Arm Cortex-A34 is a 64-bit Only Low-Power Core

Arm previously announced Cortex-A35 64-bit & 32-bit lower-power CPU core, and later on Cortex-A32 32-bit only Armv8 CPU core with the usual press release, and blog posts providing details about their new offering.

But this morning, I saw a tweet about Cortex-A34…

Based on the twitter handle, I first assumed it was “never released” ;), and was just an internal part name at Arm. But the new Cortex-A34 was actually very discreetly outed last month when Arm announced Flexible Access to lower the barrier of entry by allowing IC designers to access all Arm IP in the program, and only pay for IP blocks they actually use in the final product.

Arm Cortex-A34
Click to Enlarge

Nevertheless, the product page and developer documentation are now up – albeit with limited info for the latter -, so we have more details.

Some of the highlights of Arm Cortex-A34 core include:

  • Architecture – 64-Bit Armv8-A
  • Multicore – 1-4x Symmetrical Multiprocessing (SMP) within a single processor cluster, and multiple coherent SMP processor clusters through AMBA 4 technology
  • 8-stage, in-order pipeline
  • ISA Support
    • AArch64 for 64-bit support and new architectural features
    • TrustZone security technology
    • Neon Advanced SIMD
    • DSP and SIMD extensions
    • VFPv4 Floating point
    • Hardware virtualization support
  • Debug & Trace -CoreSight SoC-400

As far as I know, Cortex-A34 is the first and only Cortex processor that will only run 64-bit code, and be incompatible with 32-bit code. [Update: Arm Cortex-A65/A65AE is also 64-bit only]

Arm states the Cortex-A34 will be found in various applications including industrial, smart home devices (home networking devices), healthcare and cloud computing.

Support CNX Software - Donate via PayPal or become a Patron on Patreon

27
Leave a Reply

avatar
5 Comment threads
22 Thread replies
1 Followers
 
Most reacted comment
Hottest comment thread
8 Comment authors
bluwillyJerrydgpdgp Recent comment authors
  Subscribe  
newest oldest most voted
Notify of
blu
Guest
blu

Apple watch should watch out now ; )

Seriously, though, about friggin time arm came up with such a design — IL64P32 has all the advantages of aarch64 without the cost of large pointers — all it takes is compiler support.

dgp
Guest
dgp

Watch out for what? I guess you mean similar competing products based on this core? That might be true if there weren’t already tons of different smart watches out there and the major selling point of the Apple Watch being that it’s made by Apple and not some random numbers laser engraved on chips that no one cares about.

blu
Guest
blu

That was a pun, dgp.

dgp
Guest
dgp

That was pretty obvious but you clearly wrote it with a meaning within the context of the article.

blu
Guest
blu

Did I? The clear meaning is that apple had to play arm pioneers again, showing the world how small, performant cores don’t need to pay the die budget of multi-arch support, as it makes little sense.

dgp
Guest
dgp

I can’t find anything that says the S4 is ARM based. Just “custom 64 bit core”. Would be interesting if it was RISC-V. Anyhow highly integrated chips like the Apple SiPs have been around for a long time.

blu
Guest
blu

> I can’t find anything that says the S4 is ARM based.

I’m sure you could if you tried harder. Let me know when you find out.

> Anyhow highly integrated chips like the Apple SiPs have been around for a long time.

Highly-integrated SiPs of the performance caliber of S4 have been around for a long time? Interesting..

dgp
Guest
dgp

>I’m sure you could if you tried harder. Let me know when you find out.

I couldn’t find anything aside “custom 64 bit dual core” from a quick google. Maybe you could just link it if you’re in the know. (FYI I know it’s probably ARM hence saying it would be interesting if it wasn’t).

>Highly-integrated SiPs of the performance caliber
>of S4 have been around for a long time? Interesting..

The only performance metric for the S4 is that it’s apparently twice as fast as the S3. Which isn’t much to go on. But yes you can put high performance parts into a SiP if you really want the main problem is cost I guess. IIRC there are 4K camera socs with multiple processors and DDR in a single package.

blu
Guest
blu

> I couldn’t find anything aside “custom 64 bit dual core” from a quick google. Maybe you could just link it if you’re in the know. (FYI I know it’s probably ARM hence saying it would be interesting if it wasn’t).

