Apple unveiled its new M1 chip at its Peak Performance event today: the M1 Ultra. This chip is, as Apple said, the last in the M1 family, which previously included M1, M1 Pro, and M1 Max.
Ultra is based on two M1 Max matrices that use an existing but apparently inactive connection on an existing M1 Max chip. This interconnection allows 2.5TB / s bandwidth between two chips. Apple calls this “Ultra Fusion”.
Since these are two M1 Max chips in combination, it is not surprising that the Ultra version has twice as many CPU and GPU cores. It has 20 CPU cores – 16 high-performance cores and 4 high-efficiency cores – and 64 GPU cores. The chip can support up to 128 GB of unified memory. There is also a 32-core Neural Engine for machine learning workloads. Together they provide 114 billion transistors.
All of this, Apple says, should make the Ultra 8 times faster than the M1, while the chip still outperforms even 10-core desktop chips in CPU performance per watt, both in terms of CPU and GPU. However, Apple did not say with which desktop chips the M1 Ultra compares.
Apple launched the original M1 chip in November 2020, marking the company’s first steps away from Intel chips in its portfolio.
Launched in 2020, the 8-core M1 debuted in the Mac mini, Macbook Air and MacBook Pro. Apple then followed suit with much more powerful M1 Pro and M1 Max chips, with up to 10-core CPUs, 32-core GPUs, and support for 64 GB of shared memory (unlike the original M1, which reached 16 GB). Those chips debuted in the 14- and 16-inch MacBook Pro.
The M1 Ultra will debut in the new Mac studio.
Chances are we’ll see how Apple unveils the M2 Pro and Max chips at a later event, assuming the company sticks to this naming scheme in the future.