Analysis affirms Apple's A7 processor closer to a desktop CPU than regular mobile chip
An independent analysis of the microarchitecture behind Apple's latest A7 processor has shown that the company was not overstating when it called the design "desktop class," with the new silicon matching up well against Intel's recent desktop components.
The A7 sports the same number of execution ports as Intel's Ivy Bridge chips and a reorder buffer equal to that found in the Haswell architecture, according to Anand Shimpi of AnandTech. Shimpi arrived at his conclusions by studying the A7 itself as well as Apple code commitments to the LLVM compiler project.
"Apple didn't build a Krait/Silvermont competitor, it built something much closer to Intel's big cores," Shimpi wrote, referring to Intel and Qualcomm's ultra-mobile CPU designs. "At the launch of the iPhone 5s, Apple referred to the A7 as being "desktop class" - it turns out that wasn't an exaggeration."
The number of execution ports is important because it defines how many instructions the processor can handle concurrently. Apple's A7 can process six instructions per clock cycle, the same as Intel's Ivy Bridge chips found in previous-generation Apple laptops and twice the capacity of the A6.
Similarly, a larger reorder buffer gives the processor a bigger pool of instructions to choose from when deciding how to most efficiently complete its tasks. The A7's 192-instruction buffer matches Intel's Haswell designs and is more than four times the A6's 45-instruction buffer.
Apple's SVP of worldwide marketing Phil Schiller introduces the A7 system-on-chip.
Shimpi believes that the A7 was designed to be forward-looking, with room to increase performance as Apple moves to smaller fabrication processes. He also raised the possibility that Apple may choose to release yet another new architecture design with the A8, rather than simply refining the A7's "Cyclone" core.
As it stands, Shimpi added, most of the A7's processing power remains untapped due to battery life concerns. Current-generation iOS devices will run out of RAM, he predicts, long before reaching the A7's performance ceiling.
When it debuted in the iPhone 5s last fall, Apple's custom A7 processor was said to have caught the chipmaking industry off guard, sending competitors into an alleged "panic." One unnamed person at Qualcomm reportedly said that the A7 left the company "slack-jawed, and stunned, and unprepared."
The A7 sports the same number of execution ports as Intel's Ivy Bridge chips and a reorder buffer equal to that found in the Haswell architecture, according to Anand Shimpi of AnandTech. Shimpi arrived at his conclusions by studying the A7 itself as well as Apple code commitments to the LLVM compiler project.
"Apple didn't build a Krait/Silvermont competitor, it built something much closer to Intel's big cores," Shimpi wrote, referring to Intel and Qualcomm's ultra-mobile CPU designs. "At the launch of the iPhone 5s, Apple referred to the A7 as being "desktop class" - it turns out that wasn't an exaggeration."
The number of execution ports is important because it defines how many instructions the processor can handle concurrently. Apple's A7 can process six instructions per clock cycle, the same as Intel's Ivy Bridge chips found in previous-generation Apple laptops and twice the capacity of the A6.
Similarly, a larger reorder buffer gives the processor a bigger pool of instructions to choose from when deciding how to most efficiently complete its tasks. The A7's 192-instruction buffer matches Intel's Haswell designs and is more than four times the A6's 45-instruction buffer.
Apple's SVP of worldwide marketing Phil Schiller introduces the A7 system-on-chip.
Shimpi believes that the A7 was designed to be forward-looking, with room to increase performance as Apple moves to smaller fabrication processes. He also raised the possibility that Apple may choose to release yet another new architecture design with the A8, rather than simply refining the A7's "Cyclone" core.
As it stands, Shimpi added, most of the A7's processing power remains untapped due to battery life concerns. Current-generation iOS devices will run out of RAM, he predicts, long before reaching the A7's performance ceiling.
When it debuted in the iPhone 5s last fall, Apple's custom A7 processor was said to have caught the chipmaking industry off guard, sending competitors into an alleged "panic." One unnamed person at Qualcomm reportedly said that the A7 left the company "slack-jawed, and stunned, and unprepared."
Comments
Battery life and weight on the Air is already terrific, and performance has been improving. Why harm performance by going to ARM, around the same timeframe as MB Airs NEED more performance due to going retina (whenever that happens)?
And your Intel software would no longer run. And workarounds would involve major work for developers, and fat binaries what waste expensive SSD space.
So it would mean MASSIVE fragmentation for developers, and massive headache for users. The kind of thing you only do if the benefit is HUGE, or if you HAVE to (like the PPC->Intel transition). And you do it for ALL the Macs in the lineup. ARM-based MacBook Pros, iMacs and Mac Pros? Makes no sense any time soon.
Apple already has an ARM-based portable that IS a good idea. They don't need to add one that isn't.
