CogView4-6B (6B DiT + 9B GLM-4) VRAM & GPU Requirement Calculator

Interactive hardware sizing for CogView4-6B (6B DiT + 9B GLM-4) (6B by Zhipu AI). Estimated 8K context footprint: 5.2 GB (INT4) • 7.8 GB (FP8) • 13.8 GB (FP16).

GPUs

Memory Requirements 7.75 GB

Requires 1 GPUs (based on memory capacity)

6 GB

All model weights

0 GB

Conversation history cache

1.2 GB

Concurrent processing cache

0.55 GB

Temporary computation cache

ComfyUI Pipeline & Hardware Estimates
Est. Gen Time
11.3 s/img
Min System RAM
32 GB RAM
Resolution
1 MP
Offload Mode
Smart Offload
Quick Deploy CLI Command
python main.py --normalvram --fp8_e4m3fn-unet --fp8_e4m3fn-text-enc --preview-method auto
Flags reflect your selected precision, context length, and tensor parallel GPU count.Open CogView4-6B spec page →

CogView4-6B (6B DiT + 9B GLM-4) Architecture & Memory Profile

Total Parameters
6B
Architecture
Dense Transformer
Developer / Provider
Zhipu AI
Recommended Precision
FP8

CogView4-6B (6B DiT + 9B GLM-4) ComfyUI VRAM Matrix by Precision & Resolution

Precomputed 3-stage ComfyUI Smart Offload + VAE Tiling VRAM peak (recommended host RAM: 32 GB).

Precision1024x1024 (1MP)1536x1536 (2.25MP)2048x2048 (4MP)LoRA Train (1MP)
BF1613.8 GB15.5 GB18.5 GB16.6 GB
FP87.8 GB9.5 GB12.5 GB10.6 GB
INT87.8 GB9.5 GB12.5 GB10.6 GB
INT45.2 GB7.0 GB10.0 GB8.1 GB

Recommended GPU Hardware for CogView4-6B (6B DiT + 9B GLM-4) (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 GB11.3 s/img
NVIDIA RTX 5090 (32GB)1×32 GB8.2 s/img
Apple M4 Max (128GB) (128GB)1×128 GB13.7 s/img
NVIDIA A100 (80GB) (80GB)1×80 GB12 s/img
NVIDIA H100 (80GB)1×80 GB3.3 s/img
NVIDIA H200 (141GB)1×141 GB3.2 s/img
Huawei Ascend 910B (64GB)1×64 GB6.4 s/img

CogView4-6B (6B DiT + 9B GLM-4) Deployment FAQ

How much VRAM is needed to run CogView4-6B (6B DiT + 9B GLM-4)?
At 1024x1024 with ComfyUI Smart Offload and VAE Tiling enabled, CogView4-6B (6B DiT + 9B GLM-4) needs about 13.8 GB in BF16/FP16, 7.8 GB in FP8, and 5.2 GB in INT4/GGUF (paired with ~32 GB system RAM).
Can I run CogView4-6B (6B DiT + 9B GLM-4) on RTX 4090 or H100 GPUs?
Running CogView4-6B (6B DiT + 9B GLM-4) in FP8 precision requires at least 1x RTX 4090 (24GB) or 1x NVIDIA H100 (80GB) GPUs.
What precision should I choose for CogView4-6B (6B DiT + 9B GLM-4) in production?
For CogView4-6B (6B DiT + 9B GLM-4) (6B by Zhipu AI), FP8 provides the best balance between inference speed, VRAM footprint, and output quality.