Because of the broad capabilities of VDI from a security, operational simplicity, and total cost of ownership (TCO) perspective, it is often an integral part of the IT infrastructure of healthcare organizations. VDI environments are used by a range of organizational users, which increasingly includes users with requirements for graphics-rich environments that require GPU resources. GPU environments are also widely deployed for AI activities such as machine learning and deep learning. These AI activities are seeing a significant increase in usage across multiple settings, with healthcare, which is an intensely data-driven field, being one of the primary settings.
This design guide presents the results of work carried out by the Dell Technologies VDI Solutions team to demonstrate the sharing of GPU resources in a healthcare setting between graphics-accelerated VDI, including Digital Imaging and Communications in Medicine (DICOM) X-Ray viewing, and healthcare-related AI. The tests included the prediction of diseases from public-domain chest X-Ray datasets released by the US National Institutes of Health (NIH).
This process of sharing GPU resources is sometimes called “compute cycle harvesting” or “VDI by day, compute by night.” In this guide we have chosen to describe it as GPU resource sharing for VDI and AI workloads in a healthcare environment. The architecture and the performance testing described in this document for a shared VDI and AI infrastructure include scenarios where the VDI and AI workloads are run as dual workloads simultaneously, and scenarios where VDI and AI are run as single workloads consecutively. It should be noted that the mechanism for switching between the workloads when run consecutively, whether done manually or with automation, is not within the scope of this paper.
By implementing a common platform for VDI and AI workloads to increase GPU utilization, organizations in the healthcare industry can improve IT efficiency and reduce TCO. Dell Technologies VDI Solutions offers a tested and validated VMware Horizon solution based on the hyperconverged Dell EMC VxRail platform, configured with NVIDIA RTX 8000 GPUs that can run mixed workloads including VDI and AI. By sharing the GPU cards between the workloads to meet various business requirements, around the clock, you get the maximum return on investment (ROI) from your GPU investment.
This design guide provides:
- A common infrastructure stack for VDI and AI—We designed a solution stack based on a density-optimized configuration with Dell EMC VxRail V570F, NVIDIA RTX 8000 GPUs, Dell EMC PowerSwitch networking, VMware vSAN, VMware vSphere, and VMware Horizon, and NVIDIA RTX Virtual Workstation (vWS) software to support the combination of VDI and AI activities.
- Optimal sizing for VDI—We performed sizing for the VDI environment using Login VSI testing with a customized Login VSI healthcare workload.
- A customized Login VSI workload for healthcare—We customized a Login VSI Knowledge Worker workload and validated it with viewing activities based on clinical images from the NIH database, using MicroDicom as the image viewing software.
- AI frameworks—We designed and implemented AI frameworks based on standard toolsets, including TensorFlow, in collaboration with the Dell Technologies AI Solutions team.
- AI workloads (learning and inferencing)—We performed AI deep-learning activities, including learning and inferencing, to validate the AI/deep-learning capabilities of the environment.
Dell Technologies has extensive experience in VDI solutions and a broad selection of compute, storage, and networking for VDI. Tested and validated VMware Horizon-based Dell Technologies VDI Solutions configured with NVIDIA virtual GPUs offer exceptional graphics performance and predictable cost. Dell Technologies provides a single-vendor support experience.
As the hyperconverged platform for the solution, Dell EMC VxRail delivers a highly differentiated, turnkey experience, with fully automated lifecycle management, to offer a fast and simple path to IT outcomes. VMware Cloud Foundation leverages native integration between VxRail Manager and VMware Software-Defined Data Center (SDDC) Manager to build a fully automated deployment of SDDC architecture. VxRail provides the capability to host up to three NVIDIA RTX 8000 GPUs per VxRail, providing a common platform to run both VDI graphic acceleration and AI compute workloads that share GPU cards.