Next-Gen iPhone & iPhone Nano Speculation
- Thread starter Arun
- Start date
ltcommander.data
Regular
We don't know that Apple has actually customized Rogue for the A7. I don't believe it's even been 100% confirmed it is Rogue, just that it seems to match.
I believe IMG provides options like different cache sizes that can be chosen so different implementations can perform differently even if they have the same official name and clock speed, but really wouldn't be considered custom like the SGX543MP4+ in the Vita.
While good CPU design requires as much expertise as the design of any part, GPU design requires a whole other level of specialization (and perhaps refinement) in order to produce a competitive processor. I have a hard time imagining Apple wanting to design a GPU architecture themselves rather than customizing a PowerVR part (in coordination with Imagination) to get the best fit for their SoCs. I suppose it's possible, though.
I give consideration to the notion that Apple did custom design work to change a PowerVR Rogue core for the A7 GPU only as one possible contributing explanation to A7's "interesting" benchmark numbers, but I believe it unlikely at best that they actually did that this generation.
I give consideration to the notion that Apple did custom design work to change a PowerVR Rogue core for the A7 GPU only as one possible contributing explanation to A7's "interesting" benchmark numbers, but I believe it unlikely at best that they actually did that this generation.
It would be startlingly impressive indeed if apple could go from zero to an IMG-beating inhouse GPU design.
Given that intel probably has as big a GPU design dept as anyone, and are at it arguably as long as anyone, and with their totally vertically integrated approach from design to process to manufacturing on class leading process, they find it as yet impossible to design out IMG in 22nm handheld, isn't it a big ask to imagine a relative startup being able to do so.
In apples experience, they have shown that there were big performance gains to be had in doing their own CPU design. Given their only non customized design options consisted of 2-3 CPU IPs from one company (ARM), it's not difficult to see how they could identify such power/performance savings early on.
Apple is a major contributor to IMG revenue, in fact last year they accounted for more than half of IMG's entire technology revenue, between licensing and royalty. Although extremely important to IMG, the revenue stream is insignificant for apple, amounting to something like 30-35c per chip average plus licensing fee. Any decision to move away would be, IMO not for financial reasons.
Given many would say that IMG have demonstrated that they have the best mobile graphics IP out of 6 independant separate design houses ( IMG, Intel, Qualcomm, Mali, Nvidia, Vivante), and that they are effectively already acting as Apple's pseudo in house graphics dept, it's not at all easy to see how apple might identify a clear technological advantage in going inhouse, that all of the graphics design depts have missed.
Last edited by a moderator:
And drivers, though that should presumably be less of a problem for Apple.
It's typical for licensed IP to have some configurable parameters. How many varies from product to product. Some make configurability a big selling point - MIPS played up its cores this way and recently Charlie mentioned this in his random gushing over Vivante.
In this case it could just be that Apple doesn't want to disclose the core family for some reason, or wants to claim more Apple branding. On the other extreme, it could be a version specially customized for Apple. They're probably IMG's largest customer, so they could probably swing this quite a bit.
ltcommander.data said:
That's true, but I don't think Apple could have a GPU that supports OpenGL ES 3.0 and PVRTC otherwise. Unless they licensed PVRTC from IMG.
I wonder if it's legal to include in the drivers conversion from PVRTC to something else..
One thing's for sure, if IMG loses Apple it's going to be a major upset to them.
Well, they do own over 12,5% stake in them. Why not just buy them?
Does intel still own a decent chunk?
They also own some stuff that's probably not terribly interesting to Apple, like MIPS processors.
Dominik D
Regular
Apple seems to like performance and this would be super slow. I don't think they'd go in this direction.
If Wikipedia is to be believed (I have no idea how accurate WP is in this regard), Apple owns less than 10% and Intel owns more than 16%. There are probably way larger investors which many or may not want to sell their shares. In order for Apple (or anyone else) to get more shares someone else would have to be able and willing to sell.
I don't know how high institutional ownership is in IMG, but any investor will sell, for the right price. And Apple sure can afford it.
The bigger problem in my opinion would be anti-trust issues. IMG is the largest mobile graphics IP licensing house and ARM is the only other one with significant market share. You could count Vivante, which apparently has about 10%, after that it's just DMP, which is a footnote, and technically NVIDIA, which has zero market share as far as public information goes, at least for now.
