M5
A desktop processor chip introduced by Apple, the first to adopt a 2 nm fabrication process. It integrates a 12-core CPU, a 12-core GPU, and a neural network engine, and was first deployed in the Mac mini.
Core Facts
The Neural Accelerator is a dedicated hardware unit first introduced by Apple in the M5 series chips, designed specifically to accelerate neural network inference tasks, operating independently of the CPU and GPU, and continued and integrated into the next-generation GPU in the M6.
The M6's CPU has 12 cores (4 performance cores + 6 efficiency cores), 2 more cores than the M5, with multi-core performance 1.2x faster than the M5.
Timeline (last 90 days)
The Neural Accelerator is a dedicated hardware unit first introduced by Apple in the M5 series chips, designed specifically to accelerate neural network inference tasks, operating independently of the CPU and GPU, and continued and integrated into the next-generation GPU in the M6.
The speculative decoding gains observed on the M5 are only between 4% and 40%, and the M3 Ultra shows almost no improvement.
The M5's INT8 compute is twice that of BF16, and a project on GitHub using the M5's INT8 compute kernels has reported a 1.6× improvement.
Apple officially stated in LM Studio testing that after adding Neural Accelerators to the M5 Ultra's GPU, large model prompt processing speed can reach up to 4 times that of the M3 Ultra.
The M6's CPU has 12 cores (4 performance cores + 6 efficiency cores), 2 more cores than the M5, with multi-core performance 1.2x faster than the M5.
MicroGPT achieved a throughput of tens of millions of tokens per second (10M tps) on an Apple M5 chip.
All Facts (7)
The Neural Accelerator is a dedicated hardware unit first introduced by Apple in the M5 series chips, designed specifically to accelerate neural network inference tasks, operating independently of the CPU and GPU, and continued and integrated into the next-generation GPU in the M6.
65%VerifiedThe M6's CPU has 12 cores (4 performance cores + 6 efficiency cores), 2 more cores than the M5, with multi-core performance 1.2x faster than the M5.
65%UnverifiedThe speculative decoding gains observed on the M5 are only between 4% and 40%, and the M3 Ultra shows almost no improvement.
50%UnverifiedThe M5's INT8 compute is twice that of BF16, and a project on GitHub using the M5's INT8 compute kernels has reported a 1.6× improvement.
50%UnverifiedApple officially stated in LM Studio testing that after adding Neural Accelerators to the M5 Ultra's GPU, large model prompt processing speed can reach up to 4 times that of the M3 Ultra.
50%UnverifiedMicroGPT achieved a throughput of tens of millions of tokens per second (10M tps) on an Apple M5 chip.
50%UnverifiedM4 Max的内存带宽可达400GB/s以上
50%