The cheapest cataloged machine that runs Qwen3 Coder 30B-A3B on the GPU is a AMD Radeon RX 7900 XTX · 24GB VRAM. Here are the most affordable rigs that fit it, smallest memory first.
| machine | best fit | verdict | speed |
|---|---|---|---|
| AMD Radeon RX 7900 XTX · 24GB VRAM | 30B-A3B (MoE) Q4_K_M | ✅ Runs (tight) | 130–220 tok/s |
| NVIDIA GeForce RTX 3090 · 24GB VRAM | 30B-A3B (MoE) Q4_K_M | ✅ Runs (tight) | 140–230 tok/s |
| NVIDIA GeForce RTX 4090 · 24GB VRAM | 30B-A3B (MoE) Q4_K_M | ✅ Runs (tight) | 150–250 tok/s |
| NVIDIA GeForce RTX 5090 · 32GB VRAM | 30B-A3B (MoE) Q6_K | ✅ Runs (tight) | 210–350 tok/s |
| Apple M3 Max · 36GB | 30B-A3B (MoE) Q4_K_M | ✅ Runs (tight) | 65–110 tok/s |
| Apple M3 Pro · 36GB | 30B-A3B (MoE) Q4_K_M | ✅ Runs (tight) | 30–50 tok/s |
| Apple M4 Max · 36GB | 30B-A3B (MoE) Q4_K_M | ✅ Runs (tight) | 65–110 tok/s |
| AMD Radeon Pro W7900 · 48GB VRAM | 30B-A3B (MoE) Q8_0 | ✅ Runs comfortably | 80–140 tok/s |
| Apple M3 Max · 48GB | 30B-A3B (MoE) Q8_0 | ✅ Runs (tight) | 40–70 tok/s |
| Apple M4 Max · 48GB | 30B-A3B (MoE) Q8_0 | ✅ Runs (tight) | 50–85 tok/s |
| Apple M4 Pro · 48GB | 30B-A3B (MoE) Q8_0 | ✅ Runs (tight) | 30–55 tok/s |
| Apple M1 Max · 64GB | 30B-A3B (MoE) Q8_0 | ✅ Runs comfortably | 45–70 tok/s |
Already have a machine? See Qwen3 Coder 30B-A3B across popular rigs → or check your exact rig →
Planning a whole workflow? Rigs for a coding agent · local chat · reasoning · local RAG setup.
Verdicts use the same GQA-aware fit math as the CLI — a AMD Radeon RX 7900 XTX verdict here matches what runlocal tells you on the machine.