Apple acquiring IMG would very severely disrupt the market and hinder competitors.
The latest figures are 14% held by Intel, 8.6% held by Apple. Both took the stakeholdings within a few weeks of each other, generally viewed as defensive holdings.
They are the only two tech companies holding notifiable interests in IMG. The rest are mostly held by institutions.
I don't agree.
While there are good reasons to assume that Apple is not going to design their own GPUs, it is also true that they have been quite active in recruiting graphics competence for some time. It is difficult not to see similarities with how many insisted that they would be content simply integrating different IP blocks and would never design their own cores. Now, for CPUs staying with an ISA rather than rolling your own is more important than with a GPU that is pretty much always accessed through an API so they are definitely more free to go their own way with the GPU parts of their chips.
There is obviously a reason why they have been recruiting graphics people.
It could be to maintain a high level of competence in their SoC group.
It could be to modify existing IP to fit their SoCs in terms of on chip communication, power draw optimization on the hardware level, et cetera.
It could be to research and/or design their own GPUs, for whatever reason.
We just don't know. The competence is there though, for all of the above.
As I said earlier up in the thread, its easy to see how Apple would have viewed the offerings from ARM as limited. In the timescale of the last two A series socs, the options were A15, A7 or combos of those. I don't know much at all about CPUs, but most of what I read says that A15 was primarily designed for higher up than mobiles. Fundamentally, ARM is providing generic CPU IP.
So pull enough people together, throw enough resources at it, and I guess it is comparatively easy to take an ARM arch license and design something that more meets your needs than the limited options available.
I'm not sure that the same scenario at all applies to the GPU realm, particularly when I'd strongly suggest that IMG are already designing GPU roadmaps, IP and implementations to exactly target Apple end use case, undoubtedly with heavy input from Apple themselves.
I'm currently in the camp of not impossible, but quite unlikely (ref: Apple going in-house on GPU).
So pull enough people together, throw enough resources at it, and I guess it is comparatively easy to take an ARM arch license and design something that more meets your needs than the limited options available.
I'm not sure that the same scenario at all applies to the GPU realm, particularly when I'd strongly suggest that IMG are already designing GPU roadmaps, IP and implementations to exactly target Apple end use case, undoubtedly with heavy input from Apple themselves.
I'm currently in the camp of not impossible, but quite unlikely (ref: Apple going in-house on GPU).
Another piece of food for thought:
When it comes to control of their graphics IP, Apple is actually an outlier in having to rely on an outside source. Qualcom, Intel, even AMD, use in-house solutions in their SoCs. And it's easy to see why as the GPU is a major component in terms of die size, cost and power draw, and the trend is tighter integration between CPU and graphics.
I'm simply saying that I doubt the graphics design group they have assembled in Orlando are sitting around idle, and that Apple relying on an outside source for their graphics IP is atypical. I wouldn't be so quick in dismissing the notion that Apple may start to roll their own, particularly as the binary compatibility that is so important on the CPU side is much less of an issue for their graphics.
When it comes to control of their graphics IP, Apple is actually an outlier in having to rely on an outside source. Qualcom, Intel, even AMD, use in-house solutions in their SoCs. And it's easy to see why as the GPU is a major component in terms of die size, cost and power draw, and the trend is tighter integration between CPU and graphics.
I'm simply saying that I doubt the graphics design group they have assembled in Orlando are sitting around idle, and that Apple relying on an outside source for their graphics IP is atypical. I wouldn't be so quick in dismissing the notion that Apple may start to roll their own, particularly as the binary compatibility that is so important on the CPU side is much less of an issue for their graphics.
Samsung, Mediatek, HiSilicon rely on outside source(s).
anexanhume
Veteran
Update on SRAM - they now say 4MB http://www.chipworks.com/en/technical-competitive-analysis/resources/blog/inside-the-a7/
Which is what I suspected in the first place. They still think it's for the fingerprints, and thus, posit it's completely lost when the device battery dies. I'm still skeptical.
The PoP memory interface also nearly doubled. No longer dual 32 bit channels?
Which is what I suspected in the first place. They still think it's for the fingerprints, and thus, posit it's completely lost when the device battery dies. I'm still skeptical.
The PoP memory interface also nearly doubled. No longer dual 32 bit channels?