What is VRAM?
Short definition of VRAM and why GPU memory capacity matters for real workloads.
VRAM (Video RAM) is the dedicated memory on a GPU used to store data that the GPU is actively processing, such as model weights, activations, and intermediate tensors. If you are asking what is vram, think of it as the working memory for GPU tasks.
Why VRAM Matters
VRAM capacity determines how large a model or dataset you can fit on the GPU and how much work you can do without slow CPU offloading, so it is a practical limit for many workloads.
NVIDIA GPU VRAM & Memory Comparison
| Graphics Card | VRAM Capacity | Memory Type | Memory Bandwidth |
|---|---|---|---|
| RTX 3090 | 24 GB | GDDR6X | 936 GB/s |
| RTX 4090 | 24 GB | GDDR6X | 1,008 GB/s |
| RTX 5090 | 32 GB | GDDR7 | 1,792 GB/s |
| RTX A6000 | 48 GB | GDDR6 (ECC) | 768 GB/s |
| A100 | 80 GB | HBM2 / HBM2e | Up to 2,039 GB/s |
| H100 | 80 GB | HBM3 | Up to 3,352 GB/s |
How to Check VRAM
On NVIDIA GPUs, you can check VRAM capacity and usage with nvidia-smi, or inside Python with torch.cuda.get_device_properties(0).total_memory.
How to Increase VRAM
You cannot increase physical VRAM on a GPU, but you can reduce usage with smaller models, lower batch sizes, or offloading to CPU memory. You can also choose a GPU with more VRAM.