At the Apple keynote yesterday, folks like us were on the edge of our seats to finally discover what TSMC's 3nm technology would offer consumers in performance and efficiency. We expected much talk about a monumental leap forward for mobile performance, and we got it—just not where we thought. Apple focused almost solely on the GPU performance of its A17 Pro silicon while noting the uplift of its CPU is modest and a bit underwhelming given the node jump involved. Although, it's not like people needed their phones to be any faster for day-to-day use, so Apple's decision kind of makes sense.
At a high level, Apple has bumped up the number of transistors in the SoC from 16 billion to 19 billion, which is quite a bump given the size of the chip. It only added 1 billion transistors when it transitioned from the A15 to the A16, but it was also stuck on the same TSMC node for both designs: TSMC 5nm to 4nm.
Apple has kept the number and types of CPU cores from the A16 Bionic to the A17 Pro, sticking with two high-performance and four efficiency cores as before. Apple says its new performance cores are up to 10% faster than before, which can be attributed to increased logic density, clock speeds, or both. Apple says they can challenge "high-end, desktop PCs," but we'll need to see those benchmarks. It also says its efficiency cores are 3x more efficient than "the competition," so it's unclear if they've improved from the A16 Bionic.
The GPU, however, is where Apple focused a lot of its attention. It has gained a single core, going from five to six, which Apple says is 20% faster than the GPU in the A16 Bionic. The company created an entirely new shader architecture for A17 Pro, which it says is the biggest GPU redesign it has ever done. To that end, Apple says it's now a "pro-class GPU."
To prove it, the company showed it's now capable of hardware ray tracing, which is impressive on a phone. The company showed a phone running a ray tracing demo at 32fps, compared with an unplayable 8fps using software-based ray tracing. Apple says it's also added upscaling, allowing the phone to run games at high resolution while using less power.
Apple also gave a major shot in the arm to its Neural Engine, taking it from 16 trillion operations per second to 35 trillion despite featuring the same 16-core design as the previous chip. Apple says the Neural Engine performs on-device machine learning, which powers its new "personal voice" feature and removes photo wizardry and autocorrect when typing.
Overall, the A17 Pro seems like a big step forward for mobile performance, but for most of us, the M3 chips that utilize TSMC's 3nm process are where we'll see more real-world gains. Apple always touts next-gen gaming at its iPhone launches, but most people still play casual games on phones in the real world. Perhaps Apple wants to change that with the 15 Pro and Pro Max phones, as it has certainly paid attention to GPU performance with its shiny new wafers from TSMC.
