Only with CPU offload
Qwen3.8 27B at Q4_K_M needs 16.7 GB but only 5.3 GB is addressable, so about 73% of the layers would stream from system RAM at roughly 60 GB/s. Expect around 3.0 tokens per second — usable for batch work, painful for chat.
The current default local Qwen: dense 27B, text + image + video, 262K context. Only 16 of its 64 blocks keep a KV cache, so long context is cheap.
Hardware requirements What hardware do I need to run Qwen3.8 27B? →At Q4_K_M with an 8K context: a 24 GB GPU (RTX 4090) or a 32 GB Mac (M4). Minimum, recommended, high-quality and long-context tiers, with all 114 devices checked at every quantisation.
The VRAM budget
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| Q8_0 | 27.5 GB | 28.6 GB | — | ~1.5 | −0.1% ppl | 23.3 GB over |
| Q6_K | 21.2 GB | 22.3 GB | — | ~2.0 | −0.4% ppl | 17.0 GB over |
| Q5_K_M | 18.4 GB | 19.5 GB | — | ~2.4 | −0.8% ppl | 14.2 GB over |
| Q4_K_M | 15.6 GB | 16.7 GB | — | ~3.0 | −1.9% ppl | 11.4 GB over |
| Q3_K_M | 12.7 GB | 13.8 GB | — | ~3.9 | −5.4% ppl | 8.5 GB over |
| Q2_K | 10.8 GB | 11.9 GB | — | ~4.9 | −15% ppl | 6.6 GB over |
Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it. Only 16 of its 64 blocks keep a per-token KV cache; the rest are linear-attention, Mamba or convolution blocks with a fixed-size state.
How to run it
$ llama-server \
-hf Qwen/Qwen3.8-27B:Q4_K_M \
-c 8192 -ngl 17
The engine underneath most of the others. Every knob is exposed. More on llama.cpp.
Related models on a GeForce GTX 1660 Super
| Model | Total | Tok/s est. | |
|---|---|---|---|
| Qwen3.8-Flash-Next 180B-A6B 180B MoE · same family |
Q4_K_M · 102.0 GB | ~4.3 | CPU offload |
| Qwen3.8 2.4T-A95B 2446B MoE · same family |
Q4_K_M · 1376.7 GB | — | Won't fit |
| Qwen3.6 27B 27.8B · similar size |
Q4_K_M · 16.7 GB | ~3.0 | CPU offload |
| Gemma 3 27B 27.4B · similar size |
Q4_K_M · 17.0 GB | ~2.9 | CPU offload |
| Granite 4.1 30B 28.9B · similar size |
Q4_K_M · 18.9 GB | ~2.6 | CPU offload |
| Gemma 4 26B-A4B 26.5B MoE · similar size |
Q4_K_M · 16.0 GB | ~8.5 | CPU offload |
| Muse Glimmer 30B 29.8B · similar size |
Q4_K_M · 17.5 GB | ~2.8 | CPU offload |
Qwen3.8 27B on other hardware
| Device | Total | Tok/s est. | |
|---|---|---|---|
| GeForce RTX 4060 Laptop 7.0 GB usable · 256 GB/s |
16.7 GB | ~3.3 | CPU offload |
| GeForce RTX 3060 12 GB 10.6 GB usable · 360 GB/s |
16.7 GB | ~4.7 | CPU offload |
| M4 · 16 GB 10.7 GB usable · 120 GB/s |
16.7 GB | — | Won't fit |
| GeForce RTX 4090 22.4 GB usable · 1008 GB/s |
16.7 GB | 39 | Runs great |
| Radeon RX 7900 XTX 22.4 GB usable · 960 GB/s |
16.7 GB | 37 | Runs great |
| M4 · 32 GB 24.0 GB usable · 120 GB/s |
16.7 GB | 4.3 | Runs great |
| M3 Max · 64 GB 48.0 GB usable · 400 GB/s |
16.7 GB | 14 | Runs great |