@musicalmac said:
Anand says Apple's A7 is already on par with what we saw in the Core2Duo era, and it's likely the A8 will double CPU performance and as much as quadruple GPU performance (giant leaps that have been similar throughout Apple's SoC development). In fact, the A7 looks more like a Haswell chip than a mobile chip under a microscope, it's already 64bit and everything.
It's coming sooner than you all anticipate.
That's possible, but there are a number of factors that weighed in on previous generations that cannot be relied upon this time around. Let's take a look back on previous Ax SoCs.
A5 --> A6 --> A7
45nm --> 32nm --> 28nm
Cortex A9 (ARMv7 32-bit) --> Swift (custom ARMv7 32-bit) --> Cyclone (custom ARMv8 64-bit)
PowerVR 5 SGX543MP2 --> PowerVR 5 SGX543MP3 --> PowerVR 6 G6430MP4
The jump in CPU performance was due to first moving to a custom architecture, and then moving from ARMv7 to ARMv8. There is no new ARMv9 for Apple to utilize, so in order to double performance they would need to take advantage of a 20nm process (which is expected for the A8) to allow for higher clock speeds, and further customize their in-house ARMv8 architecture. Using LPDDR4 might also help reduce power consumption and improve performance, but we'll have to see if it will be ready for the mass market later this year.
There was a relatively small jump in GPU performance when moving from the A5 to the A6, but the A7 introduced a brand new PowerVR 6 architecture (replacing a GLES 2 based PowerVR 5/5 XT which had been around since Apple's A4). Again, there is no new PowerVR 7 architecture (with new API support) for Apple to utilize. Working with PowerVR 6 or more likely a 6 XT model such as the GX6450MP4, Apple might have to increase clock speed by a reasonable margin in order to achieve quadruple the GPU performance.
One other factor worth noting is newer versions of iOS have been known to improve performance, so an A6 running iOS 7 will outperform an A6 running iOS 6.
Log in to comment