2014年9月23日星期二

Chipworks Disassembles Apple's A8 SoC: GX6450, 4MB L3 collection & new

Chipworks Disassembles Apple's A8 SoC: GX6450, 4MB L3 collection & new

Individual of the new enjoyable rituals with Apple’s yearly iPhone launch is the decapping, deconstruction, and photographing of the CPU depart this life on the feeling of Apple’s newest SoC.  While we can understand a group all but the SoC from software, in favor of around things there’s entirely veto replacement in favor of looking on the hardware itself and together with the functional blocks hand over. And this day, in the role of voguish gone years, the honor of being the in the beginning to tear apart the SoC goes to Chipworks.

In favor of determining the blueprint of A8, Chipworks reached elsewhere to us to solicit our input on their depart this life shot, and following around rounds of untaken back and forth we believe we’ve take place to a solid determination of around of A8’s skin and how it has been configured. So let's dive voguish.

In the beginning and foremost we’ll start with A8’s GPU, in the role of this was individual of the hardest elements to consider voguish software. Based on Apple’s 50% performance convalescence we had previously speculated with the purpose of A8 restricted an Imagination PowerVR GX6650. However in the role of we well-known back next, a depart this life shot would make public all, and well on schedule it has.

A close analysis of the depart this life shot makes it plain with the purpose of in attendance are no more than 4 GPU cores on hand and not 6, which instantly rules elsewhere the 6 nucleus GX6650 we were previously expecting. In its place with 4 cores hand over this is conclusive resistant with the purpose of Apple is using the less important 4 nucleus GX6450 on A8, the give directions successor to the G6430 used on the A7. GX6450 induces around performance optimizations along with around appear updates – counting ASTC support, which Apple’s citations has already incorrigible is hand over – so its inclusion at this juncture is a natural movement in favor of Apple.

On A8 and its 20nm process this measures on 19.1mm2, versus A7’s 22.1mm2 G6430. In the role of a conclusion Apple is saving around depart this life room compared to A7, but this is being to some extent offset by the greater complexity of GX6450 and perhaps further SRAM in favor of superior caches on the GPU. Meanwhile looking on the symmetry of the blockade, it’s motivating with the purpose of the blocks of texturing means with the purpose of all two of a kind of GPU cores share is so visible and so huge. With these means being so vast family member to the GPU cores themselves, you can go with why Imagination would like to share them in the role of divergent to building them 1:1 with the GPU cores.

Meanwhile opposite the GPU we take the CPU blockade. Unlike the GPU the CPU blockade has seen around momentous reduction, which Chipworks estimates is down from 17.1mm2 voguish A7 to 12.2mm2 voguish A8. Voguish A7 twister did not let somebody have temporarily itself to certainly alternative apart the personage CPU cores, and neither does the CPU at this juncture voguish A8. We’ll be real looking on the just starting out CPU’s architecture in-depth voguish our iPhone 6 reassessment, but in favor of at once it’s safe to say with the purpose of while this is categorically derived from twister, Apple has added a little tweaks greater than the ultimate day with the purpose of formulate it an even new forceful CPU than the in the beginning twister. Meanwhile based on this depart this life shot Chipworks believes with the purpose of the L2 collection has been simplified to a per-core design, in the role of in attendance is veto obvious single blockade of L2 on A8 like in attendance was A7.

The final foremost specialized blockade on A8 is time was again the SRAM collection recollection. On A7 we exposed with the purpose of this blockade was 4MB and was answerable in favor of servicing the GPU and CPU. On A8 this blockade is similarly hand over and serving the same role. This 4MB of SRAM tops up being quite vast despite the dwindle from 28nm to 20nm, and while on in the beginning glance it seems like it be supposed to be real superior than 4MB set the family member size, voguish practice could you repeat that? Has happened is with the purpose of the personage SRAM cells take not shrunk by a rounded 50%. Chipworks estimates the cell size to at once be real all but 0.08µm2, versus 0.12µm2 on A7, which is closer to a 33% dwindle with the purpose of a 50% dwindle. In the role of a conclusion the SRAM collection still takes up a good spot of room, but the respect of being able to work for superior recollection needs lacking having to verve off-die continues to be real immense.

Overall, Chipworks’ analysis points to A8 being fabbed on TSMC’s 20nm process. This makes A8 together with the in the beginning SoCs to receive the 20nm healing. Merit to this less important node Apple has been able to build voguish further skin to the SoC while at once chip rotten around 15% of their depart this life size. Chipworks estimates the final depart this life size of A8 to booth on 89mm2, versus the 104mm2 in favor of the Samsung 28nm based A7. Chipworks explanation with the purpose of if this were a straight dwindle with the purpose of individual would expect the A8 to be real closer to 50% the size of A7 (though not all logic can dwindle quite with the purpose of well), which indicates with the purpose of Apple has spent quite a spot of depart this life size on civilizing performance through new knotty CPU and GPU architectures and miscellaneous appear additions.

Wrapping things up, we’ll be real back later on this month with our reassessment of the iPhone 6 children and our rounded analysis of the A8 SoC. So until next stay tuned.

Tags : Apple , SoC



没有评论:

发表评论