VRAM
Video RAM — dedicated memory on a GPU
VRAM (Video RAM) is the dedicated memory on a GPU used to store data that the GPU is actively processing — model weights, activations, intermediate tensors, and more.
Example
import torch
# Allocate a tensor on GPU and check memory usage
t = torch.randn(1024, 1024, device="cuda")
print(f"{torch.cuda.memory_allocated() / 1e6:.1f} MB used")Why It Matters
| GPU | VRAM | Typical Use |
|---|---|---|
| RTX 4090 | 24 GB | Fine-tuning 7B models |
| A100 | 80 GB | Training large models |
| H100 | 80 GB | Large-scale training & inference |