AMDDesktop GPUPrevious generation

AMD Instinct MI50 32GB for local LLMs

Written by Jakub Rusinowski · Last updated

With 32 GB of HBM2 at 1,024 GB/s, the AMD Instinct MI50 32GB runs 139 catalogued models at Q4_K_M with 8K context. The largest that fits is K2 Horizon MoVA 36B-A4B (~25 GB), and the top pick is K2 Horizon MoVA 36B-A4B.

32 GB of HBM2 at about 1 TB/s, 300 W, passively cooled. Cheap on the used market. AMD has ended official ROCm support for this card; Vulkan builds of llama.cpp and community ROCm builds still run it. No speed calibration for Vega: fit verdicts only.

Models that run on the AMD Instinct MI50 32GB

Q4_K_M, 8K context, 32 GB usable. Ranked by quality and speed.

ModelVRAMSpeed
K2 Horizon MoVA 36B-A4B
K2 Horizon
25 GB—
Qwen 3.6 35B-A3B
Qwen 3.6
22.2 GB—
Nex-N2.5 mini
Nex-N2.5
21.9 GB—
Nex-N2 mini
Nex-N2
21.9 GB—
Qwen 3.5 35B-A3B
Qwen 3.5
21.9 GB—
Jais 3 32B
Jais 3
24 GB—
Laguna XS 2.1 33B-A3B
Poolside Laguna XS 2.1
20.7 GB—
EXAONE 4.5 33B
EXAONE 4
22.3 GB—
Qwen 3 32B
Qwen 3
22.8 GB—
Aya Expanse 32B
Aya Expanse
21.6 GB—
Olmo 3.1 32B Instruct
Olmo 3
21.7 GB—
Olmo 3.1 32B Think
Olmo 3
21.7 GB—
Showing 12 of 139

Buy it or rent the same memory

Buy the card, or rent a GPU with the same memory by the hour to try models first.

Buy this hardware NVIDIA GeForce RTX 5090 32GB — 32 GB VRAM · 575 W board powerDeploy in the cloud now NVIDIA A40 on RunPod — from $0.44/hr · rate checked 2026-08

or compare on Vast.ai

As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.

Speed vs other GPUs

Llama 3.1 8B, Q4_K_M. Estimated ranges. How this is calculated

Specifications

Specs last updated 2026-10-07.

Memory
32 GB HBM2
Memory bandwidth
1,024 GB/s
Memory bus
4096-bit
Architecture
GCN 5 Vega 20
Series
Instinct
Board power
300 W
Release year
2018
Compute backends
ROCM, VULKAN
Usable for models
32 GB
Best forcheap used 32 GBVulkan

Similar GPUs

Frequently asked questions

Can the AMD Instinct MI50 32GB run local LLMs?

Yes. With 32 GB (32 GB usable by a model) it runs 139 of the catalogued models at Q4_K_M with 8K context; the largest is K2 Horizon MoVA 36B-A4B, needing about 25 GB.

How fast is the AMD Instinct MI50 32GB for AI inference?

This site has no speed calibration for the GCN 5 Vega 20 architecture, so it publishes fit verdicts for this card but no tokens-per-second estimate. Llama 3.3 70B does not fit: it needs about 44 GB against 32 GB usable.

What LLMs can I run on 32 GB?

Among the best that fit: K2 Horizon MoVA 36B-A4B, Qwen 3.6 35B-A3B, Nex-N2.5 mini, Nex-N2 mini, Qwen 3.5 35B-A3B.