In his GTC keynote, NVIDIA CEO Jensen Huang highlighted the progress of the NVIDIA BlueField DPU to offload, accelerate, and isolate data center infrastructure. Nutanix is proud to collaborate with NVIDIA to drive innovation and unite accelerated computing with hyperconvergence in a unified platform. This is a step that builds upon our years of history enabling Nutanix customers to run high-end applications for various uses including visual computing, virtual desktop infrastructure (VDI), AI/ML, and HPC using the latest GPUs — most recently the NVIDIA Ampere architecture generation — and high-performance NVIDIA networking.
NVIDIA Accelerated Computing
Our relationship with NVIDIA dates back nearly to our inception as a company. The first workload Nutanix tackled was quite naturally a perfect fit for hyperconverged infrastructure (HCI): virtual desktops and powerful virtual workstations. Immediately, customers realized the benefits of running desktop workloads where the processing, storage, and networking were fully converged and wanted to bring their most demanding design and visualization applications to the platform. This requires NVIDIA Virtual GPU (vGPU) software to manage GPU-accelerated VMs and enable advanced computer graphics capabilities like generative design, AI-denoising, and real-time ray-tracing, ..
NVIDIA Ampere architecture brings features like Multi-Instance GPU (MIG), which lets infrastructure managers partition a single GPU into multiple instances, each of which gets its own compute, cache, and GPU memory. The architecture also features third-generation Tensor Cores supporting higher performance with different precisions, third-generation NVLink for faster data transfers, structural sparsity, and second-generation RT Cores for faster and more realistic graphics. Nutanix and NVIDIA are working to bring compute and graphics with NVIDIA Ampere architecture-based GPUs to the Nutanix environment.
NVIDIA vGPU technology includes optimizations for accelerating virtual desktop graphics and UI performance. By deploying Nutanix with NVIDIAGPUs optimized for virtualization, organizations can enable a faster, smoother virtual desktop experience and a more productive workforce.
Our collaboration may have started with GPUs for virtual desktops, but as HCI gained traction in the data center, traditional networking was not the right fit for the demands of high performance computing and AI.
In an HCI environment, network traffic flows now move in an east-west pattern, because hosts span multiple access layers and racks, using a converged network for both storage and application services. Adding remote direct memory access (RDMA) using RoCE eliminates the performance and scalability limitations of traditional networks.. Because virtualization and hyperconvergence now extend across data center boundaries and into multi-clouds, the latency and throughput penalties of traditional networking are exacerbated.
NVIDIA-network adapters, such as the ConnectX-6 Lx/Dx and Spectrum switches enable a network fabric that offers predictable, low-latency traffic while achieving maximum throughput and linear scalability. Nutanix is an early adopter of NVIDIA’s end-to-end RoCE RDMA technology for storage traffic acceleration, dating back to 2017. Using RoCE allows us to achieve the highest throughput of data replication within our platform, decreasing rebuild times to ensure reliability and resilience. The NVIDIA Networking with Nutanix Solution Brief,details the best methods to leverage NVIDIA networking solutions for Nutanix.
Nutanix is working alongside NVIDIA to provide customers the ultimate performance for any workload, from virtual desktops and business-critical databases to AI and machine learning. As virtual networking and built-in network security become cornerstones of HCI and hybrid multi-cloud platforms, DPUs are poised to play a critical role to provide hardware based acceleration to these functions. With years of collaboration with NVIDIA on several generations of GPUs, Ethernet switches, NICs, RDMA-enabled NICs, and now DPUs, Nutanix is looking ahead to further improve compute and network performance for hyperconverged infrastructure, making it the de facto standard in the enterprise for all workloads including AI/ML.