I could, but that won’t change anything in your ‘fly-in hot and ignorant’ style of discussion . I don’t appreciate discussions where the other party does not bother to at least have a clue before jumping in. So, FYI I suspect you suspect it’s arm (everybody and their cat does), but the fact you didn’t bother to do your minimal diligence disinclines me of wasting any further time with you on the subject.

dgp
Guest
dgp

So that’s a long way of saying you don’t actually know. Wikipedia doesn’t know either so you could have just said so and not done you’re weird saving face thing like you do.

Fyi I googled a few different things and even checked if there are any mentions in the Apple extranet sites I have access to and saw nothing yet you can’t just paste a link. *Shrug*

blu
Guest
blu

Really? Here’s the md5 of the two keywords googling which yields a full page of apple S4 arch hits: ed0a06424bab98467a33e28bc3bd1508

Stay adorable, dgp.

dgp
Guest
dgp

See you could have just said to search for arm64_32 to get anything useful out of google but no you went with that silly “due diligence” waffle on and how you were too good to have a discussion with me only to come back with the silly md5 hash.

Stay old and cranky matey.

blu
Guest
blu

I didn’t say it was difficult for me. I said it shouldn’t be difficult for you. Protip: Next time educate yourself *before* flying high into a discussion. You may look a tad less silly.

dgp
Guest
dgp

>Protip: Next time educate yourself *before* flying high into a discussion.

Eh? come again buddy. I made one point about, extrapolating here, there being almost no public information about the S4 and how it would be interesting if it wasn’t ARM because we were talking about high level integration and you dropped everything else and instead of just pasting a link you spent 3 or more comments trying to say I hadn’t put in the effort to be worthy of your big boy debating skills. I wasn’t even countering something you had said and you went off on one.

May I remind you before you start posting md5 sums again that you wrote: “Highly-integrated SiPs of the performance caliber of S4 have been around for a long time? Interesting..”.

Do your research maybe? “hybrids” (multiple dies on a ceramic substrate, multiple dies on an interposer) running at very high frequencies and so on have been around for ages. Don’t you think it would be weird if all of those high frequency components could work on a big noisy PCB with (relatively) bad tolerances in a phone but not work on a smaller PCB with much much better tolerances? That said do we even know what frequencies the components are running at? It seems a bit silly to make a statement about the “performance calibre” of something when those details aren’t available anywhere. But again, what is impressive about something working in almost perfect conditions i.e. the DDR traces aren’t running through multiple layers of China’s finest auto-routed PCB via a socket 10cm or more away.
Maybe that they can build something like that and still make profit on it?

>You may look a tad less silly.

Do yourself and those around you a favour and take a look in the mirror sometime.

blu
Guest
blu

dgp, buddy, I see the intellectual effort of discovering the very secret ISA of apple’s S4 has strained you a lot. Take a nap, we’ll carry on tomorrow.

blu
Guest
blu

dgp, buddy, I hope you’re recovered to intellectual adequacy.

> Eh? come again buddy. I made one point about, extrapolating here, there being almost no public information about the S4

The device has been in dev’s hands for almost a year now. The chances of ‘there being almost no public information about S4’ are nil. You clearly need to know what to look for, which seems challenging for you, but ‘dgp does not know the first thing about s4’ is a far cry from ‘no public info’.

> and how it would be interesting if it wasn’t ARM because we were talking about high level integration

It would also be interesting if it was a quantum computer running on steam energy. It’s neither. Given you admitted you suspected it was arm and yet went off on a baseless supposition shows the levels of whataboutism you’re willing to go at, rather than spend the minimal effort to confirm or refute such an elementary supposition.

> and you dropped everything else and instead of just pasting a link you spent 3 or more comments trying to say I hadn’t put in the effort to be worthy of your big boy debating skills. I wasn’t even countering something you had said and you went off on one.

