ARM announces the successor of the A57: Cortex A72

[lopri]

TechReport reports:

The CPU core is called the Cortex-A72, and it supplants the Cortex-A57 as ARM's fastest CPU core. The Cortex-A72 will support big.LITTLE multi-architecture power-saving configurations by pairing up with the "little" Cortex-A53 core. Both of these cores support the 64-bit ARMv8-A instruction set.

http://techreport.com/news/27767/arm-unveils-cortex-a72-cpu-mali-t880-graphics-and-more

So I checked out ARM's site.

http://www.arm.com/markets/mobile/premium-mobile-ip-suite.php

The ARM Cortex-A72 processor is based on the ARMv8-A architecture that delivers energy-efficient 64-bit processing whilst providing full backward compatibility to existing 32-bit software.

The Cortex-A72 processor delivers substantial new benefits:

• Sustained operation within the constrained mobile power envelope at frequencies of 2.5 GHz in a 16nm FF+ process and scalable to higher frequencies for deployment in larger form factor devices
• 3.5X the performance of 2014 devices based on the Cortex-A15 processor
• Improved energy efficiency that delivers a 75 percent reduction in energy consumption when matching performance of these 2014 devices
• Extended performance and efficiency when the Cortex-A72 CPU is combined with a Cortex-A53 CPU in ARM big.LITTLE™ processor configurations

 

[poofyhairguy]

Nice, that is desktop class. Lets see how it gets implemented though, that is the beef with vanilla ARM cores.

 

[ChronoReverse]

Nice, that is desktop class. Lets see how it gets implemented though, that is the beef with vanilla ARM cores.

Indeed. I think we have enough precedent to take these sorts of announcements with heaps of salt.

 

[Suspicious-Teach8788]

Indeed. I think we have enough precedent to take these sorts of announcements with heaps of salt.

Really? I think the talk about Intel finally making it into mobile is something that requires mountains of salt.

 

[ChronoReverse]

Really? I think the talk about Intel finally making it into mobile is something that requires mountains of salt.

Intel has chips that fit the technical requirements at this point (or rather, in the previous generation they did, but the ARM/Apple/Qualcomm chips have step up to another level since). Whether it's economically reasonable for anyone to actually use them is a separate issue.

And even if they weren't technically fit, it still has nothing to do with this since the comment was about the ARM chips having a lot of trouble from the get-go. Whether or not Intel farts rainbows from their chips doesn't really relate.

 

[Mopetar]

Really depends on what they mean by 3.5x performance improvement. Typically this is some obscure benchmark (that may or may not have much real-world application) that showed some significant performance improvements rather than an average.

I'll reserve judgement until we start saying actual benchmarks, but I remain skeptical that it will actually be as powerful as they would claim.

 

[Midwayman]

Really depends on what they mean by 3.5x performance improvement. Typically this is some obscure benchmark (that may or may not have much real-world application) that showed some significant performance improvements rather than an average.

I'll reserve judgement until we start saying actual benchmarks, but I remain skeptical that it will actually be as powerful as they would claim.

If they don't bother to specify, its always some best case theoretic increase. Upto 3.5x would be more appropriate.

 

[lopri]

I hope that x3.5 best-case scenario hasn't factored in big.LITTLE configuration. That would be messed up.

If A72 is as good as ARM claims it to be, Qualcomm and Apple will learn and copy over good stuff for their own SOCs.

 

[darkswordsman17]

I don't really know what's to be so skeptical about? The only real issue is how long it will take for us to see implementations of this. Considering it's aimed at 16nm and rumors about process tech are all over the place right now, no real idea when it might show up.

A53/A57 were announced like 2 years before we got them, so I'd be shocked if it can't manage that level of performance in 1-2 years timeframe. I'm actually a little surprised that we're getting this announcement since it seems like A5x implementations are just now starting to propagate, but I guess if it takes that long for them to iterate the design then it makes sense. But maybe ARM is looking to speed up their development.

And lopri, I'd actually guess that you have it backwards, that plenty of the performance improvements in the ARM cores are basically the same things that Apple and Qualcomm have been doing for their custom chips, paired with the usual improvements (process tech improvements, GPU robustness, faster memory, newer interconnects/interfaces, more robust co-processors for things like media encode/decode). I think A57 for instance was this way. There's a lot of general things that can be improved about processors that would be feasible for basically anyone to implement. I wouldn't think big-little has anything to do with it as they compared it specifically to A57 and A15 and not A53/A57 which is how the big-little core implementation works.

 

[lopri]

You may be right, darkswordsman17. I do not know exactly which way it goes.

But didn't Qualcomm and Apple develop after ARM announced Cortex-A15? From what I remember, ARM was aiming servers in addition to mobile with A15, and Qualcomm and Apple were not happy with its power characteristics, which prompted them to go custom routes. In the course of customization they took the good and threw away the bad from A15 - or so I've heard.

In the grand scheme of things I do not think this detail is important, however.