I am more than certain that it's not 4 cores, but 4 threadsbelonging to two physical cores which are bundled with a Zephyr core each. SMT is low hanging fruit and it was just a question of time when apple would pick it.
The problem is SMT is power hungry. Further Apple has a ways to go with the CPU internals, as good as the cores are they still need more competitive IPC.
The distribution of cores seems to be still in question with this reveal but I wouldn't be surprised it there are 4 low power cores. The thought here is that more low power cores can handle more of the background behaviors Apple is building into the device. This would work out OK if the high power cores give use another 35% to 50% increase in compute power for application. Even 35%, in the high power CPU's, would be significant because we are sure to see a vastly enhanced GPU.
Whatever is delivered it is certain to be an interesting discussion.
You guys can't resist putting all the spoilers in the headlines. Instead of a headline like "Leak provides new insight into A11 chip" you just dumped the equivalent of "Leak reveals Darth Vader is Luke's father in Star Wars sequel" right in the headline. Is that really necessary? Can't you put the spoilers in the article and tag it with "potential spoiler"? I would only apply that to information that came from firmware leaks, stolen Apple prototypes--any asset that came form Apple, Inc. Hearsay, analyst musings, and unreliable reports from third parties still fall into rumor category, but this is not. You are obviously treating all info gleaned from the iOS 11 leak as fact, so you know it is not the same category as a Ming-Chi Kuo sourced story.
I'm not buying the idea there's 4 high performance cores. Maybe it's:
2 High performance cores 2 High efficiency cores 2 Nueral processing cores
Yeah, I'm in agreement, albeit the Neural processing cores will be fairly small in this generation; just enough to power Siri in standalone mode in the future.
Apple is building increased specializing onto the die, simply because they have the architectural team. the OS and the development system, to create the best tradeoff between performance and efficiency.
Remember a few weeks ago around here, when there were a couple of folks touting the Snapdragon 835 and 845, and the Kirin 970? I'm guessing that they will have to move the goal posts yet again.
Edit;
Okay, so "Longhorn" states that he has the details on the cpu's, and it's 4 big and 2 little, in which case, the Neural processing cores would have to be on the SOC if they exist. That would also imply that the interconnections among the 4 large CPU's would have to be of an improved, evolved, design, and I'm guessing at a new level of bandwidth; maybe we are seeing the first processor suitable for an ARM analog to the Mac Book, less x86 of course. Where would Apple go for an 11X version but to beef up the number of GPGPU's?
"Longhorn" just corrected himself. It's 2 big cores & 4 little cores
It would make a lot more sense this way. Taskes that need big core need them because mutli-core don't answer the problem very well. Taskes that do responed well to multi-core generally respond very well. Also having more smaller cores to handle the multitude of things your smartphone is doing at the same time without tripping over other would be good.
I see 4 low power cores making more sense due to Apple having more and more happening in background. Especially things like face recognition which will likely run in conjunction with specialized hardware. Then you have Siri, the normal apps running in background like Mail, the music player and whatever else they burden the system with. I can even see Apple opening up the possibility of one or two user apps running in background. It really comes down to just how low power they can get and frankly I think they have a ways to go before they hit the wall.
I'm not buying the idea there's 4 high performance cores. Maybe it's:
2 High performance cores 2 High efficiency cores 2 Nueral processing cores
Yeah, I'm in agreement, albeit the Neural processing cores will be fairly small in this generation; just enough to power Siri in standalone mode in the future.
Apple is building increased specializing onto the die, simply because they have the architectural team. the OS and the development system, to create the best tradeoff between performance and efficiency.
Remember a few weeks ago around here, when there were a couple of folks touting the Snapdragon 835 and 845, and the Kirin 970? I'm guessing that they will have to move the goal posts yet again.
Edit;
Okay, so "Longhorn" states that he has the details on the cpu's, and it's 4 big and 2 little, in which case, the Neural processing cores would have to be on the SOC if they exist. That would also imply that the interconnections among the 4 large CPU's would have to be of an improved, evolved, design, and I'm guessing at a new level of bandwidth; maybe we are seeing the first processor suitable for an ARM analog to the Mac Book, less x86 of course. Where would Apple go for an 11X version but to beef up the number of GPGPU's?
"Longhorn" just corrected himself. It's 2 big cores & 4 little cores
It would make a lot more sense this way. Taskes that need big core need them because mutli-core don't answer the problem very well. Taskes that do responed well to multi-core generally respond very well. Also having more smaller cores to handle the multitude of things your smartphone is doing at the same time without tripping over other would be good.
I see 4 low power cores making more sense due to Apple having more and more happening in background. Especially things like face recognition which will likely run in conjunction with specialized hardware. Then you have Siri, the normal apps running in background like Mail, the music player and whatever else they burden the system with. I can even see Apple opening up the possibility of one or two user apps running in background. It really comes down to just how low power they can get and frankly I think they have a ways to go before they hit the wall.
Apple has allowed multiple App Store apps using background resources since they first offered that many years ago and default apps have running in the background since day one, where appropriate.
Interesting that they're jumping to type 3 seemingly. When A10 was restricted to 2 cores at any time, I figured Apple didn't see a point to it and wasn't just in it for marketing core count like most others. Wonder if anything changed, like bandwidth, or if they wanted A10 to do that but didn't have time or something (not that it didn't crush at launch anyways).
