Best GPU for CogVideoX 5B: VRAM Requirements
CogVideoX 5B needs about 23.1 GB of GPU memory in BF16 and 5.8 GB at 4 bits. See which GPUs fit it for inference and fine-tuning.
CogVideoX 5B needs about 23.1 GB of VRAM in FP16, 11.6 GB in FP8 and 5.8 GB at 4 bits for inference. In FP16 it fits on one NVIDIA GeForce RTX 5090 32GB; a 4-bit build fits on one NVIDIA GeForce RTX 5080 16GB. Generating frames adds activation memory on top, which grows with resolution and clip length.
CogVideoX 5B is a 10.5B-parameter video generation model from Zhipu AI and Tsinghua University, released in August 2024 and published under the CogVideoX License. Its weights alone take 21 GB in BF16 and 5.3 GB at 4 bits per parameter; the tables below add runtime memory and fine-tuning, then list the GPUs that hold each setup.
Recommended GPUs for CogVideoX 5B
Best for inference
1 GPU × NVIDIA L4 24GB
FP8 or INT8 weights, 11.6 GB.
Best for fine-tuning
1 GPU × NVIDIA A100 PCIe 40GB
LoRA on BF16 weights, 27.3 GB.
Budget pick
1 GPU × NVIDIA GeForce RTX 5080 16GB
Smallest card class that holds a 4-bit build, 5.8 GB.
CogVideoX 5B VRAM requirements by precision
Model weights plus a 10 % margin for activations and runtime buffers, at one request.
| Precision | Weights | KV cache | Total for inference |
|---|---|---|---|
| BF16 / FP16 | 21 GB | Not applicable | 23.1 GB |
| FP8 | 10.5 GB | Not applicable | 11.6 GB |
| INT4 | 5.3 GB | Not applicable | 5.8 GB |
CogVideoX 5B VRAM by task
Fine-tuning rows assume a short sequence, batch 1 and gradient checkpointing; longer sequences add activation memory.
| Task | GPU memory |
|---|---|
| QLoRA (4-bit base) | 6.8 GB |
| LoRA (BF16 base) | 27.3 GB |
| Full fine-tuning (BF16, AdamW) | 184.8 GB |
GPUs that fit CogVideoX 5B
Number of cards whose combined memory holds each setup, assuming the model is split across them with tensor or pipeline parallelism. A dash means more than eight cards.
| GPU | Memory | BF16 inference (23.1 GB) | FP8 inference (11.6 GB) | INT4 inference (5.8 GB) | QLoRA (6.8 GB) | LoRA (27.3 GB) |
|---|---|---|---|---|---|---|
| NVIDIA GeForce RTX 5080 16GB | 16 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA GeForce RTX 4090 24GB | 24 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA L4 24GB | 24 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA GeForce RTX 5090 32GB | 32 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA Tesla V100 32GB | 32 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA A100 PCIe 40GB | 40 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA L40S 48GB | 48 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA Quadro RTX 8000 48GB | 48 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA RTX 6000 Ada 48GB | 48 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA RTX A6000 48GB | 48 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA A100 PCIe 80GB | 80 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA A100 SXM 80GB | 80 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA H100 PCIe 80GB | 80 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA H100 SXM 80GB | 80 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA RTX PRO 6000 Blackwell 96GB | 96 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA H200 141GB | 141 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA B200 180GB | 180 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA B300 288GB | 288 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA GB300 NVL | 288 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
What is CogVideoX 5B?
CogVideoX 5B is the larger open text-to-video model of the CogVideoX family from Zhipu AI and Tsinghua University, a diffusion transformer with 3D full attention that generates 6-second clips at 720 × 480 pixels and 8 frames per second.
- Developer
- Zhipu AI and Tsinghua University
- Released
- August 2024
- Parameters
- 10.5B
- License
- CogVideoX License
About 10.5 billion parameters for the pipeline: the 5.6B transformer, the T5-XXL text encoder (4.8B) and a 3D VAE. The model card lists about 5 GB of GPU memory with CPU offloading and VAE tiling in diffusers, which trade speed for memory.
Source: CogVideoX 5B model card (as of November 17, 2026).
Frequently asked questions
- How much VRAM does CogVideoX 5B need?
- For inference, CogVideoX 5B needs about 23.1 GB in BF16, 11.6 GB in FP8, 5.8 GB in INT4, including a 10 % runtime margin. Real usage depends on the inference engine and its settings.
- Can CogVideoX 5B run on a single GPU?
- Yes. In BF16 it needs about 23.1 GB, which fits one NVIDIA GeForce RTX 5090 32GB (32 GB).
- How much GPU memory does fine-tuning CogVideoX 5B take?
- QLoRA (4-bit base): about 6.8 GB; LoRA (BF16 base): about 27.3 GB; Full fine-tuning (BF16, AdamW): about 184.8 GB. Longer training sequences and larger batches add activation memory on top.
- What is the best GPU for CogVideoX 5B?
- Best for inference: 1 GPU × NVIDIA L4 24GB (FP8 or INT8 weights, 11.6 GB). Best for fine-tuning: 1 GPU × NVIDIA A100 PCIe 40GB (LoRA on BF16 weights, 27.3 GB). Budget pick: 1 GPU × NVIDIA GeForce RTX 5080 16GB (Smallest card class that holds a 4-bit build, 5.8 GB). Check live offers before booking, as availability changes.
- Can CogVideoX 5B run on an RTX 4090?
- Yes. In FP8 it needs about 11.6 GB, which fits one RTX 4090 (24 GB). FP16 needs about 23.1 GB, which leaves less than 10 % of the card free. Generating frames adds activation memory on top, which grows with resolution and clip length.
- Can CogVideoX 5B run on an RTX 5090?
- Yes. In FP16 it needs about 23.1 GB, which fits one RTX 5090 (32 GB). Generating frames adds activation memory on top, which grows with resolution and clip length.
Estimates use the formulas of the LLM VRAM calculator (decimal GB). Real usage depends on the inference engine, quantization format and memory fragmentation: test on the target GPU before a long rental.
Explore the live market
Every offer on the live market comes from a reviewed supplier and is sold by EverXCloud Inc.