4 ms·

Hence why the best processors (apple, by miles IMO) are ARM?

by mhh__ 3y ago

snvzz 3y ago
>the best processors (apple, by miles IMO) are ARM? I notice your "the best" actually is artificially restricted to something very narrow, I have to assume you mean "the fastest" at some specific TDP, which happens to be the one M1/M2 target. This is evident, as if you had more watts available, there are faster chips out there which were not designed by Apple. And the reason there's no RISC-V chips targeting M1 out there yet is quite simple: Hardware development cycles are long. Most RISC-V CPUs out there are based on the very first ratified privileged and unprivileged specs from 2019. Next year, the first ones based on the batch of specs ratified in 2021, which are important for performance (bit manipulation, vector, hypervisor and so on) are expected to show up. One of them is Tenstorrent's Ascalon, and the lead of the team that designed that one is the same man who led M1 at Apple. They've done presentations in which they present performance as competitive with projected Zen5, yet at lower power and smaller area. Yet there are a lot of definitions for "the best" outside of your narrow focus. For instance, ARM has a range of processors targeting a range of markets, including microcontroller, LITTLE cores, realtime, fault-tolerant and so on. Most of these are bested in Power Performance Area by RISC-V designs already in the market. SiFive does, by itself, manage this much, but there are already tens of companies offering hundreds of competitive designs for licensing. RISC-V is inevitable.
deaddodo 3y ago
What is even your metric for this? Apple CPUs perform well because they are specifically tuned for their use cases with software stacks designed to go hand-in-hand with them. Other than that, they are fairly generic fat cores with wide execution ports. There are a plethora of benchmarks that fall apart for Apple processors, and they hardly compare to most x86 or Power processors when you change the TDP envelope. As with any chip design, they're targeted for a specific use case and they shine in that realm. That hardly makes them "the best" and that claim does a disservice to a multitude of engineers at other firms with other design considerations.
- mhh__ 3y ago
  That "specific use case" being power efficient general purpose computing is not a triviality.
  - deaddodo 3y ago
    > power efficient general purpose computing No, that is not the use case the M cores are tuned for. "Power efficient Apple-based computing" would be a better descriptor. It's useful to keep the entire context of the response in mind, when replying.
    - mhh__ 3y ago
      Most of my use of them is a web browser and a compiler, is that Apple based? It wipes the floor with my Intel &. AMD boxes. > It's useful to keep the entire context of the response in mind, when replying. Don't be glib.

Hence why the best processors (apple, by miles IMO) are ARM?

by mhh__ 3y ago

Hence why the best processors (apple, by miles IMO) are ARM?

>the best processors (apple, by miles IMO) are ARM? I notice your "the best" actually is artificially restricted to something very narrow, I have to assume you mean "the fastest" at some specific TDP, which happens to be the one M1/M2 target. This is evident, as if you had more watts available, there are faster chips out there which were not designed by Apple. And the reason there's no RISC-V chips targeting M1 out there yet is quite simple: Hardware development cycles are long. Most RISC-V CPUs out there are based on the very first ratified privileged and unprivileged specs from 2019. Next year, the first ones based on the batch of specs ratified in 2021, which are important for performance (bit manipulation, vector, hypervisor and so on) are expected to show up. One of them is Tenstorrent's Ascalon, and the lead of the team that designed that one is the same man who led M1 at Apple. They've done presentations in which they present performance as competitive with projected Zen5, yet at lower power and smaller area. Yet there are a lot of definitions for "the best" outside of your narrow focus. For instance, ARM has a range of processors targeting a range of markets, including microcontroller, LITTLE cores, realtime, fault-tolerant and so on. Most of these are bested in Power Performance Area by RISC-V designs already in the market. SiFive does, by itself, manage this much, but there are already tens of companies offering hundreds of competitive designs for licensing. RISC-V is inevitable.

deaddodo 3y ago

What is even your metric for this? Apple CPUs perform well because they are specifically tuned for their use cases with software stacks designed to go hand-in-hand with them. Other than that, they are fairly generic fat cores with wide execution ports. There are a plethora of benchmarks that fall apart for Apple processors, and they hardly compare to most x86 or Power processors when you change the TDP envelope. As with any chip design, they're targeted for a specific use case and they shine in that realm. That hardly makes them "the best" and that claim does a disservice to a multitude of engineers at other firms with other design considerations.

mhh__ 3y ago
That "specific use case" being power efficient general purpose computing is not a triviality.
- deaddodo 3y ago
  > power efficient general purpose computing No, that is not the use case the M cores are tuned for. "Power efficient Apple-based computing" would be a better descriptor. It's useful to keep the entire context of the response in mind, when replying.
  - mhh__ 3y ago
    Most of my use of them is a web browser and a compiler, is that Apple based? It wipes the floor with my Intel &. AMD boxes. > It's useful to keep the entire context of the response in mind, when replying. Don't be glib.