Skip to main content
To reduce the cost of using instances and provide a more flexible experience, GPUhub adopts a shared bandwidth scheme for instances in the same region, instead of charging separately for network bandwidth and traffic.
Each region has approximately 1–2 Gbps of shared bandwidth, with symmetric upload and download capacity. In typical usage, upload and download speeds of around 100–300 MB/s can be considered normal.
During peak hours, if other users on the same host are pulling large datasets at the same time, it may further slow down shell responsiveness and file transfers. This only affects SSH and uploads/downloads—not actual training speed.