- The 671 billion parameters of Deepseek R1 are executed without problems in the unified memory of the ultra M3
- Apple Mac Studio demonstrates that AI work loads do not require expensive GPU with energy
- M3 Ultra consumes less than 200W, much less than traditional multi-gpu settings
Apple’s Mac Studio with the M3 Ultra Chip has demonstrated a capacity that no other personal computer can match, executing the Deepseek R1 AI tool with 671 billion parameters completely in memory.
A test of the YouTube Dave2D reviewer showed despite using a quantified 4 -bit version of the model, retained its complete parameter count and served without problems.
The DEPEEEK R1 model, a strong 404 GB of storage and high bandwidth memory that is typically found in GPU VRM, is usually executed in multi -GPU configurations that distribute the processing in several high -end graphics cards.
A unique feat: Running Deepseek R1 in memory
However, the unified M3 Ultra system, instead of trusting external GPUs, uses its 512 GB of unified memory to store and process the AI model in one way that no other personal computer can.
Although macOS imposes a default vram limit, Dave Lee increased manually through the terminal to assign up to 448 GB for AI processing, eliminating memory bottlenecks and reducing the need for multiple components to expedite AI performance in a single system.
One of the most striking aspects of this test was the energy efficiency of the Ultra M3, since it consumed less than 200W while executing Depseek R1.
The ability to execute such a demanding model without a multi -GPU configuration defies the industry standard, which is based on Nvidia and AMD graphics cards of high -end, since the best work stations and server farms generally use GPU groups that consume large amounts of electricity.
Apple’s unified memory architecture allows significant energy saving by sharing the M3 Ultra memory group through CPU and GPU workloads, unlike conventional PC settings where VRM is separated from the system memory, maximizing bandwidth at the time that minimizes energy use.
Mac Studio of Apple, launched with the M3 Ultra Chip, presents a 32 -core CPU and a GPU of 80 cores, which makes it one of the best LLM work stations and one of the best video editing computers.
Through WCCFtech
