Best GPU for Qwen3-32B
The smallest memory budget in our catalog that runs Qwen3-32B cleanly at 8,192 tokens is 30 GiB net, with Q5_K_M (23.7 GiB–24.7 GiB total). Cards at that budget: GeForce RTX 5090 32GB. The largest file, BF16, is comfortable from 96 GiB net: Ryzen AI Max+ 395.
Which card, by net memory
| Net budget | Devices | Best file that runs | Total at 8,192 | Result | Same file at 32,768 |
|---|---|---|---|---|---|
| 22 GiB | GeForce RTX 3090 24GB, GeForce RTX 4090 24GB, Radeon RX 7900 XTX 24GB | Q4_K_M | 20.5 GiB–21.4 GiB | Tight | No |
| 30 GiB | GeForce RTX 5090 32GB | Q5_K_M | 23.7 GiB–24.7 GiB | Fits | Tight |
| 46 GiB | NVIDIA L40S 48GB | Q8_0 | 34.6 GiB–35.7 GiB | Fits | Tight |
| 54 GiB | Apple M2 Max 64GB unified (MacBook Pro / Mac Studio) | Q8_0 | 34.6 GiB–35.7 GiB | Comfortable | Fits |
| 78 GiB | NVIDIA A100 80GB SXM, NVIDIA H100 80GB SXM | BF16 | 63.6 GiB–65.6 GiB | Fits | Tight |
| 96 GiB | Ryzen AI Max+ 395 | BF16 | 63.6 GiB–65.6 GiB | Comfortable | Fits |
Rows start at the first budget where the smallest official file is at least tight and stop at the first where the largest file is comfortable. Devices in one row have the same net budget (capacity minus the catalog's default reserve), so they get the same verdict; choose between them on bandwidth, software support and price, which this page does not rank. llama.cpp, one sequence, fp16 KV cache, typical extras. Unified-memory devices use the reserve in their catalog record. "n/a" in the last column means the model's native window is shorter than 32,768 tokens.
Files this ranking uses
| File | Weights | Source repository |
|---|---|---|
| BF16 | 61.02 GiB (65,524,246,528 bytes) | Qwen/Qwen3-32B |
| Q8_0 | 32.43 GiB (34,817,718,912 bytes) | Qwen/Qwen3-32B-GGUF |
| Q6_K | 25.04 GiB (26,883,306,112 bytes) | Qwen/Qwen3-32B-GGUF |
| Q5_K_M | 21.62 GiB (23,214,831,232 bytes) | Qwen/Qwen3-32B-GGUF |
| Q4_K_M | 18.40 GiB (19,762,149,024 bytes) | Qwen/Qwen3-32B-GGUF |
Byte counts come from the repository file listings recorded on the model page. Memory estimates are formulas, not measurements on these cards. No speed figure is given: none was measured here.
Data sources
- Qwen3-32B model card
- config.json — revision 9216db5781bf21249d130ec9da846c4624c16137
- safetensors index — metadata.total_size=65524246528
- Method and reserves — catalog dataset 2026-10-02.1; hardware capacities from vendor spec pages listed on each device page