Artificial intelligence is rapidly shifting the landscape of scientific computing, enabling researchers to tackle complex problems with unprecedented speed and precision. Just as high performance computing has accelerated scientific experimentation with simulation capabilities, newer AI-driven systems are now poised to push the boundaries even further. AI can assist scientists by automating data analysis, optimizing simulations, and uncovering patterns and insights in vast data sets, complementing and enhancing the powerful capabilities of traditional HPC. 

One impressive and unique new system is Venado, an AI supercomputer recently installed and launched at Los Alamos National Laboratory in collaboration with Hewlett Packard Enterprise and Nvidia. 

Venado is an exascale-class HPE Cray EX supercomputer that is all liquid-cooled and houses 2,560 NVIDIA GH200 Grace Hopper superchips. These energy-efficient GH200 chips combine an Arm-based NVIDIA CPU with an H100 GPU and can execute millions more instructions per second compared to older chip technology. 

Venado also contains 920 NVIDIA Grace CPU superchips, lending an architecture that is heavier in CPUs than one might expect, standing out among other GPU-centric AI systems. The Grace CPU superchips swap the GPU for a second Grace CPU for a total of 144 Arm cores linked by a NVLink-C2C interconnect. 

This Grace-Grace architecture is especially beneficial for scientific applications that are not well suited for GPU accelerators. Venado was built to address larger and more intricate problems such as those in multi-physics and coupled physics that cannot be fully accelerated due to the complexity of the physics involved. Many of LANL’s applications can run in the millions in terms of lines of code, causing memory challenges and slower results due to sparse or irregular access to memory. As a large boost to memory bandwidth and performance, the Grace CPU superchips feature up to 960GB of LPDDR5x memory capable of delivering upwards of 1TB/sec of bandwidth. 

Performance like this ensures Venado will accelerate LANL’s integration of AI capabilities for research in areas like materials science, renewable energy, and astrophysics. LANL noted that early testing has shown promising results for atomistic simulations for materials science and high-resolution astrophysics simulations. The system is networked with HPE Slingshot 11 interconnects and features HPE Cray supercomputing software for modeling and simulation workloads. 

The speedy new supercomputer is installed in LANL’s Nicholas C. Metropolis Center for Modeling and Simulation. Venado’s name is fitting, as it is the Spanish word for deer and also the name of one of the highest peaks in the Sangre de Cristo mountains in New Mexico. 

LANL is no stranger to supercomputing and has housed many groundbreaking systems over the decades, many of which have been dedicated to specific missions. Venado is unique in its multi-disciplinary potential and will serve as an institutional resource that will allow researchers the freedom to explore a diverse range of problems in various fields with an array of computational modalities. 

LANL says Venado is the outcome of a codesign process that will form the basis for ongoing collaboration focused on developing a broad spectrum of computing, memory, and software technologies. The lab says this codesign process seeks to fuse the combined knowledge of vendors, hardware architects, system software developers, domain scientists, computer scientists, and applied mathematicians. This collaborative process allows Venado to optimize both its hardware and software components for specific, intricate research challenges, resulting in a highly customized machine. 

Venado’s distinctive architecture, multi-disciplinary focus, and unique codesign process ensure it is positioned to address complex research challenges in this new era of scientific computing.