ARM Announces Cortex-A32 As A Tiny, 32-bit ARMv8 CPU

[duby229]

That is an Artifact of the embedded systems focus for most of these processors. Nothing abiutnAR mm processors prevents expandable RAM systems. Expandable RAM thought just does make sense for the targeted use cases.

This is bad how?

I do wonder why they bother with 32 bit hardware, in the form of an applications processor.


On the flip side at least ARM is actually delivering new hardware. Even if it is 32 bit, we are still seeing innovation. This is more that what we see out of AMD and Intel of late.

In a way that is perfectly fair and I actually agree with you to some extent. We all know what AMD's problems are and we all know that without AMD competing Intel doesn't even try. So yeah, I would love ARM to totally displace x86. In which case there would need to be both laptop and desktop arm devices.... In which case you can be damn sure Intel would then try very hard to compete.

[Marc Guillot]

I used to think Intel had confusing product names, but I just can't keep up with ARM's anymore. The Cortex names seem so meaningless now.

The number on the name is an indication of its performance and complexity in the same generation. A Cortex-A32 means that it's slightly slower and smaller than the Cortex-A35.

We have this generation : Cortex-A32 < Cortex-A35 < Cortex-A53 < Cortex-A57 < Cortex-A72

Last generation was also similar : Cortex-A5 < Cortex-A7 < Cortex-A9 < Cortex-A15 < Cortex-A17

[droidhacker]

The thing is that most arm devices don't have user upgradeable RAM. So basically all of them will have the same amount of RAM on it's last day that it had on it's first day.I really don't think many end users would notice one tiny little bit of a difference. The people that will notice the difference are the hackers unlocking bootloaders and compiling new ROMS. So I don't see much of a point. It seems like a waste engineering resources and fab allocation.

It's something that will make zero difference except irk the poeple that have another platform to support.

WTF are you talking about? What about this has anything to do with RAM, upgradable or not?

[schmidtbag]

The number on the name is an indication of its performance and complexity in the same generation. A Cortex-A32 means that it's slightly slower and smaller than the Cortex-A35.

We have this generation : Cortex-A32 < Cortex-A35 < Cortex-A53 < Cortex-A57 < Cortex-A72

Last generation was also similar : Cortex-A5 < Cortex-A7 < Cortex-A9 < Cortex-A15 < Cortex-A17

I'm not sure this is how it works. For example, an A7 (maybe even an A5) can outperform an A8.

If the names are based on a technical perspective, the numbers still make no sense at all. There is no consistency, rhyme, or reason. For example:
* The A15 is effectively the successor to the A9. A15 is faster and more efficient per-core, and is a big leap ahead. Sounds good so far, but...
* The A9 is pretty much just a multi-core A8, yet it's only an increment of 1.
* The A7 is the successor to the A8. It is faster and more efficient per-core, and can be multi-core, yet the number is a "step back".
* The A15 has a smaller L1 cache than the A12 and lacks an L2 cache, yet is "ranked higher". The A15 is overall slower/worse.
* To my knowledge, all ARMv8 processors are 64 bit, but the Cortex A32 comes along and is 32 bit.

So, the numbers aren't based on total performance and aren't based on technical features. The numbers don't [accurately] represent per-core performance either.

[duby229]

WTF are you talking about? What about this has anything to do with RAM, upgradable or not?

Because it is -the- one and only thing an end user will notice the difference between a 32bit CPU and a 64bit one. For end users this processor doesn't matter one tiny little bit.

[ldesnogu]

* The A9 is pretty much just a multi-core A8, yet it's only an increment of 1.

The A9 has nothing to do with A8 except for its NEON unit which was inherited from A8. All the rest was done by another team and is much more advanced than A8 (OoOE, stream detection, etc.) The FP unit is also much faster than the A8 one, since the A8 FPU isn't pipelined, while the A9 one is. All of this makes the A9 significantly faster than A8 for most apps.

[schmidtbag]

The A9 has nothing to do with A8 except for its NEON unit which was inherited from A8. All the rest was done by another team and is much more advanced than A8 (OoOE, stream detection, etc.) The FP unit is also much faster than the A8 one, since the A8 FPU isn't pipelined, while the A9 one is. All of this makes the A9 significantly faster than A8 for most apps.

Well then, I guess that makes my point even more prominent.

[liam]

The number on the name is an indication of its performance and complexity in the same generation. A Cortex-A32 means that it's slightly slower and smaller than the Cortex-A35.

We have this generation : Cortex-A32 < Cortex-A35 < Cortex-A53 < Cortex-A57 < Cortex-A72

Last generation was also similar : Cortex-A5 < Cortex-A7 < Cortex-A9 < Cortex-A15 < Cortex-A17

It's actually more confusing than that.
A32 is faster at executing 32bit code than the A35, and uses a surprisingly smaller amount of energy, as well. The A53 is a bit faster than the A35, but uses a good deal more energy to get to than point. Then there's a massive leap in power to the A57, which, from EVERYTHING I've read, is a pretty poor design. As a result the A72 is much faster and uses less energy.
Then, you've the separation of core types with big.LITTLE, which would result in two branches, at least, for each generation, but the A32 mucks this up because it doesn't have an identical ISA to the other cores, so it may belong on its own branch (given than this is intended for iot, this may make sense), but I don't know why you couldn't match the A32 with the A72, and run the latter in aarch32 only.
Ofc, none of this really matters as this info isn't for consumer consumption.

[Nelson]

It's actually more confusing than that.
A32 is faster at executing 32bit code than the A35, and uses a surprisingly smaller amount of energy, as well. The A53 is a bit faster than the A35, but uses a good deal more energy to get to than point. Then there's a massive leap in power to the A57, which, from EVERYTHING I've read, is a pretty poor design. As a result the A72 is much faster and uses less energy.
Then, you've the separation of core types with big.LITTLE, which would result in two branches, at least, for each generation, but the A32 mucks this up because it doesn't have an identical ISA to the other cores, so it may belong on its own branch (given than this is intended for iot, this may make sense), but I don't know why you couldn't match the A32 with the A72, and run the latter in aarch32 only.
Ofc, none of this really matters as this info isn't for consumer consumption.

They make parts for purposes. There are at least 3 dimensions: performance, power usage, size/complexity (think cost.) Then as manufacturing technology improves there are shrinks and clock increases and such. It's not possible to just have a simple A7 is faster than A5 comparison because 7>5. An A5 clocked higher can outperform an A7 or A9.

Generally the bigger number means faster, generally but there can be trade-offs.

[liam]

They make parts for purposes. There are at least 3 dimensions: performance, power usage, size/complexity (think cost.) Then as manufacturing technology improves there are shrinks and clock increases and such. It's not possible to just have a simple A7 is faster than A5 comparison because 7>5. An A5 clocked higher can outperform an A7 or A9.

Generally the bigger number means faster, generally but there can be trade-offs.

I think you're mostly correct, but I'm not at all certain that the a5 could practically be made to outperform an a7. However, I've endeavoured to find out more definitive information about this but haven't been able.