Apple Unveils M5 Pro and M5 Max Chips
Apple has announced its new M5 Pro and M5 Max chips, targeting professional creative workflows with a new "Fusion Architecture" and specialized "super cores" for parallel AI tasks. Early reports position the M5 Max as a local AI powerhouse, reportedly running a 70B parameter model at 20-30 tokens/sec on a laptop.
The new M5 Pro and M5 Max chips are built on an innovative "Fusion Architecture," which combines two 3nm dies into a single System on a Chip (SoC). This design integrates an up-to-18-core CPU, a scalable GPU with up to 40 cores, a Media Engine, Neural Engine, and Thunderbolt 5 capabilities. The new CPU architecture features six high-performance "super cores" and twelve "performance cores" optimized for multithreaded workloads, delivering up to a 30% boost in CPU performance over the M4 Pro. A significant leap in AI and machine learning capability comes from the redesigned GPU, which now includes a Neural Accelerator in every single core. This, combined with a faster 16-core Neural Engine, results in what Apple claims is over four times the peak GPU compute for AI tasks compared to the previous M4 generation and over six times that of the M1 generation. These architectural changes are specifically engineered to accelerate on-device AI, enabling faster processing for large language models (LLMs) and AI-powered creative tools. The M5 Max, in particular, solidifies its position for high-end creative and AI development by supporting up to 128GB of unified memory with a bandwidth reaching 614GB/s. This substantial memory bandwidth is crucial for handling large datasets and complex scenes, and it directly impacts the speed of token generation for large language models. This configuration makes running a 70B parameter model, which requires around 55GB for a Q6 quantized version, entirely feasible on a laptop. When compared to desktop GPUs like Nvidia's RTX series, the M5 Max offers a compelling proposition in power efficiency and portability for running large AI models. While an RTX 5090 may offer higher raw token generation speed for models that fit within its VRAM, the M5 Max's 128GB unified memory avoids the need for CPU offloading or complex multi-GPU setups for 70B models. The M5 Max accomplishes this with a power consumption of 60-90W under load, a fraction of the 600-800W required by a high-end desktop system. The new chips are featured in the latest 14-inch and 16-inch MacBook Pro models, which also gain Wi-Fi 7 and Bluetooth 6 support thanks to Apple's new N1 wireless chip. The storage is also faster, with read/write speeds up to 2x faster than the M4 generation, and base storage has been increased. Pre-orders for the new MacBook Pro models began on March 4, with availability starting March 11.