NVIDIA at

A team of NASA scientists and engineers are using Summit, the world's fastest supercomputer at Oak Ridge National Laboratory (ORNL) to simulate retropropulsion using NASA's FUN3D, NVIDIA V100 GPUs, NVIDIA IndeX and GPUDirect Storage. This allows researchers to fly through the massive dataset in real time, volumetrically, and even navigate through it while the simulation data continuously updates.

The stakes are high as scientists and researchers are racing against the clock to find a cure for COVID-19. Finding the right drug requires screening billions of drug candidates and identifying which drugs will most favorably bind with the virus, which is extremely computationally intensive. NVIDIA GPUs can accelerate the process of screening drug candidates by 33x and scanning over 1B compounds in 12 hours.

Combining the compute power of more than 200,000 NVIDIA GPUs and additional computing resources contributed by volunteers, the Folding@Home project has performed an exascale simulation of the COVID-19 spike. Using NVIDIA Omniverse to visualize this simulation output gives us an idea of the dangerous dance performed by the virus’ spikes, but also demonstrates regions where potential treatments could attack.

Climate simulations produce large amounts of inherently 3D data, and yet the analysis is often limited to 2D projections. The NVIDIA Omniverse platform enables the fusion of large-scale scientific data with cinematic rendering capabilities, allowing for interactive exploration of complex climate phenomena. In this demo, we'll show the power of Omniverse to visualize high-resolution climate simulation data.

HPC users often don’t always have easy access to the latest software, and system admins have limited resources to maintain complex software environments. Containers address these challenges by packaging the application and its dependencies. The NVIDIA NGC catalog offers containers for the latest versions of AI, HPC, and visualization software.

NVTAGS is a toolset that enables faster simulation of HPC applications with high GPU-communication-to-application-runtime ratios. It automatically profiles application communication, extracts the system GPU communication topology, and intelligently finds an efficient GPU assignment that minimizes communication congestion, thereby improving application performance.

The NVIDIA Mellanox UFM Cyber-AI platform harnesses AI-powered analytics to minimize downtime in InfiniBand data centers. System admins can now quickly detect and respond to potential security threats and address upcoming failures, saving costs and ensuring consistent customer service.

Modern-day machine learning data centers require complex computations and fast, efficient data delivery. See how NVIDIA Mellanox Scalable Hierarchical Aggregation and Reduction Protocol, or SHARP dramatically improves the performance of machine learning workloads.

NVIDIA SimNet is a GPU-accelerated simulation toolkit based on physics-informed neural networks. This demo shows SimNet using medical imaging data to calculate the full spatial-temporal velocity field in a blood vessel aneurysm. With SimNet, medical researchers can get a far better estimation of the velocities, pressures, and wall shear stresses in the blood vessel, which can help lead to better, more personalized treatment.

View the latest CUDA kernel analysis capabilities in NVIDIA Nsight Compute, including the popular Roofline Analysis Method and a new feature for the NVIDIA Ampere GPU Architecture. We'll demonstrate profiling the hardware-supported asynchronous data copy feature, which can boost the performance of your workloads.

Introducing Application Replay in NVIDIA Nsight Compute, opening the door for new workloads and workflows to take advantage of the powerful CUDA kernel profiling capabilities in Nsight Compute. You'll see how Nsight Compute replays kernels to get accurate performance data and how Application Replay can improve performance and unlock new options for CUDA kernel analysis.

NVIDIA Nsight Systems now includes support for tracing NCCL usage in your CUDA application, enabling users to identify NCCL activity on the CPU timeline and correlate it to the associated GPU CUDA kernels and memory copies. Equipped with this information, users can validate and improve NCCL usage by identifying CPU and GPU cold spots and hot spots.

NVIDIA Nsight Systems now traces CUDA memory allocation to ensure optimal memory usage. Effective memory management is key to ensuring efficient application performance. With this information, users can ensure that their application is reclaiming available memory to avoid any out-of-memory starvation or stalls. Nsight Systems is the premier profiling tool for gaining a holistic view of CUDA applications.