Comment by corresation
13 years ago
Most prior Apple designs have been close to stock ARM core implementations coupled with a PowerVR GPU. There hasn't been much if anything to steal, given that Samsung also licenses PowerVR GPUs and is also an ARM licensee.
The A7 is interesting in that it's the first mainstream 64-bit part, but I see Samsung being far more interested in the A53/A57 (designs that Samsung gets straight from ARM) than the A7. The A7, if rumors are true, is a sort of hybrid approach to use one of the next generation, 64-bit ARM cores early, similar to what Qualcomm does.
EDIT: ARM nomenclature is such a mess. The A7 runs ARMv7 or ARMv8 using a sort-of A57 dual-core architecture, but should not be confused with the ARMv7 Cortex-A7.
The A7 is not a hybrid approach. A7 is an ARMv8 chip (it implements the ARMv8 ISA), but not an ARM design (that's the difference between the processor license (can use ARM-designed cores) and the architecture license (can design own core implementing ARM ISA). Apple and Qualcomm have both licenses. There's nothing "hybrid" about using your architecture license.
ARMv8 provides two architectures: AArch64 and AArch32. Implementors can implement either or both. AArch64 is the brand new architecture while AArch32 is backwards-compatible with ARMv7-A. If both are implemented, it's possible to switch between AArch32 and AArch64 on the fly at specific change points, giving the ability to run AArch32 (= ARMv7) applications seamlessly on an AArch64 kernel (or an AArch32 guest os in an AArch64 hypervisor).
The A7 implements both, so do the A53 and A57.
The A7 is not a hybrid approach.
Unless you work at Apple in silicon design, you don't know that, and to be fair neither do I. But history has shown that every A# release gets greeted with incredible fanfare about the completely-custom CPU work at Apple, to later quietly get corrected when it turns out that it is at most a marginally derived ARM core.
I'm just going with history. Given that the A57 finished design last year, and started taping out early this year, it seems unlikely -- if not strategically risky -- that Apple just went their own way. From a pure performance perspective, ARM is hyping a clock-for-clock tripling of performance with the A57 over a Cortex-A15, or a quintupling of performance at a given power usage level.
>> Unless you work at Apple in silicon design, you don't know that, and to be fair neither do I. But history has shown that every A# release gets greeted with incredible fanfare about the completely-custom CPU work at Apple, to later quietly get corrected when it turns out that it is at most a marginally derived ARM core.
His hasn't been true since the A6, which (amongst others) chipworks confirmed is a full custom design, manually laid out even, which proves it is not a modified reference ARM design. I don't think Apple would invest hundreds of millions acquiring chip design shops, gradually move from increasingly customized reference designs to full custom designs, to throw their brand new Swift core out after one generation and start over with a reference design. So I think it's pretty safe to assume the A7 is a pimped up A6 that implements armv8.
I'm not sure why you're so skeptical about Apple designing their own ARM cores, they've been going further down this path since after acquiring Intrinsity and PA-Semi, and they are not the only ones doing this, Qualcomm, nvidia and previously TI also design full custom ARM cores.
1 reply →
OK, here's the miscommunication: due to your edit
> The A7 runs ARMv7 or ARMv8 using a sort-of A57 dual-core architecture
I understood your use of the word "hybrid" as applying to the ARMv7 compatibility in an ARMv8 chip (which there's nothing hybrid about, since it's part of ARMv8's features).
And apparently you intended it as "Apple's custom chips are relatively minor customizations of standard ARM designs", which you seem to think of as a hybrid between using standard designs and designing from scratch.
Carry on.
Or indeed the ARM7, which is ARMv4.