Nov 8, 2020
What to expect from Apple Silicon GPUs?
A direct comparison of Metal performance across Apple’s platforms today and what we can extrapolate about the future.
GPUs are an embarassingly parallel problem to solve. If you increase the number of cores on your chip, you get a proportional amount of performance gain. The restriction hence is the amount of space on your chip.
Apple’s A14 series of chips are based on TSMC 5nm manufacturing process which gives the largest amount of transistors in the same space that we know of today. That means when it comes to packing many more GPU cores on the same silicon die, Apple is in the right spot to make it happen.
Another benefit of GPUs is that Apple’s Metal framework (analogous to OpenGL, Vulcan and DirectX on Windows) is available across macOS, iPadOS and iOS. Hence a Metal benchmark actually serves as a decent performance comparison.
Year-on-year improvements made by Apple gives us a good sense of extrapolating what to expect. So let’s get started!
The X variants
Occasionally Apple takes an A series chip and makes an X variant of it. Typically this includes more CPU and GPU cores. Let’s take a look at the history here:
- A10 (3196) and A10X (6956)
- A12 (5343) and A12X (10936)
- A14 (12513)
Apple is typically able to double the number of GPU cores in these eXpanded variants, and also double the performance. With this, I predict that the entry-level MacBook Air and next iPad Pro both will push A14X at around 26000 GeekBench score.
How does this compare to Intel and AMD today?
First the portable lineup scores:
and then the desktop scores:
Key Takeaways
- Intel Integrated Graphics are a joke. The high-end MacBook Pro 13" maxes out at 8498, while the iPad Air with A14 is pushing 12513!
- Entry level graphics in base model of MacBook 16" (the only portable with discrete GPUs) and iMac 21" from AMD are at par with our hypothetical A14X at 26000 score.
This lines up with the expectation that we are going to see new MacBook Air and MacBook Pro 13" this year with the graphics performance of today’s MacBook 16" with discrete GPUs! And this is without having any discrete GPU, but by sharing memory with the main CPU itself (both are on the same physical chip) and the additional performance benefits that such an architecture gives.
Addendum: After the Announcement (12th Nov)
While Apple has given it a dedicated name M1, instead of the A14X in the article above, the conclusion remains. The MacBook Air and MacBook Pro 13" are faster than all MacBook 16" devices!
This is mind-boggling!