Of course you weren’t, and I wasn’t debating you. I was filling in your information gaps, until I realized (yet again) it’s a lost cause — dgp will dgp, rather than try to have an intelligent discussion.

> May I remind you before you start posting md5 sums again that you wrote: “Highly-integrated SiPs of the performance caliber of S4 have been around for a long time? Interesting..”.

May I remind you why I posted the hash? You tried to ‘nyah nyah, you don’t know’ me, kindergarten style. You’re such an adorable kid.

> It seems a bit silly to make a statement about the “performance calibre” of something when those details aren’t available anywhere.

See, this is what we’re talking about — dgp doesn’t know something, ergo nobody knows anything related to s4 performance, and it’s silly to talk about performance levels..

Could it be that, perhaps, the public know something? Could it be that competing products (as in wearable SiPs featuring application processors capable of running entire OS ecosystems) are available, to which we could compare the s4? Well, let’s use our super-dgp powers, and google for that, shall we? No, wait, let’s not. We’ll leave that to dgp with his newly-discovered google powers.

dgp, I’ll let you demonstrate you’re a person worthy discussing the subject with, as you come back with a clue what s4 competitors are, what they contain (perhaps competitors are more open about what their wearable SiPs contain), and why s4 massively outperforms them.

Talk to you soon (or not — it’s up to you).

dgp
Guest
dgp

For any on lookers here are Xrays of the S4:
comment image

The “SiP” is actually fairly big.. like the size of the whole device big. The level of integration is still impressive but this is the same sort of “multiple packages on a high density PCB covered in epoxy” like murata, octavo etc do. They still needed PoP inside the SiP to make it possible to get the DDR and NAND into the space.

Laurent
Guest
Laurent

My understanding is that there already is support for AArch64 ILP32 in gcc, kernel, etc. Never tried it though.

willy
Guest
willy

I did for my build farm, resulting in a ~20% faster compiler. But it’s pointless in the end, it’s exactly the same perf as running in armv7 mode which requires less efforts.

maz
Guest
maz

Cortex-A65 is also a pure 64bit CPU.

miguel angel
Guest
miguel angel

What about Apple A12?
iOS is a 64 bits only operating system, and AppStore only have 64 bits apps.
May be A12 use 100% of silicon surface for aarch64, with the lack of armV7 support …
Seems a great advantage about others systems like android (32 bits/64 bits)
2019: widows 10 has 32bit support, android has 32bit support …

dgp
Guest
dgp

My bet: It’ll be too expensive and not well packaged (high density BGA) for really high volume/low cost/battery powered stuff where Cortex-M is popular. I can imagine there will be an imx based on this it won’t be cost effective in any application you’d consider it.

blu
Guest
blu

This is not even competing with M — they are sw-incompatible due to M being thumb-only and this core being thumb-less.

dgp
Guest
dgp

>they are sw-incompatible due to M being thumb-only and this core being thumb-less.

Just like many other situations the instruction set doesn’t matter here. Based on the list of applications this targets from the article this is a low power stripped down core for cost/power constrained applications between MCU and high end smart phone SoC.. ARM’s product page about it is all IoT hypetrain stuff.

For smart home devices you’re looking a complete BOM that’s in the tens of dollars and that’s why Cortex M and R are popular there even though it’s insanity to put them on a network. If you really need Linux you can get that with some cheap Chinese quad core A7 chip.

Jerry
Guest
Jerry

Why would one want a low power 64 bit processor? Isn’t Cortex A-32 perfect for the task? It’s probably even more power efficient.

willy
Guest
willy

It could be the necessary step to force everyone to drop 32-bit in fact. This way we can think of new high-end CPUs not supporting 32-bit with an excuse that “you have 64-bit support down the whole chain, why would you want 32-bit there?”.

blu
Guest
blu

pundits: Why would anyone want a 64-bit processor in a phone?

Apple releases A7. We know what follows.

pundits: Why would anyone want 64-bit processor in a wearable?

Apple releases S4. We’re yet to see what follows, but early signs indicate it might be an A7 repeat.