Los Alamos National Laboratory (LANL) has debuted its Venado supercomputer. The beast is capable of ten exaFLOPS of performance, specifically designed for AI workloads for energy department. The supercomputer will be utilized to integrate artificial intelligence and machine learning for fundamental science research and national security.

Installed at the Nicholas C. Metropolis Center for Modeling and Simulation, Venado is designed and installed by Los Alamos National Laboratory in collaboration with Nvidia and Hewlett Packard Enterprise (HPE).

Venado Supercomputer’s capacity and architecture

Stationed at New Mexico LANL, the Venado supercomputer is outfitted with 2560 of Nvidia’s Grace Hopper Superchips and Hewlett Packard Enterprise’s proprietary Slingshot 11 Network connectivity, which is especially designed for the AI computational requirements of supercomputers. Venado is also one of the first supercomputers built with Nvidia’s designed superchips. 

The system is entirely liquid cooled and it boasts 920 Grace-Grace CPUs and 2560 GH200 (Grace Hopper) super chips, totaling 3480 of Nvidia’s Superchips. The GH200 is the main system module for processing AI tasks, it has a 72-core Grace CPU and 480GB of memory. Grace CPUs are capable of a bandwidth of 1TB per second as they swap the GPU for  a second Grace CPU. Thom Mason, director of LANL, said,

“Venado adds to our cutting-edge supercomputing that advances national security and basic research, and it will accelerate how we integrate artificial intelligence into meeting those challenges.”

Source: LANL.

Source: Straits research. Venado lives up to LANL expectations

The overall system is quite compact, despite the fact that Venado is capable of handling giant scale artificial intelligence applications with its Nvidia Grace Hopper Superchips, an Nvidia Arm-based super computing unit, and an Nvidia Hopper architecture. Early tests have shown that Venado can produce significant results in simulations for astrophysics and simulations for material science. The focus is on workflow efficiency rather than extreme accuracy. 

Venado is a powerful supercomputer for what it is, suited for large language models processing, as the claimed score of ten exaflops it achieved is by compromising on accuracy for a higher flow rate or processing speed, but this is what LLMs require, so a suitable candidate for artificial intelligence. But compared to the AMD Frontier system, it’s still not a match, but not a ‘Slowpoke Piston’ by any means. As it will do fair in managing lower precision workloads, which are the demand of the day due to artificial intelligence requirements. The deputy secretary of the U.S. Department of Energy, David Turk, said,

“Our supercomputing capabilities represent a critical component of how national laboratories tackle important problems.”

Adding further,

“With its ability to incorporate artificial intelligence approaches, we are looking forward to seeing how the Venado system at Los Alamos can deliver new and meaningful results for areas of interest.”

Source: LANL.

Venado is a beautiful example of codesign by the combined expertise of different vendors. The LANL has been getting a fair share of the federal government’s support, especially under the Biden administration. As the US government has seriously taken the artificial intelligence and has been preferring research and development regarding machine learning. Venado is a mighty supercomputer with 2560 Nvidia GH200 Superchips, but the British government’s project Isambard will house 5448 of the same Superchips and is expected to debut this year.

Find Los Alamos National Laboratory’s note here.