FitMyLLM

Running LLMs on AMD Radeon Instinct MI350X: What You Can Run

The AMD Radeon Instinct MI350X packs 288GB of HBM3e with 8190 GB/s of memory bandwidth, making it one of the fastest cards you can run a local model on. Here is which models fit, how fast they run, and what quantisation gives the best balance of quality and speed.

Already have it? Find the best model for it. Still choosing? See what else $25,000 buys. Or the full specifications.

Loading the model tables…

What determines LLM performance on the AMD Radeon Instinct MI350X?

Two specs decide almost everything when you run a language model locally. Memory capacity decides which models fit at all, and memory bandwidth decides how fast they write, because every generated token reads the whole model once.

With 288GB of VRAM, the AMD Radeon Instinct MI350X holds models up to roughly 507B parameters at Q4_K_M, the most common quantisation, once the runtime has taken its own share. The context you ask for needs room on top of that, so a model near the limit runs with a short context.

At 8190 GB/s, speed falls as models grow: a model twice the size generates at roughly half the tokens per second. The estimates in the tables come from the bandwidth and each model's size, and real runs differ by 10–20% with context length and the inference backend.

Our recommendation

For general chat and everyday use on the AMD Radeon Instinct MI350X, start with a 14B–32B model at Q4_K_M. It is the best balance of answer quality and generation speed this card offers.

For coding, pick a model built for it (“coder” in the name). For reasoning-heavy work, reasoning models such as DeepSeek R1 distills or Qwen with thinking mode do best, at the cost of longer answers.

The AMD Radeon Instinct MI350X specs page has the full specifications and the MSRP ($25,000).