Can the GeForce RTX 4070 Super run Qwen3.6 27B?
Only with CPU offload at Q4_K_M.
Qwen3.6 27B is 27.8B parameters. At Q4_K_M the weights are about 15.5 GiB; with the 1.5 GiB runtime reserve that is 17.0 GiB against the card's 12GB, about 5.0 GiB more than the card has.
All three quants
12GB card| Quant | Weights | With reserve | Verdict | Headroom |
|---|---|---|---|---|
| Q4_K_Mthe everyday quant | 15.5 GiB | 17.0 GiB | Only with CPU offload | short 5.0 GiB |
| Q8_0near-lossless | 27.4 GiB | 28.9 GiB | Does not fit | short 16.9 GiB |
| FP16full weights | 51.8 GiB | 53.3 GiB | Does not fit | short 41.3 GiB |
Verdict rules: fits comfortably when weights plus reserve sit within 85% of VRAM; tight when they fit with little left for context; CPU offload when they exceed VRAM by up to 2x (runs, slowly, with layers in system RAM); no beyond that. The KV cache per token depends on the model's architecture and is not modelled here.
The numbers for this pairing
At Q4_K_M the weights take about 15.5 GiB, so with the 1.5 GiB runtime reserve it needs 17.0 GB: 5.0 GB more than the card's 12 GB.
No version of it fits entirely in this card's memory, so it would lean on system RAM and run far slower.
The cheapest card on our list that runs it comfortably at Q4 is the Radeon RX 7900 XTX at about $850, against about $965 for this card.
At about $965, this card costs roughly $80 per GB of VRAM.
Running it at its full 220 W board power for 4 hours a day would use about 26 kWh a month, about $4.86 at the US average residential rate of 18.4 cents per kWh (2026-05); real use sits below that ceiling.
Prices are the lowest current listings our tracker saw on Oct 7, 2026, since averages include premium variants; launch prices are marked where we have no listing.
Not this card. These fit it comfortably at Q4.
- GeForce RTX 5090 32GB · headroom 15.0 GiB
- GeForce RTX 3090 (used) 24GB · headroom 7.0 GiB
- GeForce RTX 4090 (used) 24GB · headroom 7.0 GiB
- Radeon RX 7900 XTX 24GB · headroom 7.0 GiB
What the GeForce RTX 4070 Super does run well → · Qwen3.6 27B on every card →
About the model
Qwen3.6 27B by Alibaba Qwen: 262,144-token native context, text + vision, released 2026 under the Apache-2.0. Model card → · Hosted pricing on PicksByModel →