YouTuber Dave Lee of Dave2D fame has demonstrated how Apple’s new Mac Studio geared up with an M3 Extremely chip can effectively run an enormous model of the DeepSeek R1 AI mannequin regionally, supplied that customers spec the machine with the utmost 512GB of reminiscence.
In line with Lee’s testing, the 671 billion parameter AI mannequin might be executed straight on Apple’s high-end workstation, however it requires substantial reminiscence sources, consuming 404GB of storage and requiring the guide allocation of 448GB of video RAM via Terminal instructions.
The M3 Extremely’s unified reminiscence structure is essential to this efficiency, permitting the system to deal with a 4-bit quantized model of DeepSeek R1 effectively. The quantization barely reduces accuracy, however it maintains all parameters and delivers roughly 17-18 tokens per second, which is ample for a lot of sensible purposes.
Maybe most impressively, the Mac Studio accomplishes this whereas consuming underneath 200 watts of energy. Comparable efficiency on conventional PC {hardware} would require a number of GPUs drawing roughly ten instances extra electrical energy.
The aptitude to run such superior AI fashions regionally provides privateness benefits for delicate purposes like healthcare information evaluation, the place sending info to cloud providers raises safety issues.
Nonetheless, this efficiency would not come low-cost – a Mac Studio configured with M3 Extremely and 512GB of RAM begins at round $10,000. Absolutely maxed out, an M3 Extremely Mac Studio with 16TB of SSD storage and an Apple M3 Extremely chip with 32-core CPU, 80-core GPU, and 32-core Neural Engine prices a cool $14,099. After all, for organizations requiring native AI processing of delicate information, the Mac Studio provides a comparatively power-efficient answer in comparison with various {hardware} configurations.
Apple says the M3 Extremely is the quickest Mac chip it has ever launched, due to its technique of fusing two M3 Max chips collectively utilizing the corporate’s “UltraFusion” know-how. This makes the chip’s specs double that of the M3 Max.
