A14 chip could make the iPhone 12 as powerful as the 15-inch MacBook Pro

This is a big upgrade. The 5 nm mode is not a half-step by any stretch, but it is the next “full node” after 7nm […]
If we take TSMC at its word about the improved transistor density of the 5 nm process, we’re looking at an incredible 15 billion transistors. That’s more than all but the largest high-end desktop and server CPUs and GPUs. It’s huge. It’s so big that I wouldn’t be entirely surprised if Apple shrunk the total chip area a bit to around 85 mm square  and around 12.5 billion transistors.
This could enable Apple to do better than the trend line would suggest when it comes to multi-core performance in particular.
The trend line gives us a score around 4,500, but I think a combination of architectural changes and clock speed will give us a lot more. I wouldn’t be surprised if the Geekbench 5 multi-core score creeps up to 5,000 or so.
For what it’s worth, the fastest Android phones score around 3,000 on this test, and a score of 5,000 would be similar to 6-core mainstream desktop CPUs or high-end laptop CPUs. It’s 15-inch MacBook Pro territory.
More transistors in the GPU and the rumored (faster) 6 GB RAM could see games performance boosted by about 50%.
I expect Apple to expend significant transistor budget making the GPU more powerful. Along with more memory bandwidth, we can probably expect GPU performance well beyond the trend line’s prediction of the low 7,000 range in this test. Barring some new performance bottleneck, I think a score over 9,500 is certainly possible. In other words, I think we can expect a 50% improvement in graphics performance for the kind of high-end graphics used in games.
Cross expects even more of an improvement in another area of the A14 chip: the neural engine.
With the higher transistor budget afforded by the 5 nm manufacturing process, I think Apple will add Neural Engine cores this time, and may make other architectural improvements as well. It wouldn’t surprise me to see Apple claim that machine learning tasks are at least twice as fast as on the A13

Post a Comment

0 Comments