Review of Mac Studio (M5 Ultra) with 256 GB of RAM: a dream machine to run local AI agents and a massive leap over M3 Ultra for prompt processing and generation

Local AI models now process prompts and generate responses up to 70% faster than before.

The M5 Ultra Mac Studio, equipped with 256 GB of RAM, is presented as a powerful machine for running local AI agents, significantly outperforming its predecessor, the M3 Ultra. Tested against the M3 Ultra with 512 GB RAM and a desktop PC with an RTX 5090, the M5 Ultra demonstrates exceptional performance for AI tasks. Its key improvements lie in a new Apple silicon architecture featuring UltraFusion, a next-gen 80-core GPU with enhanced Neural Accelerators providing up to 4.5x AI compute, and a 50% increase in memory bandwidth to 1.2 TB/s. These advancements translate to faster prompt processing and quicker response generation, making it viable for complex agentic workflows and large context windows. The reviewer found local AI agents on the M5 Ultra to be approximately 70% faster on average than on the M3 Ultra, enabling seamless interactions and efficient processing of up to 256K context windows.

The M5 Ultra Mac Studio's leap in AI performance, driven by its new architecture and doubled GPU cores, positions it as a viable alternative to expensive cloud-based AI services for complex, continuous tasks. This shift toward powerful local processing makes sophisticated AI capabilities accessible without recurring cloud costs, a significant factor for users handling sensitive data or undertaking large-scale projects. The improved speed and efficiency directly support more complex agentic loops, enabling faster iteration and larger context windows which were previously impractical on local hardware.

This development signifies a maturation of Apple's silicon for specialized AI workloads, challenging the dominance of dedicated GPUs in the consumer market for certain applications. By offering comparable or superior performance in specific areas like prompt prefill and generation speed, especially within its integrated ecosystem, the M5 Ultra prompts a re-evaluation of hardware investment for AI enthusiasts and professionals. The focus on memory bandwidth and unified memory architecture also highlights a design philosophy that prioritizes seamless data flow for AI models, potentially influencing future hardware development across the industry.