Guides to running large language models on unified-memory machines: Mac Studio with MLX, NVIDIA DGX Spark with CUDA, memory sizing and the economics of renting.
5 min read
Two desktop machines with 128 GB of unified memory, built on very different bets. How the Mac Studio M5 Max and the NVIDIA DGX Spark compare on bandwidth, compute, software and tooling, and when to pick which.
5 min read
What's inside NVIDIA's desktop AI computer: the GB10 Grace Blackwell superchip, 128 GB of LPDDR5x at 273 GB/s, up to 1 PFLOP of FP4, the CUDA software stack, and which models fit.
5 min read
A 128 GB Mac Studio is a big purchase. Here's the break-even against renting one at €20 a day, the costs people forget, and the jobs where renting simply makes more sense.
5 min read
How to work out whether a model fits in 96, 128 or 256 GB of unified memory — weights, KV cache and headroom — and what the M5 Ultra's 1.2 TB/s buys you over the M5 Max's 614 GB/s.
5 min read
From a fresh Mac Studio to a working model server: installing mlx-lm, choosing a model and quantisation, generating, chatting, serving an OpenAI-compatible API, and what speed to expect.
6 min read
Large language models are limited by memory before anything else. Here's how the Mac Studio's shared memory pool changes what fits, and why bandwidth then decides how fast it runs.