[octotoad]

> Dan Dobberpuhl was one of the original Alpha designers before he left DEC to start a chip design firm called P.A. Semi, which Apple bought in 1998 for its talent to work on their Arm processors.

P.A. Semi didn't even exist in 1998. It was founded in 2003, and was acquired by Apple in 2008.

[leejoramo]

I also found this line referring to the advantages of 64 vs 32 bits questionable:

> Bits don’t change the processing power; they just change the amount of addressable memory

Now in my personal experience since the early 1980s, increased addressable memory is the biggest advantage for most personal computer workloads. (At least until software is rewritten to use 64-bit instructions) But there are definite processing power advantages too, and for the target market of DEC’s Alpha chip this was likely an important advantage.

[daggersandscars]

It is questionable and indicates the author doesn't understand CPU bit width.

There are three key bit widths that impact a CPU

Address bus: how much RAM the CPU could theoretically address.

Data bus: how many bits at a time the CPU can read from the RAM it can address.

Register: how much data an instruction can operate on at a time.

The author appears to understand only address bus width and may be influenced by early (valid, depending) complaints that the transition from 32 to 64 bit didn't benefit end users.

Going from 32-bit to 64-bit registers can provide a significant performance improvement. But only if you're doing math on 64-bit values. Without specialized instructions, 32-bit and smaller math won't improve due to the register width improvement.

Increasing the size of the data bus can provide a big performance improvement if the previous bus was narrower than the register width.

Many (most? citation needed) current, mainstream CPUs use the same bit width for all three (address, data, register).

The 8088 was notable back in the day because it had a 20 bit address bus, 16 bit registers, and an 8 bit data bus.

Edit: The Alpha had 64 bit registers -- it wasn't just a 32-bit register / data bus CPU with a 64-bit address bus.

[zeusk]

> Many (most? citation needed) current, mainstream CPUs use the same bit width for all three (address, data, register).

This is quite wrong, for most x86_64 processors:

Address bus is 42-48 bits

Data bus is 32-128 bits

Registers are 64 bits but usable as 32/16/8 bits

Even on most ARM processors, neither the address bus nor the data bus are same width as either or the register width.

[fanf2]

Don’t forget the SIMD registers which can be 128/256/512 wide (neon/avx2/avx512)

It’s very difficult to give just a single number for data bus width. Do you add all the DRAM channels? DDR5 has 2 x 32 bit channels where DDR4 has a 64 bit channel. What does bus width mean for PCIe, which uses very fast differential serial lines? Really, it hasn’t been a useful way to describe system performance since the 1990s.

[daggersandscars]

Awesome! Thank you for posting this. I'd edit my response to include this, but it's too late.

[leejoramo]

Great more detailed yet succinct explanation of what I was trying to say

[amalcon]

That confused me a bit, thanks. In '98, Macs still used PPC architecture, and the IPhone didn't exist. I don't know what kind of CPU the IPod was using at the time -- I'd guess ARM. Still, that seemed like it could get away with an off the shelf CPU -- buying a bunch of chip design talent doesn't seem like it would be super rational then.

[classichasclass]

For that matter, PA Semi was itself a Power licensee. They were developing the PA6T, which might have been in whatever the next PowerBook would have been. After the purchase, Apple still had to make promises about its availability because it was being used in some military applications even though they would never use the PA6T in one of their own machines. I think the AmigaOne X1000 was the only generally available computer that ever used it.

[leejoramo]

The Newton was Apple’s first big usage of ARM

https://en.m.wikipedia.org/wiki/Apple_Newton

[masklinn]

iPod was only released in 2001, though, and the idea to make a PMP only really got steam in 2000, with much of the work only happening after Fadell was urgency is how Apple got to contract out much of the work, even he was initially contracted before getting fully hired when his proof of concept got the nod).

OTOH the ill-fated Newton was also ARM-based. As well as the emate pseudo-laptop.