Slashleaks has what is supposedly a leak of the A11:

005c1ddygy1fio0aksg2oj30m80m8myf.jpg

005c1ddygy1fio0al29sbj30m80m8ac7.jpg


Is it possible to estimate the approximate die size of the A11 from these pictures (assuming they are real), even with a large error margin?
 

lanek

Veteran
Slashleaks has what is supposedly a leak of the A11:


Is it possible to estimate the approximate die size of the A11 from these pictures (assuming they are real), even with a large error margin?

Without any reference point, this will be hard, untill they have keep something identical .
 

Grall

Invisible Member
Legend
The inner set of pads should correspond reasonably well with the die, seeing as those are the power and ground connections... Scale might be an issue, although is not maximum pad density relatively standardized/constricted within a set range, for manufacturability reasons?

Then there's the remarkable fuzziness of those pictures, which suggest fakery at work.
 
The inner set of pads should correspond reasonably well with the die, seeing as those are the power and ground connections...
There's also a picture of the board this supposed A11 slots into https://pbs.twimg.com/media/DH_zYjxXUAEuRz4.jpg; screw hole placement is claimed to be consistent with the iPhone 7 iPhone 7s Plus. So I guess some kind of estimate should be possible.
Thanks for the info.

From the above information and the 7 Plus teardown I made a very quick and rough estimate of the A11 die size: ≈ 86 mm^2.

UPDATE: There's a better picture here, and I got ≈ 83 mm^2 from that picture.

UPDATE 2: From this picture of a 7 Plus PCB and the die size of the A10, I got ≈ 88 mm^2. I think it's reasonable to assume a die size of 80 something mm^2.
 
Last edited:

psurge

Regular
A10X is just 96.4 mm2, so that seems quite large if (close to) correct. Maybe we can expect substantial GPU improvements, or perhaps dedicated HW for deep learning?
 

Entropy

Veteran
Or it could be the first chip with their new GPU design, which may not be as area efficient as ImgTechs per fps on GfxBench. I agree, if it is just a cut down version of the A10x (and the size estimate is correct of course), it seems somewhat large, but not by a lot necessarily.
 
The iPhone (2017 OLED) [commonly referred to as the "iPhone 8"] PCB has apparently been leaked, and it appears to use the same SoC as the iPhone (2017 LCD 5.5") ["iPhone 7s"].

There could still be clock speed differences though.
 
Sorry I've got nothing ;)
But iMacmatian's estimate seems reasonable.

As for the name, mistral to me sounds more like an evolution of apple's 'little core' Zephyr, meaning 'a light wind'
Going from a Hurricane to a strong wind somehow doesn't sound right :)

As for the A11, I guess we'll find out the performance numbers tuesday, which may be the only interesting bit of the show, as it feels like everything else has already leaked
 
So a 4+2 design where all 6 can run independently would be very interesting because it means they changed the Architecture for their cores. I used to think the A10X was built on the successor of hurricane, because I didn't expect apple to go through the trouble of shrinking an existing design, but after taking a closer look (and this would reinforce that) I now believe A10X is still hurricane, and the new stuff is yet to come. Apple is essentially running a half-year tick-tock cadence, which is extremely impressive.

PS: You gotta love that twitter handle :D
 

Grall

Invisible Member
Legend
Why don't mobile processors use SMT? I've been waiting for years, and it just isn't happening.

Any particular reason why, or is it more like, there's no genuine need for it? :p
 

pcchen

Moderator
Moderator
Veteran
Subscriber
Why don't mobile processors use SMT? I've been waiting for years, and it just isn't happening.

Any particular reason why, or is it more like, there's no genuine need for it? :p

SMT is about hiding memory latency by switching between threads. Current mobile workloads tend to not have many concurrent working threads.
Mobile games could be a different animal, but on the other hand mobile phones can't be running at full speed for too long, otherwise it will drain the battery in a very short time (and also could burn the user's hand :) ). So I think we'll probably see more cores before SMT in mobile space (of course, some crazy SoC vendors already have 8 cores SoC, but still).
 

Gubbi

Veteran
Any particular reason why, or is it more like, there's no genuine need for it?

SMT is all about maximizing the performance of the silicon. In a phone environment you don't have the power to that anyway, so are better off with silicon that is mostly dark (10 cores, yay!!). SMT makes the most sense when you already have a very wide CPU core. With a wide core you have a greater chance of under-utilizing your execution units. SMT is a fairly cheap way to keep those idle execution units busy.

But as Pcchen says, multi threaded workloads are rare on phones; Or at least multi threaded workloads that require high single thread performance as well are rare. Nowadays I'd expect the Javascript engines in browsers and WebViews embedded in apps is where the biggest demand for high performance lies, - and those are strictly single threaded.

Cheers
 

Grall

Invisible Member
Legend
SMT makes the most sense when you already have a very wide CPU core.
Well, Apple's cores are already quite wide, 6-issue IIRC. Seeing as they tend to favor smaller SoCs over larger, I thought maybe it would be an easy solution to avoid having to drop in more cores... :p

Nowadays I'd expect the Javascript engines in browsers and WebViews embedded in apps is where the biggest demand for high performance lies, - and those are strictly single threaded.
Yeah, javascript... Is there a particular reason it doesn't scale across cores, or is it just that it was conceived at a time when there were only single-core computers and the overall design has been stuck since then?

If so, is there any efforts at overhauling javascript in an age where CPUs are getting increasingly multi-cored (up to what now, 28 in skylake-X?) And, as cores go up, individual core performance/single-thread tends to go down, thus counterproductive to what javascript in current incarnation needs... :p

SMT and power efficiency are two opposing things and why we'll never see it in a proper mobile design.
Why would SMT be power inefficient? It makes the core draw more power, but that doesn't automatically mean increased inefficiency.
 

Gubbi

Veteran
Yeah, javascript... Is there a particular reason it doesn't scale across cores, or is it just that it was conceived at a time when there were only single-core computers and the overall design has been stuck since then?

Javascript was designed and implemented in three weeks. Keeping the thread model to a single context was probably wise considering the average frontend developers' skill wrt. concurrency issues.,

If so, is there any efforts at overhauling javascript in an age where CPUs are getting increasingly multi-cored (up to what now, 28 in skylake-X?) And, as cores go up, individual core performance/single-thread tends to go down, thus counterproductive to what javascript in current incarnation needs...

Be careful what you wish for, you'll be loading webpages that max 8 CPU cores, because some developer don't know what they are doing.

Also, all massively multicored CPUs today turbo like crazy, exactly to accomodate the single thread scenario.
 
Top