AMD Radeon Instinct MI50 32GB
Compare AMD Radeon Instinct MI50 32GB against another machine in the Arena
Reference: AMD Instinct on Wikipedia
- Class
- GPU
- Memory
- 32 GB HBM2
- Runs (Q4_K_M)
- —
- Bandwidth
- 1024 GB/s
- TDP
- 300 W
- Released
- 2018-11-01
- Price (US)
- $199 used · $433 new as of 2026-08
- Price (Canada)
- CA$600 used sourced, as of 2026-08
Sourcing and disambiguation notes
The prices above ($433 new / $199 used, gpudojo.com) are for the 16GB MI50, NOT the 32GB variant this entry documents and that the benchmarks below were run on. No verified current listing for the 32GB card was found, so treat the USD figures as a floor: the card actually benchmarked here almost certainly costs more. The Canadian figure IS for the 32GB card; other eBay.ca listings, mostly shipping from China, range roughly CA$600-1,424 by condition. ROCm support for gfx906 is being phased out upstream, and recent llama.cpp testing finds Vulkan often faster and more reliable for short-context dense models, with ROCm ahead only at long context or on MoE models. Needs a blower shroud for airflow outside a server chassis, like the Nvidia Teslas.
Specification sources
- https://cputronic.com/gpu/amd-radeon-instinct-mi50 — press, 2026-08-22
- https://katalog.atcomp.cz/katalog/2078603/radeon-instinct-mi50-datasheet.pdf — vendor, 2018-11-01
Measurements
Decode is token generation — the speed you feel while an answer streams. Prefill is prompt processing — the wait before it starts. Why bandwidth predicts decode speed.
No records yet for this hardware. Know of a published benchmark on it? Submit the link.
What verified, single-source and estimated mean, and the same rows with every filter and sort in the benchmarks explorer.
Speed over time
What it can run
Fit is arithmetic, not a measurement — how it is computed. Against 32 GB; models that fit are listed largest first.
No modeled quant fits in 32 GB.