ARM releases Cortex A15 info

[pxc]

http://www.arm.com/files/pdf/AT-Exploring_the_Design_of_the_Cortex-A15.pdf

Target frequencies are 1.5GHz in low power 32/28nm processes and 2.5GHz in high performance 32/28nm processes. A quad core block diagram is shown in the document.

New power saving features.

New instruction extensions including FMAC @ 4 per clock

Massive per core performance increases and lower power usage per task at target frequency (maximum power seems to have gone up significantly, about 2x higher than a 45nm A8):

A9 is not shown, so it looks like A15 is an improvement, but not a gigantic leap over A9.

Several of the new features seem aimed at server functionality:
40-bit/1TB physical memory addressing (in 64-bit mode I guess, one slide lists 4GB max for 32-bit)
modern virtualization and protection features
ECC L1 & L2 caches

See the document for details of new features that improve performance. The A15 is basically a 3-way OOO execution processor with 8 execution units. That's more complex than either Atom (2-way in-order) or Bobcat (2-way out of order).

 

[TheBlueChanell]

Things are going to get really interesting soon. Smart move by Microsoft, jumping on the ARM wagon. Pretty impressive performance metrics for such a little guy.

 

[450]

I want an iPad with this chip.

 

[rflcptr]

This would make an interesting MacBook Air.

 

[Untitledone]

Gotta love the speed at which phones, and other mobile divices are progressing. Like SSD's did, pretty close to breaking moore's law.

 

[/rant/]

The Cortex A15 is where Microsoft can bring 64-bit Windows for ARM as far as I understand. With that, the possibilities are endless. Whether you want a miniature phone that runs two weeks without recharge or some almighty Server, it basically will be all the same architecture.

It is also the architecture that will attack Intel's main market and pose a real threat to them. Intel doesn't seem to have any aces up their sleves, so they better watch out for this big boy here coming. Maybe they'll license it themselves and bring their own ARM SoC instead of old Atoms and castrated ULVs. I'm looking forward to this.

 

[pxc]

The Cortex A15 is where Microsoft can bring 64-bit Windows for ARM as far as I understand.

It is also the architecture that will attack Intel's main market and pose a real threat to them. Intel doesn't seem to have any aces up their sleves, so they better watch out for this big boy here coming. Maybe they'll license it themselves and bring their own ARM SoC instead of old Atoms and castrated ULVs. I'm looking forward to this.

Disagree with all of the above.

Windows 8 for ARM will almost certainly have a single 32-bit version, customized for the 4 SoCs/architectures MS announced support for. 64-bit is only for servers in the foreseeable future, IMO.

Intel is starting to ship new 32nm Medfield Atoms already, first of several tiny system on a chip versions with ARM-like power consumption and suitable for phones. I think it's not going to be a real challenge in handsets because x86 compatibility is not needed in that market, and the price is too high for a commodity smartphone. Windows can run on it, but it requires an obscene amount of flash memory (32GB in the phones demonstrated so far) for the OS. Intel's "ace" is manufacturing advantages and a long overdue update of the Atom architecture next year.

Intel was an ARM licensee. It had success in the PDA market with its StrongARM models. It was sold off years ago in favor of the current Atom strategy (which has sold over 100 million devices as of earlier this year). Not exactly the same volume as all ARM-based chips, but much more profitable. With Atom's roadmap, I don't think Intel is considering becoming an ARM licensee again.

I'm not saying Intel will be successful in the handheld market (reasons stated above), but it should remain a competitive player in tablets. Also, while Atom may not carve into ARM strongholds, it should help keep market share for downsized PCs (netbooks, tablets/slates) with broader software compatibility with full Windows than the zero support Windows for ARM will have for legacy Windows applications.

The problem with ARM affecting x86 sales will hit both Intel and AMD, with Intel in a slightly better position since it is starting to have products competitive with ARM chips (besides price). AMD needs to catch up quickly with sub-1W SoCs, or it's going to repeat missing out on the tablet boom* like it did for the netbook boom.

* unsurprisingly it's been underwhelming so far. the sales volume analysts predicted is actually closer to 1/2 or fewer ARM tablets sold... non-iPad models are not taking off. who would have known?!?!?

 

[user21]

Well this is gr8 as most of the demanding future devices are now carrying on with ARM and they are too trying to enter desktop market! thats the reason micrsoft is up releasing their windows 8 with much more hardware support
check out the link guyz
[ame="http://www.youtube.com/watch?v=UYSSdSNFjhU"]http://www.youtube.com/watch?v=UYSSdSNFjhU[/ame]
windows getting flexible even more! imagine the oldest operating system now going integrated everywhere with any hardware possible especially ARM

 

[user21]

Disagree with all of the above.

Windows 8 for ARM will almost certainly have a single 32-bit version, customized for the 4 SoCs/architectures MS announced support for. 64-bit is only for servers in the foreseeable future, IMO.

Intel is starting to ship new 32nm Medfield Atoms already, first of several tiny system on a chip versions with ARM-like power consumption and suitable for phones. I think it's not going to be a real challenge in handsets because x86 compatibility is not needed in that market, and the price is too high for a commodity smartphone. Windows can run on it, but it requires an obscene amount of flash memory (32GB in the phones demonstrated so far) for the OS. Intel's "ace" is manufacturing advantages and a long overdue update of the Atom architecture next year.

Intel was an ARM licensee. It had success in the PDA market with its StrongARM models. It was sold off years ago in favor of the current Atom strategy (which has sold over 100 million devices as of earlier this year). Not exactly the same volume as all ARM-based chips, but much more profitable. With Atom's roadmap, I don't think Intel is considering becoming an ARM licensee again.

I'm not saying Intel will be successful in the handheld market (reasons stated above), but it should remain a competitive player in tablets. Also, while Atom may not carve into ARM strongholds, it should help keep market share for downsized PCs (netbooks, tablets/slates) with broader software compatibility with full Windows than the zero support Windows for ARM will have for legacy Windows applications.

The problem with ARM affecting x86 sales will hit both Intel and AMD, with Intel in a slightly better position since it is starting to have products competitive with ARM chips (besides price). AMD needs to catch up quickly with sub-1W SoCs, or it's going to repeat missing out on the tablet boom* like it did for the netbook boom.

* unsurprisingly it's been underwhelming so far. the sales volume analysts predicted is actually closer to 1/2 or fewer ARM tablets sold... non-iPad models are not taking off. who would have known?!?!?

you just reminded me LARABI a code name for intel graphics to minimize both amd and nvidia! intel is still the biggest player around and maximum market segment taker! amd is up with same AM3+ socket with its 8core processor! pre leaked benches are no good i dont know if anything is work needy of 8cores right now! ARM is famous for its low power solution tablets! the conclusion still remains that taking down the blue is far from difficult right now but ARM might have a chance in mobility

 

[pxc]

you just reminded me LARABI a code name for intel graphics to minimize both amd and nvidia!

Funny you should mention Larrabee. The successor to it was never cancelled as most dopey web sites reported. It's back in the news and codenamed Knights Corner, a "many integrated core" (MIC) 50+ core chip aimed primarily at a different purpose than desktop PC graphics. The seeded Larrabee systems went out to researchers and for use in render farms IIRC. There is a HPC press release here with a mention of the new chip: http://www.hpcwire.com/hpcwire/2011...deal_with_three_us_national_laboratories.html

Anyways, Knights Corner, although not made as a discrete GPU, nonetheless has GPU functionality left in it. Intel may have dumped the discrete GPU usage role of those chips for desktops, but doesn't seem to be abandoning the HPC market which is shifting to GPU computing. The desktop GPU path just seems to be completely separate.