AMD Instinct MI350X
No editorial image yet — generic vendor mark shown. Credentials in spec table below.
The air-coolable sibling of the listed MI355X — same CDNA 4 silicon, 288GB HBM3E, ~8 TB/s, at a lower (~1,000W-class) power profile. The variant most non-hyperscale on-prem deployments would actually buy. (TDP figure pending confirmation against AMD's official datasheet.)
Sub-scores sum to 894 / 1000. Headline = 894 × 0.70 (Estimated-confidence discount) = 626. This is an algorithmic performance-tier score — distinct from, and often lower than, the editorial “Our verdict” below, which weighs value and real-world fit (especially for hardware we haven’t measured yet). How scoring works →
Extrapolated from 8000 GB/s bandwidth — 800.0 tok/s estimated. No measured benchmarks yet.
Plain-English: Runs 70B comfortably — snappy enough for a coding agent.
Verdicts extrapolated from catalog VRAM + bandwidth + ecosystem flags. Hover any chip for the rationale. Want measured numbers? Submit your own run with runlocalai-bench --submit.
What it is
The MI350X is AMD's air-cooled CDNA 4 datacenter GPU — same 288GB HBM3E and 8 TB/s as the liquid-cooled MI355X already in the catalog, but at a lower, air-coolable power envelope (1,000W class). The air-vs-liquid split matters operationally: the MI350X is the one that drops into standard air-cooled racks, which is what most non-hyperscale on-prem buyers actually run.
Relevance to local AI
AMD's 288GB-per-GPU on ROCm is the credible non-NVIDIA option for serving very large models on-prem, and the air-cooled MI350X is the realistic procurement target for an enterprise that can't deploy liquid cooling. The catch remains software: ROCm has closed much of the gap for inference (vLLM, SGLang support is solid) but still trails CUDA on the long tail. For an org standardizing on AMD Instinct for cost or supply reasons, this is the practical SKU.
Bottom line
The air-cooled, on-prem-friendly 288GB AMD datacenter GPU. Pairs with the MI355X row; choose AMD here for VRAM/cost on ROCm-supported inference, NVIDIA for maximum software compatibility. Confirm exact TDP against AMD's datasheet before relying on the power figure.
Overview
The air-coolable sibling of the listed MI355X — same CDNA 4 silicon, 288GB HBM3E, ~8 TB/s, at a lower (~1,000W-class) power profile. The variant most non-hyperscale on-prem deployments would actually buy. (TDP figure pending confirmation against AMD's official datasheet.)
Search-fallback link — editorial hasn't yet curated a retailer URL for this card.
Some links above are affiliate links. We may earn a commission at no extra cost to you. How we make money.
Specs
| VRAM | 288 GB |
| Power draw (peak) | 1000 W |
| Released | 2025 |
| Backends | ROCm |
Models that fit
Open-weight models small enough to run on AMD Instinct MI350X with usable context.
Hardware worth comparing
The closest alternatives by price, memory bandwidth, and form factor, plus a step up and down — so you can frame the buying decision against real options.
Frequently asked
What models can AMD Instinct MI350X run?
Does AMD Instinct MI350X support CUDA?
Where next?
Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify hardware specifications.