video model · cogvideox · Windows
CV Can I run CogVideoX-2B on Nvidia GeForce RTX 4090 (24GB)?
Yes. CogVideoX-2B runs on Nvidia GeForce RTX 4090 (24GB) at fp16 + offload (~8 GB of ~23 GB usable).
Runs at fp16 + offload using ~8 GB of ~23 GB usable.
- Peak VRAM
- ~8 GB
- Usable on device
- ~23 GB
- Device memory
- 24 GB
- Quant
- fp16 + offload
How to run it
Use Diffusers or ComfyUI at fp16 + offload. The big text encoder is loaded to encode your prompt, then offloaded before generation, which is why peak VRAM stays near the backbone size rather than the sum of every file.
- Type
- video (DIT)
- Parameters
- 2B
- Peak VRAM
- ~8 GB at fp16 + offload
- Resolution
- 720×480
- License
- Apache-2.0
- Memory
- 24 GB vram
- Usable for weights
- ~23 GB
- Best runtime
- vLLM (Linux) / Ollama (CUDA)
You could also run
Run CogVideoX-2B on other hardware
FAQ
Can Nvidia GeForce RTX 4090 (24GB) run CogVideoX-2B?
Yes. CogVideoX-2B runs on Nvidia GeForce RTX 4090 (24GB) at fp16 + offload (~8 GB of ~23 GB usable).
How much VRAM does CogVideoX-2B need?
Nvidia GeForce RTX 4090 (24GB) has room to spare. At fp16 + offload the realistic peak is ~8 GB of VRAM, versus ~18 GB with every component kept resident (no offload). With aggressive CPU offload it drops to ~4 GB, much slower.
What do I use to run CogVideoX-2B locally?
CogVideoX-2B runs in Diffusers or ComfyUI. It loads as a video diffusion checkpoint plus its text encoder and VAE, not a single chat command.
Sources
VRAM figures are sourced peak-usage anchors at the noted quant, validated 2026-06-15. See methodology.