Apple releases M1 Ultra: 20-core CPU + 64-core GPU

At Apple's press conference early this morning, the M2 chip that Apple fans were looking forward to was not released, but Apple released a more powerful M1 Ultra, which is the fourth member of Apple's M1 family.

The M1 Ultra is not a completely new design. Instead, it utilizes the die-to-die connector that has been hidden in the M1 Max until now to integrate two M1 Max chips together. Apple calls it "Ultra Fusion". It has more than 10,000 signal points, an interconnection bandwidth of up to 2.5TB/s, and very low latency and power consumption.

With the same 5nm manufacturing process, the number of transistors in the M1 Ultra reaches 114 billion, which is a very terrifying number.

In terms of specifications, the M1 Ultra integrates 20 CPU cores, including 4 energy-efficiency cores and 16 performance cores; each performance core has 192KB instruction cache, 128KB data cache, and a shared 48MB of L2 cache; each energy-efficiency core has 128KB instruction cache, 64KB data cache, and a shared 8MB of L2 cache.

The number of GPU cores reaches 64, with a total of 8192 execution units, and the concurrent threads can support up to 196608. The theoretical computing power reaches 21TFlops, which is definitely a new generation of film editing artifact.

The number of neural network engines has also reached 32, capable of achieving a budget of 22 trillion times per second, which is eight times the performance of M1.

There is no upgrade in the media engine. It supports 2 video decoding engines, 4 video encoding engines, and 4 ProRes codec engines, supporting H.264, H.265, ProRes, ProRes RAW and other codecs.

In terms of performance, Apple said that the M1 Ultra has 65% lower power consumption than a 10-core desktop processor at the same performance; compared with a 16-core desktop processor, it has 100W lower power consumption at the same performance, and its performance is 90% stronger at the same power consumption.

In terms of GPU, Apple claims that the M1 Ultra GPU consumes 200W less power than high-end independent GPUs with the same performance. Compared with mainstream independent GPUs, it has stronger performance and only 1/3 of the power consumption.