Cloud GPU VM FAQ
Last updated
Was this helpful?
Cloud Graphics Processing Units (GPU) Cloud Virtual Machines (VMs) provide access to powerful GPUs through a virtualized cloud environment, enabling massive parallel processing. This specialized hardware accelerates compute-intensive workloads, including training complex AI models, machine learning inference, and high-speed 3D rendering. Direct GPU access delivers the low latency that real-time applications require, while predictable, straightforward pricing makes continuous operation affordable.
While based on shared resources, the Cloud GPU VM can rival physical servers through a platform design that can redistribute available performance capacities among individual instances. At the same time, reduced operational complexity and highly optimized resource usage translate into lower operating costs.
The Cloud GPU connects directly to the VM via Peripheral Component Interconnect Express (PCIe) passthrough for near-native performance.
Cloud GPU VMs come complete with vCPUs and RAM. You can choose a template that suits your needs. For more information, see Cloud GPU VM workflow. You can also create and manage your Cloud GPU VM from the DCD.
By default, you can deploy only 1 Cloud GPU VM using the H200–S template. To deploy templates sized M, L, or XL, or to run multiple S instances, you must first request a resource limit increase through the IONOS Cloud Support.
Cloud GPU VM uses a PCIe passthrough architecture. It provides direct hardware access to ensure optimal performance, simplify deployment, and accelerate your production readiness.
Cloud GPU VM organizes the eight GPUs into two clusters of four (2 x 4 GPUs), connected through NVLink within each cluster.
Because the GPUs are split into two groups, address each cluster separately rather than as a single 8-GPU pool. To maximize performance, split your workloads into smaller chunks, process them across the two GPU groups, and merge the results afterward.
Last updated
Was this helpful?
Was this helpful?