apple's mac studio just beat nvidia's own local-AI chip — and the win doesn't add up

Tom's Hardware ran the numbers on three very different local-AI machines, and the Mac won on raw speed almost across the board — just not by the… Aliteq Ravi Malhotra · Hardware Editor

Apple's M4 Max (40-core GPU) has 546 GB/s of memory bandwidth — exactly double Nvidia GB10's and AMD Strix Halo's 273 GB/s each, per Tom's Hardware's own measurements.

The M4 Max produced higher tokens-per-second than both rivals at every context depth tested.

On Qwen 3.6-35B-A3B, the M4 Max's real-world throughput advantage was just 25% — far below the 2x its bandwidth spec would suggest.

On Gemma 4 12B, the gap was much wider: 1.8x faster than GB10 and 2.26x faster than Strix Halo.

The takeaway isn't 'buy a Mac' — it's that memory bandwidth alone is a bad predictor of local-AI performance, and model architecture matters more than a single spec.

Read the full story

Aliteq apple's mac studio just beat nvidia's own local-AI chip — and the win doesn't add up Read the full story on Aliteq