MiniMax-H3-Image (19.4B Repack / 33B Full) VRAM & GPU Requirement Calculator

Interactive hardware sizing for MiniMax-H3-Image (19.4B Repack / 33B Full) (19.4B by MiniMax). Estimated 8K context footprint: 13.7 GB (INT4) • 21.8 GB (FP8) • 41.2 GB (FP16).

GPUs

Memory Requirements 21.8 GB

Requires 1 GPUs (based on memory capacity)

19.4 GB

All model weights

0 GB

Conversation history cache

1.1 GB

Concurrent processing cache

1.3 GB

Temporary computation cache

ComfyUI Pipeline & Hardware Estimates
Est. Gen Time
15.2 s/img
Min System RAM
64 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 MiniMax-H3-Image spec page →

MiniMax-H3-Image (19.4B Repack / 33B Full) Architecture & Memory Profile

Total Parameters
19.4B
Architecture
Dense Transformer
Developer / Provider
MiniMax
Recommended Precision
FP8

MiniMax-H3-Image (19.4B Repack / 33B Full) ComfyUI VRAM Matrix by Precision & Resolution

Precomputed 3-stage ComfyUI Smart Offload + VAE Tiling VRAM peak (recommended host RAM: 64 GB).

Precision1024x1024 (1MP)1536x1536 (2.25MP)2048x2048 (4MP)LoRA Train (1MP)
FP1641.2 GB42.3 GB45.0 GB44.8 GB
FP821.8 GB22.9 GB25.6 GB25.4 GB
INT821.8 GB22.9 GB25.6 GB25.4 GB
INT413.7 GB14.7 GB17.5 GB17.2 GB

Recommended GPU Hardware for MiniMax-H3-Image (19.4B Repack / 33B Full) (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)1×24 GB15.2 s/img
NVIDIA RTX 5090 (32GB)1×32 GB11 s/img
Apple M4 Max (128GB) (128GB)1×128 GB18.4 s/img
NVIDIA A100 (80GB) (80GB)1×80 GB16.1 s/img
NVIDIA H100 (80GB)1×80 GB4.4 s/img
NVIDIA H200 (141GB)1×141 GB4.4 s/img
Huawei Ascend 910B (64GB)1×64 GB8.6 s/img

MiniMax-H3-Image (19.4B Repack / 33B Full) Deployment FAQ

How much VRAM is needed to run MiniMax-H3-Image (19.4B Repack / 33B Full)?
At 1024x1024 with ComfyUI Smart Offload and VAE Tiling enabled, MiniMax-H3-Image (19.4B Repack / 33B Full) needs about 41.2 GB in BF16/FP16, 21.8 GB in FP8, and 13.7 GB in INT4/GGUF (paired with ~64 GB system RAM).
Can I run MiniMax-H3-Image (19.4B Repack / 33B Full) on RTX 4090 or H100 GPUs?
Running MiniMax-H3-Image (19.4B Repack / 33B Full) in FP8 precision requires at least 1x RTX 4090 (24GB) or 1x NVIDIA H100 (80GB) GPUs.
What precision should I choose for MiniMax-H3-Image (19.4B Repack / 33B Full) in production?
For MiniMax-H3-Image (19.4B Repack / 33B Full) (19.4B by MiniMax), FP8 provides the best balance between inference speed, VRAM footprint, and output quality.