Qwen3.8 27B DFlash2 EXL3 4.00bpw — Hardware Requirements & GPU Compatibility
Chat
Qwen3.8 27B DFlash2 EXL3 4.00bpw is a 27B-parameter open language model from r0b0tlab in the Qwen 3.8 family. It supports a context window of up to 262,144 tokens. At BF16 it needs about 54.35 GB of VRAM — see which GPUs and Macs can run it below.
Qwen3.8 27B DFlash2 EXL3 4.00bpw runs on a 64 GB+ Mac or a workstation GPU
Best local pick → Mac Pro M2 Ultra (64 GB) · BF16 · 54.7 GB · ~10 tok/s
Macs from 64 GB (BF16 · ~10 tok/s)
Apache 2.0
Not set up for it?
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Qwen3.8 27B DFlash2 EXL3 4.00bpw
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
vllm serve r0b0tlab/Qwen3.8-27B-DFlash2-EXL3-4.00bpw --max-model-len 16384vLLM repo with its own quantisation
Download: Weights on Hugging Face
Other ways
each opens step-by-stepRun in cloud
Fits on A100 80GB (25 GB headroom) · BF16
- Generation speed
- ~24 tok/s
- generation speed
- Cost per 1M output tokens
- $13.18
- per 1M output tokens
How Much VRAM Does Qwen3.8 27B DFlash2 EXL3 4.00bpw Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 54.4 GB | 55.1 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Does not fit: needs 54.7 GB |
Selected BF16
Brain floating point 1654.7GBat 16K context
- Weights
- 54.0 GB
- KV cache
- 0.4 GB
- Overhead
- 0.3 GB
Doesn't fit RTX 4090 (reference) · needs 30.7 GB more
Which Hardware Can Run Qwen3.8 27B DFlash2 EXL3 4.00bpw?
BF16 · 54.4 GBQwen3.8 27B DFlash2 EXL3 4.00bpw (BF16) requires 54.4 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 71+ GB is recommended. Using the full 262K context window can add up to 6.7 GB, bringing total usage to 61.0 GB. No consumer GPU has enough memory. 52 devices with unified memory can run Qwen3.8 27B DFlash2 EXL3 4.00bpw, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Pro M2 Ultra (64 GB).
64–80 GB
fits · 9.3 GB free · up to 256K ctx
3 Mac classes · 1 device family
M2 Ultra*64 GB~10Simulate output on Mac Pro M2 Ultra (64 GB)M5 Max64 GB~8Simulate output on Mac Studio M5 Max (64 GB)Orin~2Simulate output on NVIDIA Jetson AGX Orin 64GB
By VRAM tier · typical GPUs and Macs at BF16
Alternatives to Qwen3.8 27B DFlash2 EXL3 4.00bpw
Qwen 3.8 family · 12 models →Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Qwen3.6 27BAlibaba | 111.2 | 17.9 GB | ~38 | same size |
| Qwen3.5 27BAlibaba | 103.8 | 17.9 GB | ~38 | same size |
| Gemma 3 27B ITGoogle | 79.4 | 21.7 GB | ~36 | same size |
| Gemma 2 27B ITGoogle | 72.7 | 19.5 GB | ~37 | same size |
| Gemma 4 26B A4BGoogle | — | 16.9 GB | ~153 | same size |
| Ariarhymes-ai | — | 20.2 GB | ~168 | same size |
Specifications
- Publisher
- r0b0tlab
- Family
- Qwen 3.8
- Parameters
- 27B
- Architecture
- DFlash2DraftModel
- Context Length
- 262,144 tokens
- Vocabulary Size
- 248,320
- Release Date
- 2026-09-16
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Qwen3.8 27B DFlash2 EXL3 4.00bpw need?
Qwen3.8 27B DFlash2 EXL3 4.00bpw requires 54.4 GB of VRAM at BF16. Full 262K context adds up to 6.7 GB (61.0 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 27B × 16 bits ÷ 8 = 54 GB
KV Cache + Overhead ≈ 0.4 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
KV Cache + Overhead ≈ 7 GB (at full 262K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 5090 run Qwen3.8 27B DFlash2 EXL3 4.00bpw?
No — Qwen3.8 27B DFlash2 EXL3 4.00bpw requires at least 54.4 GB at BF16, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.
Can I run Qwen3.8 27B DFlash2 EXL3 4.00bpw on a Mac?
Yes — Mac Pro M2 Ultra (64 GB) and 41 other Macs can run Qwen3.8 27B DFlash2 EXL3 4.00bpw. Apple Silicon uses unified memory, so the model shares RAM with the system. At BF16 you need at least 54.4 GB of usable unified memory (RAM minus macOS overhead).
Can I run Qwen3.8 27B DFlash2 EXL3 4.00bpw locally?
Yes — Qwen3.8 27B DFlash2 EXL3 4.00bpw can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Qwen3.8 27B DFlash2 EXL3 4.00bpw?
At BF16, Qwen3.8 27B DFlash2 EXL3 4.00bpw can reach ~88 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: NVIDIA B200 → 8000 ÷ 54.4 × 0.65 = ~96 tok/s
Estimated speed at BF16 (54.4 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Qwen3.8 27B DFlash2 EXL3 4.00bpw?
At BF16, the download is about 54.00 GB.
Which GPUs can run Qwen3.8 27B DFlash2 EXL3 4.00bpw?
No single consumer GPU has enough VRAM to run Qwen3.8 27B DFlash2 EXL3 4.00bpw at BF16 (54.4 GB). Multi-GPU or professional hardware is required.
Which devices can run Qwen3.8 27B DFlash2 EXL3 4.00bpw?
52 devices with unified memory can run Qwen3.8 27B DFlash2 EXL3 4.00bpw at BF16 (54.4 GB), including ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB), GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.