Servers with a Graphics Card

On a regular server all computation is handled by the processor. Processors use a small number of powerful cores and finish sequential work quickly, but they are inefficient at running thousands of small calculations in parallel.

This is exactly where a graphics card helps: with thousands of CUDA cores it completes parallel workloads in a fraction of the time. Rendering, video encoding, image processing and AI inference all fall into this class. In addition, software that renders a 3D scene — a game client, for example — will not start at all on a server without a card.

A graphics card is a hardware option selected during checkout on Ryzen and Intel Core i9 dedicated servers.

Graphics cards are not supported on Xeon dedicated servers. If you need a card, choose a Ryzen or Intel Core i9 system.

No. Graphics cards are offered on dedicated servers only; they cannot be added to virtual server plans.

The reason is the hardware itself: the card is a physical component installed in the server’s motherboard and belongs to that machine directly. If your workload needs a GPU, you will need a Ryzen or Intel Core i9 dedicated server.

Yes. The card you choose at checkout can be replaced with a more powerful model later — you might start with a GTX 1660 and move to an RTX card as your needs grow.

Because the card is swapped physically, the server has to be restarted during the process, and your service is interrupted for that period. You can request an upgrade through support and schedule the work at a time that suits you.

VRAM capacity is the deciding factor: if the model does not fit in memory it will not run, and core count becomes irrelevant. As a rough measure, a 4-bit quantised model needs roughly half a byte per parameter.

On that basis 8 GB suits small models and image processing, 12–16 GB suits mid-sized models, and 24 GB or more is appropriate for large language models. In our catalogue the RTX 3090 and RTX 4090 offer 24 GB, and the RTX 5090 offers 32 GB.

For rendering, time scales largely with CUDA core count; as long as your scenes fit in memory, more cores means less time. The RTX 4080 (9,728 cores) and RTX 4090 (16,384 cores) are the strongest options in the catalogue for this.

For video encoding, core count is not the deciding factor — the work is done by a separate hardware encoder on the card. A mid-range RTX therefore performs much like a far more expensive model on stream encoding and format conversion. Buying more card than you need is wasted budget here.