Orders & servers
GPU server bandwidth and egress fees
Outbound traffic is included in your monthly price, so there is no per-gigabyte bill. Here is what the allowance covers and what happens if you need more.
On this page
Here is how GPU server bandwidth and egress fees work at FullGPU: every server includes outbound bandwidth, with no egress fees within that allowance. Single-GPU plans include 20 TB per month and 8-GPU servers include 50 TB per month, under fair use. Inbound traffic is not counted.
Included bandwidth per server
| Server type | Outbound traffic included | Roughly equal to, averaged over 30 days |
|---|---|---|
| Single-GPU plans | 20 TB per month | about 62 Mbit/s, around the clock |
| 8-GPU servers | 50 TB per month | about 154 Mbit/s, around the clock |
The allowance comes with each server and is part of every price on our pricing page. Within it, you pay no egress fees at all. Hyperscale clouds typically bill egress separately; if that is what you are used to, see how FullGPU compares with AWS.
What counts toward the allowance
- Counted: outbound traffic from your server, such as responses from an inference endpoint, images or video sent to your users, and checkpoints or results you copy off the server.
- Not counted: inbound traffic, such as downloading datasets and model weights, pulling container images or uploading data to the server.
Some rough numbers help. A 70B-parameter model in BF16 is about 140 GB, and downloading it is inbound, so it does not count. Copying a checkpoint of that size off the server ten times is about 1.4 TB, well within 20 TB. Text responses from a language model are usually a few kilobytes each, while image and video workloads use far more.
Fair use and going over
The allowance is provided under fair use. If your outbound traffic consistently exceeds it, we contact you before any change, to agree on a suitable option with you. Nothing changes before that conversation.
If you expect heavy outbound traffic from day one, such as video delivery or large dataset exports, mention it in your request through the contact form so we can plan for it together.
Tracking your traffic
You can measure outbound traffic on the server itself. On Ubuntu, vnStat keeps daily and monthly totals per network interface once it is installed:
sudo apt install vnstat
vnstat -mFirewall and interface settings are covered in our networking and firewall guide. The bandwidth commitment itself is section 6 of our service level agreement.
Frequently asked questions
Is inbound traffic counted?
No. Only outbound traffic counts toward the allowance, so downloading datasets, model weights or container images to your server is not counted.
What happens if I use more than my bandwidth allowance?
If you consistently exceed it, we contact you before any change to agree on a suitable option. There are no egress fees within the included allowance.
Do 8-GPU servers include more bandwidth?
Yes. 8-GPU servers include 50 TB of outbound traffic per month, compared with 20 TB on single-GPU plans, both under fair use.
Still stuck?
Send us the details — server IP, what you ran and any error output. An engineer replies 24/7.