The Apple M1 marks an important milestone in the company’s history. This is the beginning of an era where Apple no longer needs to rely on a third-party processor and graphics manufacturer to build its computers. The M1 may be one of the most interesting processors released in the last couple of years. The 5nm unified ARM-based SoC combines four powerful Firestorm CPU cores, four energy-efficient Icestorm CPU cores, and an eight-core graphics accelerator.

To a large extent, the design of the M1 graphics accelerator remains a mystery. So far we know that it has eight cores and offers 128 execution units (EU). Apple hasn’t disclosed clock speeds but hasn’t been shy about boasting performance numbers. The M1 can handle approximately 25,000 threads concurrently and provides a theoretical throughput of up to 2.6 teraflops. Apple is probably referring to the M1’s single-precision floating point (FP32) performance. For comparison, the theoretical performance of the M1 is roughly on par with the Radeon RX 560 (2.6 teraflops) and slightly inferior to the GeForce GTX 1650 (2.9 teraflops).

These numbers only show one part of the story. An anonymous user tested the Apple M1 in the GFXBench 5.0 benchmark using the Apple Metal API. There are also benchmark results for the GeForce GTX 1050 Ti through Metal, which allows you to evaluate the capabilities of Apple’s graphics versus Pascal.