How to upgrade GPU server plans and make changes

You can upgrade GPU server plans at FullGPU whenever you need more capacity: an upgrade takes effect right away, with a prorated charge for the rest of your current month. A move to a smaller plan applies at your next renewal. Every change is subject to stock, and when a change means moving to another machine, we can help you move your data between servers.

Upgrading to a bigger plan

Upgrades take effect right away. You pay a prorated charge for the price difference over the part of the month that remains, then the new monthly price from your next renewal.

For example, moving from a single H100 at $2,190 per month to a single H200 at $3,090 per month halfway through your month costs roughly half of the $900 difference, about $450. From your next renewal, you pay $3,090 per month.

Typical reasons to upgrade are a model that no longer fits in GPU memory, longer contexts that need a larger KV cache, or a move from one GPU to an 8-GPU server for multi-GPU training. The VRAM calculator helps you check what fits before you ask. If you need more servers rather than a bigger one, volume and term pricing is available on request, usually from three servers.

Moving to a smaller plan

A move to a smaller plan applies at your next renewal, subject to stock. You keep your current server until the end of the paid month, and the lower price starts with the next month. The remaining days of the current month are not refunded, so tell us before your renewal date. Our guide to how GPU server billing works covers renewals in detail.

How to request a change

Send your request through the contact form with:

• the IP address of the server you want to change;
• the configuration you want, as listed on our pricing page;
• when you want the change, especially for a move to a smaller plan at renewal.

All changes are subject to stock. If the configuration you want is not available, your current server stays as it is.

Moving data between servers

When a change involves a different machine, we can help you move your data between servers. A simple plan:

1. Copy datasets, checkpoints and configuration to the new server, for example with rsync over SSH.
2. Check that your workloads run correctly on the new server.
3. Tell us when you no longer need the old one. Disks are securely erased between customers, so make sure nothing you need is left on it.

rsync -aP /data/ root@NEW_SERVER_IP:/data/

Moving between NVIDIA and AMD? AMD servers use ROCm images instead of CUDA, so allow time to test your stack. If you run several servers, private networking between them is available on request. Billing rules for every plan are in our terms of service.