One of many extra intriguing mysteries within the Apple ecosystem has been the query over what course of the corporate would use for the A10X SoC, which is getting used within the newly launched 2017 iPad Professional household. Whereas the A10 used within the iPhone was a lot too early to make use of something however 16nm/14nm, the iPad Professional and A10X is coming in the course of the transition level for high-end SoCs. 16nm continues to be a excessive efficiency course of, but when an organization pushes the envelope, 10nm is accessible. So what would Apple do?
The reply, because it seems, is that they’ve opted to push the envelope. The intrepid crew over at TechInsights has lastly dissected an A10X and posted their findings, giving us our first in-depth take a look at the SoC. Most significantly then, TechInsights is confirming that the chip has been fabbed on TSMC’s new 10nm FinFet course of. Actually, the A10X is the primary TSMC 10nm chip to indicate up in a client machine, a really fascinating flip of occasions since that wasn’t what numerous manufacturing roadmaps referred to as for (that honor would have gone to MediaTek’s Helio X30)
Picture Courtesy TechInsights
Apple is after all identified for pushing the envelope on chip design and fabrication; they’ve the sources to take dangers, and the revenue margins to cowl them ought to they not pan out. Nonetheless, that the A10X is the primary 10nm SoC is an particularly fascinating growth as a result of it’s such a high-end half. Historically, smaller and cheaper components are the primary out the door as these are much less impacted by the inevitable yield and capability challenges of an early manufacturing node. As a substitute, Apple appears to have gone comparatively huge with what quantities to their 10nm pipecleaner half.
I say “comparatively huge” right here as a result of whereas the A10X is a robust half, and large for a 10nm SoC, by way of absolute die measurement it’s not all that huge of a chip. Actually by Apple X-series SoC requirements, it’s downright small: simply 96.4mm2. That is 24% smaller than the 16nm A10 SoC (125mm2), and in reality is even 9% smaller than the A9 SoC (104.5mm2). So not solely is it smaller than any of Apple’s 16nm SoCs, but it surely’s additionally about 20% smaller than the next-smaller X-series SoC, the A6X. Or, if you wish to examine it to the earlier A9X, Apple’s achieved a 34% discount in die measurement. In different phrases, Apple has by no means made an iPad SoC this small earlier than.
One key distinction right here nonetheless is that the X-series SoCs have by no means earlier than been the main half for a brand new course of node. It has all the time been iPhone SoCs which have lead the cost – A9 at 16nm, A8 at 20nm, A7 at 28nm, and so on. This does imply that as a pipecleaner half, Apple does should be particularly aware of the dangers. If an X-series SoC is to guide the cost for the 10nm technology, then it will probably’t be allowed to be too huge. Not that this has stopped Apple from packing in three CPU cores and a 12-cluster GPU design.
Talking of measurement, TechInsights’ estimates for space scaling are fairly fascinating. Based mostly on their accounting, they consider that Apple has achieved a 45% discount in characteristic measurement versus 16nm, which is in step with a full node’s enchancment. That is in step with TSMC’s earlier statements, however given the challenges concerned in bringing newer processes to market, it’s none the much less thrilling to truly see it occurring. For chip distributors designing merchandise towards 10nm and its 7nm sibling, that is excellent news, as small die sizes are the rule for just about everybody in addition to Apple.
A10X Structure: A10 Enlarged
Diving a bit deeper, maybe the largest purpose that A10X is as small as it’s, is that Apple appears to have opted to be conservative with its design. Which once more, for a pipecleaner half, is what you’d need to do.
|Apple SoC Comparability|
(Hurricane + Zephyr)
|2x Tornado||3x Storm||2x Swift|
|GPU||12 Cluster GPU||PVR 12 Cluster Series7||Apple/PVR GXA6850||PVR SGX554 MP4|
|Typical RAM||4GB LPDDR4||4GB LPDDR4||2GB LPDDR3||1GB LPDDR2|
|Reminiscence Bus Width||128-bit||128-bit||128-bit||128-bit|
|Manufacturing Course of||TSMC 10nm FinFET||TSMC 16nm FinFET||TSMC 20nm||Samsung 32nm|
We all know from Apple’s official specs that the A10X has three Fusion CPU core pairs, up from 2 pairs on A10, and a pair of Tornado CPU cores on A9X, all with 8MB of L2 cache tied to the CPU. In the meantime the GPU in A10X is comparatively unchanged; A9X shipped with a 12 cluster GPU design, and so does A10X. Which means that Apple hasn’t invested their die house features from 10nm in a lot of the best way of further . To make sure, it’s not only a smaller A9X, but it surely’s additionally not the identical form of generational leap that we noticed from A8X to A9X or related iterations.
Sadly TechInsights’ public die shot launch isn’t fairly large enough or clear sufficient to attract an in depth floorplan from, however at a really excessive stage we will make out the 12 GPU clusters on the left, together with the CPU cores to the suitable. Considerably, there aren’t any actual surprises right here. TechInsights closely compares it to the A9X and there’s good purpose to take action. IP blocks have been up to date, however the one main change is the CPU cores, and people don’t take up numerous die house relative to the GPU cores. That is what permits A10X to be extra highly effective than A9X whereas having fun with such a big die measurement lower.
As for the GPU particularly, Apple as of late is not formally specifying whether or not they’re utilizing Creativeness’s PowerVR structure of their chips. Moreover we all know that Apple is creating their very own GPU, impartial from Creativeness’s designs, and that it will likely be rolled out earlier than later. With that mentioned, even previous to at this time’s die shot launch it’s been slightly clear that A10X just isn’t that GPU, and the die shot additional proves that.
Apple’s developer documentation has lumped within the A10X’s GPU with the remainder of the iOS GPU Household three, which contains all the A9 and A10 household SoCs. So from a feature-set perspective, A10X’s GPU isn’t bringing something new to the desk. As for the die shot, as TechInsights accurately notes, the GPU clusters within the A10X look nearly precisely just like the A9X’s clusters (and the A10’s, for that matter), additional indicating it’s the identical base design.
Picture Courtesy TechInsights
In the end what this implies is that by way of design and options, A10X is comparatively easy. It’s a correct pipecleaner product for a brand new course of, and one that’s geared to take full benefit of the die house financial savings versus spending these financial savings on new options/transistors.
In any other case I’m very curious as to simply what this implies for energy consumption – is Apple gaining a lot there, or is all of it space features? A10X's CPU clockspeed is barely marginally greater than A9X's, and just about equivalent to A10, so we will see that Apple hasn't gained a lot in the best way of clockspeeds. So does that imply that Apple as an alternative invested any process-related features in decreasing energy consumption, or, as some theories go, has 10nm not considerably improved on energy consumption versus 16nm? However the reply to that should wait for an additional day.