GeForce RTX 5080 · Qwen3.8 27B · UD-Q4_K_M
Only with CPU offload. Qwen3.8 27B at UD-Q4_K_M does not fit in the 16 GB on GeForce RTX 5080, and a 32 GB system RAM pool runs it slowly, roughly 19–27 tokens per second.
| Piece | Value | Note |
|---|---|---|
| Weights | 15.3 GiB | Qwen3.8-27B-UD-Q4_K_M.gguf |
| KV cache | 0.65 GiB | 8,192 tokens |
| Engine overhead | 0.50 GiB | Flat 0.5 GiB allowance |
| 5% headroom | 0.82 GiB | On the sum above |
| Total | 17.3 GiB | Offload |
| Fast memory | 16.0 GiB | 16 GB VRAM |
| Speed | 19–27 tok/s | Dense bandwidth range |
GeForce RTX 5080 · Qwen3.8 27B ·GGUF file · Methodology
FAQ
How was this fit calculated?
Weights 15.3 GiB, KV cache 0.65 GiB at 8,192 tokens, engine overhead 0.50 GiB, then 5% headroom, for 17.3 GiB. Fast memory available is 16.0 GiB.
What if I change the quant or the context?
This page is UD-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 19–27 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.
Is GeForce RTX 5080 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.