now.
The second benefit is the one that I suspect is making AMD and Intel nervous. Since RISC instructions have a fixed length, it’s possible to load a batch of instructions, figure out which ones have to be run in order and which ones don’t, and then execute all the independent instructions in parallel. This makes it possible to just keep adding more cores and parallelize the work across them. And this scales to hundreds of cores working together.
Turns out that this is basically impossible to do at scale with CISC chips because the instructions are variable length, and calculating dependencies between them simply doesn’t scale. AMD found that the overhead of figuring out dependencies starts negating the benefits you get from parallel execution around 3-4 instructions.
This is why Apple M series chips are now in a class of their own right now. They run way faster than x86 chips and they use a lot less power.
Now that Apple showed the benefits of such architecture, I think it’s only a matter of time before we’ll see similar approach taken with RISC-V chips which will basically make x86 architecture obsolete.
Any chance you know of an archive of that link? I was enjoying it until it told me to sign in.
oh yeah here https://archive.ph/Nmgp3
I remember my dad saying much the same thing back in the mid 90s. He was certain that RISC would destroy CISC because of the obvious benefits inherent in the design. Turns out markets don’t always go for the best or most efficient solution, at least not right away.
Indeed, and I’m glad to see RISC finally vindicated. I’ve always thought it was a much more elegant architecture.