/grumbles about typo-producing fifo overruns
Right. Back on topic.
The microarchitecture is, well, not the ISA : ) Apropos, there's currently one other VLIW-backend design, which morphs a frontend ISA to its mysterious native ISA:
Elbrus 2C. I hope to get one of
these in my mittens sometime next year.
Incidentally Denver did end up showing what you and I talked about and predicted a while ago, fantastic performance in a straight line so to speak, such as set benchmarks, but its performance is far more variable than the A8X, most likely due to the code morphing engine choking up on spaghetti code.
Also, that EA guy saying the iPhone 6 is near the 8th gen...All of my wat. Look at this, this is what's in the A8X, not even the A8 in the iPhone 6. I'd say they've closed in on and in some ways (feature set, not brute performance) surpassed the 7th gen HD twins, but the 8th gen is still a while off.
http://www.notebookcheck.net/Imagination-PowerVR-GXA6850.128993.0.html
Around the 200Gflop ballpark is just where the PS360 are sitting. Granted the GPU in the A8X in the Air 2 has to share its hypothetical 25GB/s bandwidth with the CPU, so it may be behind them there still, though the L3/graphics cache will help.
http://www.anandtech.com/show/8666/the-apple-ipad-air-2-review/2
And I only mention Gflops for a ballpark, I know performance with any such rating can be quite variable, but you'll also never see, say, a 200Gflop GPU outperforming a 1800 one.