Skip to content

Best GPU for Qwen3.8-27B Locally

Real-time prices and hardware recommendations updated for August 2026.

27B
27B
dense

Qwen3.8-27B is a dense model: all 27B parameters activate on every token, so generation speed is bound by how fast your card can stream the full weights.

To run Qwen3.8-27B locally you need roughly 19.7 GB of VRAM at Q4_K_M quantization with a 128k token context. The best-value card that fits is the GeForce RTX 4090 (24 GB), which should generate around 16 tokens per second.

Adjust Context Length

Slide to popular sizes or type any value manually.

k tokens
8k
16k
32k
64k
128k
256k
Budget EntryQ3_K_M quant
11.6 GB
4.1 GB
Total VRAM:15.8 GB

Recommended Hardware

Radeon RX 9060 XT 16GB
160 tok/sprefill
6 tok/sgeneration
EUR 494.78·16GB VRAM
View Card
GeForce RTX 5060 Ti 16GB
256 tok/sprefill
10 tok/sgeneration
EUR 702.51·16GB VRAM
View Card

Q3 weights are ~12 GB. Because ~three-quarters of the layers use linear attention, the KV cache barely grows with context — even a 128k window adds only ~4 GB, so a 16 GB card holds the whole thing. This hybrid attention is why a 27B model runs long context on mainstream hardware.

Balanced Sweet SpotQ4_K_M quant
15.5 GB
4.1 GB
Total VRAM:19.7 GB

Recommended Hardware

GeForce RTX 4090
502 tok/sprefill
16 tok/sgeneration
EUR 2682.19·24GB VRAM
View Card
GeForce RTX 3090
369 tok/sprefill
11 tok/sgeneration
Out of Stock·24GB VRAM
View Card

Q4 weights (~15.5 GB) plus the small hybrid-attention KV cache stay under 20 GB at 128k context. A 24 GB card runs it at the balanced quant with room to push toward the native 262k window.

Near LosslessQ8_0 quant
31 GB
4.1 GB
Total VRAM:35.2 GB

Recommended Hardware

RTX PRO 5000 Blackwell
588 tok/sprefill
15 tok/sgeneration
EUR 8436.95·48GB VRAM
View Card
RTX A6000
218 tok/sprefill
5 tok/sgeneration
Out of Stock·48GB VRAM
View Card

Near-lossless Q8 weights are ~31 GB. A 48 GB workstation card holds the full precision model with headroom for the extended 262k-1M context this model supports.

Other models with the same VRAM requirement

Because Qwen3.8-27B's weights fit a 24 GB card, other models of a similar size run on the same GPU. Generation speed varies — mixture-of-experts models are faster, dense models slower — but any of these load in the same VRAM:

DiffusionGemma 26B-A4B25.2BNemotron-3-Nano-30B-A3B30BNorth-Mini-Code-1.030BGLM-4.7-Flash30.5BQwen3-30B-A3B30.5BQwen3-Coder-30B-A3B30.5B · Qwen3-30B-A3B

Optimizing Setup for Qwen3.8-27B

Quantization Recommendations

For daily coding and reasoning tasks, Q4_K_M (4-bit quantization) offers the best balance of quality and memory efficiency — it reduces memory requirements by over 70% with minimal quality loss compared to FP16. Q8 and higher presets preserve more fidelity at the cost of significantly higher VRAM usage, which may force layer offloading and hurt throughput.

Recommended Local Software

We recommend using Ollama as the primary runner for local inference due to its automated GPU model splitting and context cache optimizations. For advanced fine-tuning or quantization splits, llama.cpp with Flash Attention compiled natively provides the best granular control.

Running Qwen3.8-27B locally — FAQ

How much VRAM do I need to run Qwen3.8-27B?

At a 128k context with KV cache quantization on, Qwen3.8-27B needs about 19.7 GB of VRAM at Q4_K_M — the quantization most people should use. Dropping to Q3_K_M brings that down to roughly 15.8 GB at some quality cost, while Q8_0 needs about 35.2 GB for the best quality this model can give.

What size graphics card does Qwen3.8-27B fit on?

Qwen3.8-27B needs about 19.7 GB at Q4_K_M, so a 24 GB card is the smallest common size that holds it entirely in VRAM. Anything smaller has to offload layers to system RAM, which typically costs you most of your generation speed.

Which quantization should I use for Qwen3.8-27B?

Use Q4_K_M unless you have VRAM to spare. It needs about 19.7 GB and loses very little quality against full precision. Q8_0 needs about 35.2 GB for a quality gain most people cannot detect in everyday coding and chat. Spend spare VRAM on a longer context instead.

How does context length affect the VRAM Qwen3.8-27B needs?

Model weights are fixed, but the KV cache grows linearly with context. For Qwen3.8-27B at a 128k context the cache is about 4.1 GB; doubling to 256k takes it to roughly 8.3 GB. Turning KV cache quantization off doubles those figures again.

Can the same GPU run other models similar to Qwen3.8-27B?

Yes. Qwen3.8-27B needs about 19.7 GB at Q4_K_M, and any model whose weights fit the same card runs on it — including DiffusionGemma 26B-A4B (25.2B), Nemotron-3-Nano-30B-A3B (30B), North-Mini-Code-1.0 (30B), GLM-4.7-Flash (30.5B), Qwen3-30B-A3B (30.5B), Qwen3-Coder-30B-A3B (30.5B, based on Qwen3-30B-A3B). The weights fit the same GPU; generation speed varies (mixture-of-experts models are faster, dense models slower).

How token speeds are estimated

Two metrics are shown per GPU: Read tok/s (how fast the model ingests your prompt) and Decode tok/s (how fast it streams tokens back). They model fundamentally different bottlenecks.

📖

Read (Prefill)

The prompt is processed in one parallel pass. This is compute-bound: it saturates the GPU's tensor cores.

read tok/s ≈ TFLOPS × readFactor × 400 ÷ activeParams

Decode (Generation)

Each new token requires loading the entire model's active weights from VRAM. This is memory-bandwidth-bound: the GPU stalls waiting for data, not computing.

decode tok/s ≈ bandwidth × decodeFactor ÷ (weights + kv_cache)

Weights = (activeParams × bits ÷ 8) × 1.15 overhead. KV cache per step = activeParams × multiplier × contextK.

Architecture utilization factors

Architecture
Decode
Read
Blackwell, Xe2
0.45
0.55
Ada Lovelace, RDNA 4, Battlemage
0.38
0.48
Ampere, Turing, RDNA 3, Xe-HPG
0.28
0.38
Volta, RDNA 1/2
0.2
0.25
Pre-tensor-core (Pascal, Maxwell, Kepler, GCN, Alchemist)
0.12
0.15

Left: decode factor — Right: read factor

Data sources

TFLOPS and memory bandwidth are read from the GPU database. When missing, bandwidth falls back to a hardcoded dictionary.

Limitations

These are analytical estimates, not benchmark results. Use them as a relative comparison, not an absolute performance guarantee.