CUDA: NVIDIA’s software for GPU computing
CUDA is NVIDIA’s parallel computing platform and programming model for running general-purpose code on its GPUs. It is software, not a chip, and it runs only on NVIDIA hardware.
See the H100Checked 31 Oct 2026 · ZenoCloud Team
Is CUDA a GPU?
No. The GPU is hardware. CUDA is the software layer, a driver, toolkit and libraries, that lets programs use it.
Do I need to write CUDA?
Rarely. Frameworks such as PyTorch call CUDA for you. You pick a CUDA build of the framework and make sure the server’s driver is new enough.
| Layer | Example | Supplied by |
|---|---|---|
| Your code | Training or serving script | You |
| Framework | PyTorch | You |
| CUDA runtime and libraries | cuBLAS, cuDNN | Usually your framework build |
| Driver | NVIDIA driver | Included in our price |
| Hardware | The GPU card | ZenoCloud |
Our H200 guide states the card, the NVIDIA driver and a baseline host are in the monthly price. The framework layers are yours.
Does CUDA only work on NVIDIA?
Yes. AMD’s equivalent is ROCm, whose HIP interface is designed to make porting CUDA applications easier.
Which CUDA version do I need?
CUDA 13 can’t compile new code for Maxwell, Pascal or Volta. CUDA 12 software still runs, and NVIDIA’s R580 driver gets security fixes for them until October 2028.
Questions people ask
Is CUDA the same as GPU?
No. A GPU is the chip. CUDA is NVIDIA’s software platform for programming it.
Is CUDA a Python library?
No. It is a platform with a C++ programming model. Python reaches it through libraries such as PyTorch.
Is CUDA for NVIDIA only?
Yes. Other vendors have their own stacks, such as AMD’s ROCm, which can run ported CUDA-style code.
CUDA on a rented GPU
Our NVIDIA cards ship with the driver. Your framework brings the CUDA libraries. The H100 80GB starts at ₹1,80,000 a card a month, host included, before tax.
See the H100