GeForce RTX 5090 · Qwen3.6 35B-A3B · UD-Q4_K_M
Yes. A GeForce RTX 5090 runs Qwen3.6 35B-A3B at UD-Q4_K_M with room left at 8,192 tokens, roughly 414–668 tokens per second.
| Piece | Value | Note |
|---|---|---|
| Weights | 20.6 GiB | Qwen3.6-35B-A3B-UD-Q4_K_M.gguf |
| KV cache | 0.22 GiB | 8,192 tokens |
| Engine overhead | 0.50 GiB | Flat 0.5 GiB allowance |
| 5% headroom | 1.07 GiB | On the sum above |
| Total | 22.4 GiB | Fits |
| Fast memory | 32.0 GiB | 32 GB VRAM |
| Speed | 414–668 tok/s | Active-expert bandwidth range |
GeForce RTX 5090 · Qwen3.6 35B-A3B ·GGUF file · Methodology
FAQ
How was this fit calculated?
Weights 20.6 GiB, KV cache 0.22 GiB at 8,192 tokens, engine overhead 0.50 GiB, then 5% headroom, for 22.4 GiB. Fast memory available is 32.0 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 414–668 tok/s is a bandwidth roofline, 55% to 80% efficient. This MoE model only reads the active experts, using an architecture-derived fraction of the file. Measure your own box before you buy around it.
Is GeForce RTX 5090 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.