Can a NVIDIA RTX 4090 (24GB) run Qwen3.8 27B?

Yes — Qwen3.8 27B runs at Q4_K_M (17.4 GB) at about 38–57 tok/s on a NVIDIA RTX 4090 (24GB).

How much memory does Qwen3.8 27B need?

Qwen3.8 27B needs about 11.9 GB at 4K context at its smallest quantization. A NVIDIA RTX 4090 (24GB) has 24.0 GB.

Yes — Qwen3.8 27B runs on a NVIDIA RTX 4090 (24GB) at Q4_K_M

Qwen3.8 27B: pack choices on an RTX 4090

Content reviewed 11 Oct 2026

Current recommended pack · Q4_K_M

Weights: 17.4 GB
KV cache: 0.3 GB
Runtime reserve: 0.8 GB
Total requirement: 18.5 GB
Listed hardware memory: 24.0 GB
Usable model budget: 21.6 GB

Estimated at 4,096 tokens, llama.cpp, FP16 K/V cache, one stream and one GPU. Total = weights + cache + runtime reserve. The usable budget assumes 90% of device memory. Budget and reserve are site assumptions, not measured free memory. Actual peak allocation and runtime support can differ.

Budget for the whole allocation

On this single-card setup, the useful comparison is weights plus cache and runtime overhead against the GPU budget shown here. A pack that is smaller than nominal VRAM can still exceed that budget. Compare the listed quantizations before increasing context, and leave room for prompt processing and other GPU users rather than relying on file size.

Match CUDA, format and model support

llama.cpp provides an NVIDIA CUDA backend, but that does not certify every new architecture or conversion. Check support for this exact GGUF and the model chat template. Qwen’s vLLM recipe describes its own dependencies and serving configuration; it does not validate this RTX 4090 estimate. Weight quantization and K/V cache precision are independent choices.

llama.cpp backends ↗llama.cpp runtime controls ↗vLLM Qwen3.8 recipe ↗

Expand a working text setup carefully

Qwen specifies 262,144 native tokens and a configured extension to 1,000,000. Test prompt length before expanding the short-context setup. CPU offload needs sufficient host RAM and changes performance. A smaller model can leave more room for context.

Qwen3.8 model card ↗llama.cpp runtime controls ↗

This is a text-inference estimate. It excludes the vision projector and image/video workload; use the Vision calculator for that separate memory scope.

Smaller models to compare

Same-context catalog estimates; speed is unmeasured and task quality can differ.

See the calculation methodology, or contribute a measured run with setup details.

Yes — it runs great.

Runs great

Best fit: Qwen3.8 27B at Q4_K_M · 17.4 GB weights · 38–57 tok/s on a NVIDIA RTX 4090 (24GB).

Arena Text Overall · style-controlled

1,438Q4_K_M untested

qwen3.8-27b · published configuration reference · Q4_K_M not tested

Arena Text Overall (style-controlled), exact published entry qwen3.8-27b, 2026-10-08: reference rating 1,438.1. Published confidence bounds 1,432.8–1,443.4. 20,711 votes. Published qwen3.8-27b configuration reference. Reasoning mode is not disclosed by the Arena dataset. Tested precision, runtime and sampling settings are not disclosed in this dataset. Selected Q4_K_M has not been measured by this Arena entry. No quantization quality adjustment is applied.

Snapshot published 2026-10-08 · snapshot imported 2026-10-11 UTC · validated last-good data · manual refresh

Arena leaderboard dataset · Arena / lmarena-ai, Arena Leaderboard Dataset (CC BY 4.0). Selected Text Overall style-controlled rows; field names adapted, ratings unchanged. · dataset CC BY 4.0 · revision e16f90232986

Total memory at 4K context: 18.5 GB; Weights + KV cache + runtime overhead
Estimated generation: 38–57 tok/s
Estimated first token: ~3.9s

Every quantization on this machine

On small screens, each quant lists every metric below its name. On wider screens, scroll sideways for total memory, speed and fit. Keyboard: focus the table, then use the arrow keys. Quant names stay visible.

Qwen3.8 27B quantizations on NVIDIA RTX 4090 (24GB), estimated at 4K context. Weights and total memory are shown separately.
Quant	Weights	Total memory	Speed	Fit
Q2_K	10.8 GB	11.9 GB	57–85 tok/s	Runs great
Q3_K_M	13.4 GB	14.5 GB	47–71 tok/s	Runs great
Q4_K_M	17.4 GB	18.5 GB	38–57 tok/s	Runs great
Q5_K_M	20.9 GB	22.0 GB	— not resident	Won't fit
Q6_K	23.9 GB	24.9 GB	— not resident	Won't fit
Q8_0	29.1 GB	30.2 GB	— not resident	Won't fit
FP16	54.7 GB	55.7 GB	— not resident	Won't fit

Open the full calculator →Everything a NVIDIA RTX 4090 (24GB) runs Qwen3.8 27B on other machines What to buy to run it →

Ran Qwen3.8 27B on a NVIDIA RTX 4090 (24GB)? Add your real numbers →