2 high processing cores 2 efficiency cores and 2 cores dedicated to gpu?
GPU cores, and for that matter, potential neural processing ones, aren't identified as CPU0,1,2,3 etc.
So would you think they'd do a 3x3-core CPU setup?
4x2, and the reason would be the same as introducing Fusion in the first place, trying to put as much on the low performance cores as possible before turning on the big ones. The big ones may continue to take even more power, so only using them for lengthy tasks makes sense.
Being that supposedly the iPhone 8 can record in 4K 60fps, that chip must really be a beast. I guess there's nothing that can be thrown at the iPhone 8 that will slow it down. I think it's going to be hard for the Android manufacturers to keep up in terms of performance and still make decent profits. They'll be buying smaller quantities of SnapDragon 835s and paying more for them, yet selling fewer flagship models than Apple. I would think most mid-range Android smartphones aren't going to have enough processing power to be doing AR on Apple's level, so Apple should be able to dominate in terms of the number of smartphones being able to run AR apps. Apple already has a large customer base of iPhones than can immediately run AR apps and now come the A11 Fusion beasts. It appears there's going to be an iPhone AR storm on the horizon.
If Wall Street still yawns at the high-end iPhone X, then I suppose there's nothing Apple can do to impress anyone.
I am pretty sure that the limitation for 4k 60fps today is not a CPU/GPU pair. It is the camera module that gives off that data, to begin with.
When I was playing with the code for my app in 2015, it would require about 30-40% of the total CPU performance of iPhone6 to be able to record full frame low compressed footage (8Mpx) at 30 fps without dropping any frames.
I am pretty sure it could have pulled off recording at 60fps too, albeit probably would have required a slightly faster working memory.
As for CPU and GPU, they were fine for that task even in iPhone6 family.
SOOO... what's the significance of the A11, 6 cores with "no Fusion?"
Huge!. In A10, you dont get to choose ( Well you can but let simplify things here ) which core to go. So despite there are 2+2 Core, there is only "2" Core active at any time, depending on work load.
Now it has 2 + 4 Core, and you can program to use them all at the same time. If done correctly this is much better then SMT / Hyper threading.
But previously the 2+2 Core was done because there is a believe that programmers shouldn't and wont know which one to use. That is why there is a Hardware Switch to automatically choose which Core to use on the fly. Even Linus thinks this is a better way to handle it rather then the big.LITTLE implementation.
Now they have gone back, I wonder why, there must be very good reason behind this, Automatic Compiler optimization? Still having a hardware switch in place? Contention?
Note: One thing i would like to point out, ARM were never keen on SMT. It added complexity to the uArch and lots of testing needs to be done to get things right. And in modern SoC the actual size of the CPU core are already tiny compared to SRAM and GPU, it doesn't make sense to add all this to complexity for extra performance per transistor count.
SOOO... what's the significance of the A11, 6 cores with "no Fusion?"
Huge!. In A10, you dont get to choose ( Well you can but let simplify things here ) which core to go. So despite there are 2+2 Core, there is only "2" Core active at any time, depending on work load.
Now it has 2 + 4 Core, and you can program to use them all at the same time. If done correctly this is much better then SMT / Hyper threading.
But previously the 2+2 Core was done because there is a believe that programmers shouldn't and wont know which one to use. That is why there is a Hardware Switch to automatically choose which Core to use on the fly. Even Linus thinks this is a better way to handle it rather then the big.LITTLE implementation.
Now they have gone back, I wonder why, there must be very good reason behind this, Automatic Compiler optimization? Still having a hardware switch in place? Contention?
Note: One thing i would like to point out, ARM were never keen on SMT. It added complexity to the uArch and lots of testing needs to be done to get things right. And in modern SoC the actual size of the CPU core are already tiny compared to SRAM and GPU, it doesn't make sense to add all this to complexity for extra performance per transistor count.
Excellent explanation, thank you. Although I must admit, upon reading I now have tons of follow up questions
Comments
The distribution of cores seems to be still in question with this reveal but I wouldn't be surprised it there are 4 low power cores. The thought here is that more low power cores can handle more of the background behaviors Apple is building into the device. This would work out OK if the high power cores give use another 35% to 50% increase in compute power for application. Even 35%, in the high power CPU's, would be significant because we are sure to see a vastly enhanced GPU.
Whatever is delivered it is certain to be an interesting discussion.
Nooooooooooooo !!!!!!!!!!!!!!!!
I see 4 low power cores making more sense due to Apple having more and more happening in background. Especially things like face recognition which will likely run in conjunction with specialized hardware. Then you have Siri, the normal apps running in background like Mail, the music player and whatever else they burden the system with. I can even see Apple opening up the possibility of one or two user apps running in background. It really comes down to just how low power they can get and frankly I think they have a ways to go before they hit the wall.
Interesting that they're jumping to type 3 seemingly. When A10 was restricted to 2 cores at any time, I figured Apple didn't see a point to it and wasn't just in it for marketing core count like most others. Wonder if anything changed, like bandwidth, or if they wanted A10 to do that but didn't have time or something (not that it didn't crush at launch anyways).
4x2, and the reason would be the same as introducing Fusion in the first place, trying to put as much on the low performance cores as possible before turning on the big ones. The big ones may continue to take even more power, so only using them for lengthy tasks makes sense.
They're min/maxing it so to speak lol.
Makes more sense than 8 A53 SoCs at any rate.