Huh? Why would you think Apple will use last years' tech in the MBA?
Exactly. PowerPC–>Intel made a ton of sense. Intel–>ARM doesn’t, and won’t, for a very long time without drastic changes.
MacBook Air? No.
Battery life and weight on the Air is already terrific, and performance has been improving. Why harm performance by going to ARM, around the same timeframe as MB Airs NEED more performance due to going retina (whenever that happens)?
And your Intel software would no longer run. And workarounds would involve major work for developers, and fat binaries what waste expensive SSD space.
So it would mean MASSIVE fragmentation for developers, and massive headache for users. The kind of thing you only do if the benefit is HUGE, or if you HAVE to (like the PPC->Intel transition). And you do it for ALL the Macs in the lineup. ARM-based MacBook Pros, iMacs and Mac Pros? Makes no sense any time soon.
Apple already has an ARM-based portable that IS a good idea. They don't need to add one that isn't.
I really have to agree with you. Not much to add to that except some people may say it can compete with a chromebook (which is kinda worthless). But Apple could never price low enough to hurt the small chromebook market. There really doesn't seem much for Apple to gain other than making a more PC-like iPad. Which would be cool but I can imagine the price would be in the ~700 or ~800 range and by that point, might as well go up a bit more and grab the Macbook Air (which is not a laptop for me). I prefer my nice 15.4'' retina macbook pro.
All in all, I agree with you.
Well that made me sit up.
Possibilities galore. 8-)
According to the article the A7 is hampered by the amount of RAM available to it... which, in turn is limited by the drain on the battery more RAM would add. That's my take-away..
So it would mean MASSIVE fragmentation for developers, and massive headache for users.
I agree. Fragmentation is bad. Apple will not just convert MBA to A7 chip and keep Intel on MBP, I believe. Can the "Ax" chip become powerful enough for MBP in the next iteration? It would be interesting if it can. How about MacPro? Somehow, it feels hopeless for Apple to stop using Intel chips.
Could Apple stick 2X A8s in a slim MacBook Air?
The increased thermal headroom of a MacBook Air compared to an iPhone or iPad would allow a design to run at faster speeds for more of the time (most of the processing time in an iPhone or iPad is actually spent in "race to idle" mode - long battery lives are only achieved by being idle most of the time). I don't think they would use two A8s, they would probably fab an A8X or clock the A8 faster due to the increased thermal headroom.
But I don't think Apple is in the right place yet to release a ARM device running Mac OS X - all the software is compiled for x86-64 currently, not ARM. There would have to be a period of time where ARM was enabled for all software builds in XCode for Mac OS X. I believe Apple uses the threat of ARMing themselves to make Intel provide its CPUs at a reasonable price.
Yes, I agree. And the other limiting factor is that A7 is currently underclocked (with respect to its capability), to further reduce battery drain.
I'm not bright enough to say what this might look like or do, but if matched with the right new software, it could define a new market as fresh as the iPad was...
I wasn't thinking about last year tech but this new A8 that they mentioned.
Anyways we'll see...
For use in the new Macbook Air, or not?
No.
A crazy idea: A docking station with screen and keyboard, portable or desktop. The iPhone is the computer. For many people it's all the computing power they would ever need. Crazy, but well within the realm of possibility, and already predicted by others. It could even be an attractive, inexpensive option for the enterprise. Lots of docks, and everyone carries their own computer in their pocket.
"The challenge is that at full tilt a pair of Cyclone cores can consume quite a bit of power. So for now, Cyclone's performance is really used to exploit race to sleep and get the device into a low power state as quickly as possible. The other problem I see is that although Cyclone is incredibly forward looking, it launched in devices with only 1GB of RAM. It's very likely that you'll run into memory limits before you hit CPU performance limits if you plan on keeping your device for a long time."
The overall impression from this text is that Cyclone, and the iPhone/iPad, are somehow unbalanced -- that there is a lot of performance potential that isn't being fully realized, and might never be fully realized in existing devices. Yet also contained in this text is the explanation for why Cyclone makes perfect sense: "Cyclone's performance is really used to exploit race to sleep and get the device into a low power state as quickly as possible." Anand makes it sound like this is a second or third order concern, but in the context of a mobile device, it's actually one of the most important concerns. Apple is trying to design a chip that gets its work done quickly, using as little power as possible. Given the physics confronting CPU designers, that means that you'd rather have more transistors running at a lower clock speed than the reverse. And given the nature of the software run on mobile devices, you'd rather have greater instruction level parallelism than thread level parallelism.
So the A7 is the perfect mobile SOC -- low clockspeed, high ILP, low TLP. it is perfectly balanced for its job. The fact that the cyclone core could also be the basis for a very credible desktop CPU is what's secondary here.