Originally Posted by rkelsen

I don't know if you're aware of this, but the "R" in ARM stands for RISC.

RISC stands for Reduced Instruction Set Computer.

ARM processors will never be as capable as Intel or AMD. They don't have to worry, just because Linus uses one while travelling.

This article was from not long ago: https://www.zdnet.com/article/you-ca...ere-to-buy-it/

Originally Posted by pan64

probably relevant: AMD may be looking towards RISC-V in future CPUs.

Originally Posted by kernelhead

I'm most intrigued and interested in your statement, here, that ARM will never be as capable as Intel or AMD.

Originally Posted by kernelhead

When Apple, recently, put the M1 in their laptops & desktops, I did a fair amount of reading about ARM.

Originally Posted by some internet rando

ARM processor don’t tend to scale up well in the high performance segment since they don’t have extensions comparable to what SSE3/4 or AVX offers on the x86 platform. Support is being worked on to bring scalable vector extensions for data centers and HPC but no current ARM products use it.

You also have the issue of simpler instructions which are inadequate for heavier workloads, on ARMv7/v8-A they’re all 32-bits long even in 64-bit mode (Aarch64). By contrast, x86 benefits from instruction density since it has variable length instructions (up to 120-bits) and more complex instructions, meaning that less bits are required to express a single instruction. Take for example the addq (%rax,%rbx,2), %rdx instruction which is executed in one go on modern x86 cores. To do the same thing on ARM, you would have to use the load instruction before using the add instructions meaning that you’ll need more cycles to do the same thing.

Another disadvantage of ARM systems concerns device enumeration which is a no brainer on x86 systems due to industry standards like BIOS/UEFI. An OS running on x86 systems can know exactly what is attached or connected to X or Y bus. On ARM there is no such thing because of proprietary buses and non-discoverable components. Take Linux on ARM for instance, if you want to tell it what features, buses or devices are there you must use workarounds like device trees/device tree overlays and a lot of out-of-tree blobs which are not open source. ARM platforms don’t have a unified industry-wide BIOS, instead each vendor uses its own implementation and sometimes their own internal buses. On x86 most hardware device have open source drivers and is generally easy to setup, you don’t need those hacks and workarounds. ARM’s business strategy revolves around their multitude of IP (from ARM cores, interconnects, buses and Mali GPU) which they license to their ‘partners’. Each partners can then design their own SoC implementation which is not necessarily similar to one another. This is why it’s hard to have a fully functional OS+kernel for all ARM platforms.

Another thing is cache coherency and high-bandwidth buses. This is something that the ARM platform struggles with due to power constraints (buses can use a lot of power!) and ARM’s obsession with small dies, which is why until very recently ARM cores lacked L3 cache for cache coherency and instead use a fancy CCI bus to do so.

Originally Posted by kernelhead

Would you argue that Intel/AMD will be able to keep the speed advantage in their processors. I do believe that ARM has already demonstrated that, at least, it has the advantage in running at much lower temperatures.

Originally Posted by rkelsen

Maybe my opinion on that is a little too strong, and these things are being developed at break neck pace so it may be outdated soon enough.

Admittedly, my own experience with ARM CPUs is little more than in the dinky devices we tend to carry around.

To give a practical example of my own findings, virtualisation and emulation don't run well on ARM chips. These are two technologies that I use daily and cannot live without.



I found this opinion online, which sums up the problems with ARM:

I'm somewhat removed from CPU development, but I guess many of the issues are being worked on?

Either way, this move by Apple will only increase the aforementioned pace of development, and could prove to be a good thing in the long run.

And at lower rates of power consumption.

Originally Posted by kernelhead

Also, I'm curious as to how many of these "tasks" are for very high end computing - it seems that Apple's M1 processors are thus far adequate for personal/home computer users (?)