Qwen 3.8 27B VRAM Requirements

Qwen 3.8 27B needs about 16.9 GB of VRAM at Q4_K_M with an 8K context: 15.2 GB for weights, 0.6 GB for the KV cache and 1.1 GB of runtime overhead. At Q8_0 it needs 28.3 GB and at full FP16 precision 51.8 GB. The smallest GPU that runs it fully is the Radeon RX 7900 XT 20GB.

By Alibaba · Updated October 14, 2026 · Values from config.json on Hugging Face

16.9 GB VRAM

Qwen 3.8 27B at Q4_K_M with 8K context and the KV cache in FP16.

  • Weights 15.2 GB
  • KV cache 0.6 GB
  • Runtime overhead 1.1 GB

Which GPUs can run it

Qwen 3.8 27B VRAM by quantization

Total memory with an FP16 KV cache. "Smallest tier" is the smallest standard VRAM size with a 6% safety margin at 8K.

QuantizationBits/weightWeightsTotal at 8KTotal at 32KSmallest tier
Q4_K_M4.8515.2 GB16.9 GB18.4 GB20 GB
Q5_K_M5.6917.8 GB19.5 GB21.0 GB24 GB
Q6_K6.5920.6 GB22.3 GB23.8 GB24 GB
Q8_08.5026.6 GB28.3 GB29.8 GB32 GB
FP16 / BF1616.0050.1 GB51.8 GB53.3 GB64 GB
Q3_K_M3.9112.2 GB13.9 GB15.4 GB16 GB
Q2_K3.009.4 GB11.1 GB12.6 GB12 GB

VRAM by context length

At Q4_K_M. Hybrid attention: only 16 of 64 layers use full attention and keep a KV cache (4 KV heads × 256 dimensions). The other 48 use linear attention (Gated DeltaNet) with a fixed state of about 150 MB that does not grow with context.

ContextKV cache (FP16)KV cache (Q8_0)Total, FP16 cacheTotal, Q8_0 cache
4K tokens0.4 GB0.3 GB16.6 GB16.5 GB
8K tokens0.6 GB0.4 GB16.9 GB16.7 GB
32K tokens2.1 GB1.2 GB18.4 GB17.4 GB
128K tokens8.1 GB4.4 GB24.4 GB20.6 GB
256K tokens16.1 GB8.6 GB32.4 GB24.9 GB

GPUs that can run Qwen 3.8 27B

At Q4_K_M with 8K context. "Best quant" is the highest-quality format that fits fully; "max context" is at Q4_K_M.

GPUUsable memoryStatusBest quantMax contextSpeed
Radeon RX 7900 XT 20GB20.0 GBFitsQ4_K_M38K~26 tok/s
Apple M5 32GB21.3 GBFitsQ5_K_M58K~5.9 tok/s
NVIDIA L4 24GB24.0 GBFitsQ6_K98K~12 tok/s
Tesla P40 24GB24.0 GBFitsQ6_K98K~11 tok/s
Apple M4 Max 36GB24.0 GBFitsQ6_K98K~16 tok/s
Intel Arc Pro B60 24GB24.0 GBFitsQ6_K98K~14 tok/s
RTX 3090 24GB24.0 GBFitsQ6_K98K~36 tok/s
Radeon RX 7900 XTX 24GB24.0 GBFitsQ6_K98K~31 tok/s
RTX 3090 Ti 24GB24.0 GBFitsQ6_K98K~38 tok/s
RTX 4090 24GB24.0 GBFitsQ6_K98K~38 tok/s
Radeon AI PRO R9700 32GB32.0 GBFitsQ8_0219K~21 tok/s
RTX 5090 32GB32.0 GBFitsQ8_0219K~67 tok/s
Apple M4 Pro 48GB36.0 GBFitsQ8_0256K~11 tok/s
A100 40GB40.0 GBFitsQ8_0256K~58 tok/s
RTX 4060 Ti 16GB16.0 GB89% on GPUIQ4_XS–~8.1 tok/s
RTX 4070 Ti Super 16GB16.0 GB89% on GPUIQ4_XS–~13 tok/s
RTX 4080 16GB16.0 GB89% on GPUIQ4_XS–~13 tok/s
RTX 4080 Super 16GB16.0 GB89% on GPUIQ4_XS–~13 tok/s
RTX 5060 Ti 16GB16.0 GB89% on GPUIQ4_XS–~10 tok/s
RTX 5070 Ti 16GB16.0 GB89% on GPUIQ4_XS–~14 tok/s

About Qwen 3.8 27B

The latest dense Qwen at the 27B size, released in August 2026. It keeps the Qwen 3.5 hybrid layout (16 full-attention and 48 Gated DeltaNet layers) and a native 262K context, extendable toward 1M tokens with YaRN. Memory needs match Qwen 3.5 and 3.6 27B exactly; only quality changed.

SpecValue
Parameters26.9B
Layers64
Hidden size5120
Attention heads / KV heads24 / 4
Head dimension256
Vocabulary248,320
Max context (used here)256K tokens
Linear-attention layers48 of 64 (Gated DeltaNet, fixed 150 MB state)
KV cache per token (FP16)64 KB
Released2026-08
LicenseApache 2.0

Qwen 3.8 27B VRAM: common questions

How much context can Qwen 3.8 27B use on a 24GB GPU?

At Q4_K_M, about 96K tokens with an FP16 cache and about 180K with a Q8_0 cache. That is far more than a dense model without hybrid attention allows, because only 16 of 64 layers store keys and values.

How much VRAM does Qwen 3.8 27B need?

About 16.9 GB at Q4_K_M with an 8K context, 28.3 GB at Q8_0 and 51.8 GB at FP16. Longer contexts add 0.06 GB per 1,000 tokens with an FP16 cache.

What is the cheapest GPU that can run Qwen 3.8 27B?

The smallest listed device that fits it fully at Q4_K_M with 8K context is the Radeon RX 7900 XT 20GB (20.0 GB usable).

Related