NetApp and SK Telecom Test High-Performance AI in Virtualized Environments

NetApp and SK Telecom have completed a demonstration project showing that demanding AI workloads can run in virtual machines without the significant performance losses traditionally associated with virtualization.

The tests combined the NetApp AFX storage architecture, the Petasus AI Cloud platform and NVIDIA GPUDirect Storage. According to the published results, the virtualized environment achieved more than 99% of the performance delivered by bare-metal infrastructure.

During the test, a throughput of 32.7 GB/s was recorded in both the physical and virtualized environments. CPU utilization was reduced by approximately 40–50%, leaving more resources available for model training and inference workloads.

The results are important for organizations that require the flexibility of virtualization but do not want to compromise the performance needed for AI applications, financial services, telecommunications, manufacturing or advanced analytics.

BTS PRO specialists can analyze the relationship between servers, GPU accelerators, virtualization platforms, networking and storage to prevent performance bottlenecks in AI projects. This ensures that the infrastructure is designed as an integrated system rather than as a simple collection of separate components.