FLUX.2-dev (32B DiT + 24B VLM) VRAM & GPU Requirement Calculator

Interactive hardware sizing for FLUX.2-dev (32B DiT + 24B VLM) (32B by Black Forest Labs). Estimated 8K context footprint: 20.7 GB (INT4) • 34.1 GB (FP8) • 66.2 GB (FP16).

GPUs

Memory Requirements 34.15 GB

Requires 2 GPUs (based on memory capacity)

32 GB

All model weights

0 GB

Conversation history cache

1.35 GB

Concurrent processing cache

0.8 GB

Temporary computation cache

ComfyUI Pipeline & Hardware Estimates
Est. Gen Time
12 s/img
Min System RAM
80 GB RAM
Resolution
1 MP
Offload Mode
Smart Offload
Quick Deploy CLI Command
python main.py --lowvram --fp8_e4m3fn-unet --fp8_e4m3fn-text-enc --preview-method auto
Flags reflect your selected precision, context length, and tensor parallel GPU count.Open FLUX.2-dev spec page →

FLUX.2-dev (32B DiT + 24B VLM) Architecture & Memory Profile

Total Parameters
32B
Architecture
Dense Transformer
Developer / Provider
Black Forest Labs
Recommended Precision
FP8

FLUX.2-dev (32B DiT + 24B VLM) ComfyUI VRAM Matrix by Precision & Resolution

Precomputed 3-stage ComfyUI Smart Offload + VAE Tiling VRAM peak (recommended host RAM: 80 GB).

Precision1024x1024 (1MP)1536x1536 (2.25MP)2048x2048 (4MP)LoRA Train (1MP)
FP1666.2 GB68.1 GB71.5 GB70.7 GB
FP834.1 GB36.1 GB39.5 GB38.7 GB
INT834.1 GB36.1 GB39.5 GB38.7 GB
INT420.7 GB22.7 GB26.1 GB25.2 GB

Recommended GPU Hardware for FLUX.2-dev (32B DiT + 24B VLM) (FP8)

Minimum GPU count for 1024x1024 generation with ComfyUI offload and estimated seconds/image.

GPU HardwareMin GPUsTotal VRAM PoolEst. Gen Time
NVIDIA RTX 4090 (24GB)2×48 GB12 s/img
NVIDIA RTX 5090 (32GB)2×64 GB8.7 s/img
Apple M4 Max (128GB) (128GB)1×128 GB29.2 s/img
NVIDIA A100 (80GB) (80GB)1×80 GB25.5 s/img
NVIDIA H100 (80GB)1×80 GB7 s/img
NVIDIA H200 (141GB)1×141 GB6.9 s/img
Huawei Ascend 910B (64GB)1×64 GB13.7 s/img

FLUX.2-dev (32B DiT + 24B VLM) Deployment FAQ

How much VRAM is needed to run FLUX.2-dev (32B DiT + 24B VLM)?
At 1024x1024 with ComfyUI Smart Offload and VAE Tiling enabled, FLUX.2-dev (32B DiT + 24B VLM) needs about 66.2 GB in BF16/FP16, 34.1 GB in FP8, and 20.7 GB in INT4/GGUF (paired with ~80 GB system RAM).
Can I run FLUX.2-dev (32B DiT + 24B VLM) on RTX 4090 or H100 GPUs?
Running FLUX.2-dev (32B DiT + 24B VLM) in FP8 precision requires at least 2x RTX 4090 (24GB) or 1x NVIDIA H100 (80GB) GPUs.
What precision should I choose for FLUX.2-dev (32B DiT + 24B VLM) in production?
For FLUX.2-dev (32B DiT + 24B VLM) (32B by Black Forest Labs), FP8 provides the best balance between inference speed, VRAM footprint, and output quality.