Radeon RX 9070 XT · Qwen3 32B · Q4_K_M
Only with CPU offload. Qwen3 32B at Q4_K_M does not fit in the 16 GB on Radeon RX 9070 XT, and a 32 GB system RAM pool runs it slowly, roughly 4–6 tokens per second.
| Piece | Value | Note |
|---|---|---|
| Weights | 18.4 GiB | Qwen3-32B-Q4_K_M.gguf |
| KV cache | 2.00 GiB | 8,192 tokens |
| Engine overhead | 0.50 GiB | Flat 0.5 GiB allowance |
| 5% headroom | 1.05 GiB | On the sum above |
| Total | 22.0 GiB | Offload |
| Fast memory | 16.0 GiB | 16 GB VRAM |
| Speed | 4–6 tok/s | Dense bandwidth range |
ollama run qwen3:32b
Radeon RX 9070 XT · Qwen3 32B ·GGUF file · Methodology
FAQ
How was this fit calculated?
Weights 18.4 GiB, KV cache 2.00 GiB at 8,192 tokens, engine overhead 0.50 GiB, then 5% headroom, for 22.0 GiB. Fast memory available is 16.0 GiB.
What if I change the quant or the context?
This page is Q4_K_M at 8,192 tokens. A smaller quant or a shorter context lowers the total. The homepage calculator changes both without a new page.
Can I trust the tokens per second?
The range 4–6 tok/s is a bandwidth roofline, 55% to 80% efficient. This dense model reads the whole weight file each token. Measure your own box before you buy around it.
What do I type?
ollama run qwen3:32b
Is Radeon RX 9070 XT assumed to have extra RAM?
Yes. Discrete GPU pages assume 32 GB of system RAM. Offload uses whatever remains after 8 GB for Windows, and it is limited by the slower of DDR5-5600 and the PCIe link.