GPU: a chip for many sums at once
GPU stands for graphics processing unit: a chip built to run many calculations at the same time. It began as a graphics chip and is now widely used to train and run AI models.
See GPU hostingChecked 31 Oct 2026 · ZenoCloud Team
What does a GPU do?
It splits one big job into thousands of small, independent sums and runs them together. Drawing pixels is one such job. AI maths is another.
A graphics card is the board you plug in. The GPU is the chip on it. Data-centre GPUs such as the H100 are built for compute, not displays.
Why does AI use GPUs?
Models are mostly matrix multiplication, which splits into many parallel sums. GPUs suit that work. A CPU has fewer, faster cores built for general tasks.
GPU memory or RAM?
No. A GPU has its own memory, VRAM. A model runs best when it fits there. System RAM belongs to the CPU and is slower to reach.
Buy or rent a GPU?
Renting means no hardware to buy or host. Memory decides which models fit. Prices are per card per month in Mumbai, host included, before tax.
| Card | Memory | Per card a month |
|---|---|---|
| L4 | 24GB | ₹32,000 |
| L40S | 48GB | ₹55,000 |
| A100 | 80GB | ₹99,000 |
| H100 | 80GB | ₹1,80,000 |
| H200 | 141GB | ₹2,20,000 |
From our price book, dated 14 October 2026. B200 and RTX PRO 6000 are on the pricing page.
Questions people ask
What is the full form of GPU?
Graphics processing unit.
Does GPU mean RAM?
No. The GPU is the processor. Its memory is called VRAM, and it is separate from the system RAM your CPU uses.
Can I run an LLM without a GPU?
Small quantized models run on a CPU, slowly. For a 70B model you want GPU memory: its weights alone are tens of gigabytes.
Renting a GPU by the month
Cards start at ₹32,000 a card a month for the L4 24GB, before tax. Larger cards are in the table above. A written quote confirms the variant and the host.
See GPU hosting