News
DealsCorvex agrees USD 103 million take-or-pay deal to perform AI inference with Mid-Atlantic GPUsSovereignNeysa scales AI infrastructure in India with AMD Instinct MI350X GPUs on TytanInferenceMicrosoft Foundry adds Baseten, DeepSeek AI inference (MSFT:NASDAQ)ClustersDataOne and Rowan University sign engineering co-op program agreement for DataOne's Vineland AI factoryClustersLook inside Cal Poly's $3 million AI Factory (Video)DealsOpenAI Taps AMD Chips for 457 MW of AI Capacity in Six-Year TensorWave DealClustersCal Poly launches an AI data center on campus. Take a look insideInferenceArista takes aim at the next AI network battleground: distributed inferenceDealsOpenAI Signs 457 MW AI Data Center Deal as AMD Expands Its Challenge to NvidiaClustersBehind Dell AI Factory鈥檚 NVIDIA & AMD Enterprise ExpansionDealsCorvex agrees USD 103 million take-or-pay deal to perform AI inference with Mid-Atlantic GPUsSovereignNeysa scales AI infrastructure in India with AMD Instinct MI350X GPUs on TytanInferenceMicrosoft Foundry adds Baseten, DeepSeek AI inference (MSFT:NASDAQ)ClustersDataOne and Rowan University sign engineering co-op program agreement for DataOne's Vineland AI factoryClustersLook inside Cal Poly's $3 million AI Factory (Video)DealsOpenAI Taps AMD Chips for 457 MW of AI Capacity in Six-Year TensorWave DealClustersCal Poly launches an AI data center on campus. Take a look insideInferenceArista takes aim at the next AI network battleground: distributed inferenceDealsOpenAI Signs 457 MW AI Data Center Deal as AMD Expands Its Challenge to NvidiaClustersBehind Dell AI Factory鈥檚 NVIDIA & AMD Enterprise Expansion
EveryNeocloud
Selene

Selene

Nvidia's own DGX SuperPOD supercomputer, built in three months in 2020

Selene is a supercomputer built and used by Nvidia, entering service in 2020. It is a DGX SuperPOD system combining AMD CPUs, Nvidia A100 GPUs and Mellanox HDR InfiniBand networking, reaching 63.460 petaflops and placing fifth in the world when it joined the Top500 list.

Share on WhatsApp

  • Reached 63.460 petaflops, fifth fastest in the world on entry
  • Built in three months during 2020
  • 4,320 Nvidia A100 GPUs with 1,080 AMD Epyc CPUs
  • Trains BERT in under 16 seconds
  • Top MLPerf results among commercial systems through December 2021

What it is

Selene is an Nvidia supercomputer that came into service in 2020. It is built on the DGX SuperPOD design, a ready-made high performance machine that pairs DGX compute nodes with fast storage and high bandwidth networking for demanding machine learning work. The system combines 1,080 AMD Epyc CPUs with 4,320 Nvidia A100 GPUs, linked by Mellanox HDR InfiniBand. On entry to the Top500 ranking it reached 63.460 petaflops, placing fifth worldwide. It was also described at the time as the fastest industrial system in the United States and the second most energy efficient supercomputer ever measured.

How it came about

Nvidia assembled Selene in just three months during 2020. It was designed as an internal machine for the company's own AI research and as a demonstration of the DGX SuperPOD, the turnkey supercomputing package Nvidia sells to customers with heavy machine learning demands. When it entered the Top500 list it was the fifth fastest computer in the world, the fastest industrial system in the US, and second only to one other machine for energy efficiency.

The compute

The machine uses Nvidia's DGX building blocks: AMD Epyc processors paired with A100 GPUs, Mellanox HDR InfiniBand for the network fabric, and fast storage alongside the compute nodes. InfiniBand is a high speed interconnect that lets thousands of GPUs exchange data quickly. Selene trains BERT, a natural language model, in under 16 seconds, a task smaller systems take roughly 20 minutes to finish. Through December 2021 it led every commercially available system on MLPerf, the training and inference benchmark run by a consortium drawn from universities, research labs and industry.

Customers and deals

Selene is Nvidia's own system rather than a commercial service. Its work includes training BERT for language tasks and GauGAN2, the model behind Nvidia Canvas, which learned from 10 million landscape images and combines segmentation mapping, inpainting and text to image generation in one model.

Its place in AI

Selene served as both a working research machine and a reference for the DGX SuperPOD product line. It held the top MLPerf results among commercially available systems through December 2021 and was used for coronavirus research, including protein docking and quantum chemistry problems relevant to understanding the virus and possible treatments.

Questions people ask about Selene

What is Selene?

A supercomputer built by Nvidia in 2020 and used for the company's own AI research. It is based on the DGX SuperPOD design, a packaged system of DGX compute nodes, fast storage and high bandwidth networking aimed at demanding machine learning work.

How fast is it?

Selene reaches 63.460 petaflops. That placed it fifth in the world when it entered the Top500 ranking, and it was the fastest industrial system in the United States at that time.

What hardware does it use?

It combines 1,080 AMD Epyc CPUs with 4,320 Nvidia A100 GPUs, connected over Mellanox HDR InfiniBand networking, with fast storage attached to the compute nodes.

What has it been used for?

It trains language models such as BERT in under 16 seconds, powered Nvidia's GauGAN2 image model, and supported coronavirus research into protein docking and quantum chemistry.

Good reading and sources

Where this page comes from

The facts on this page come from the Wikipedia article "Selene (supercomputer)", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryNeocloud rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.

It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

Photo credits on this page

The small pictures above come from Wikimedia Commons under free licences. Some are cropped to fit.

The weekly digest

Every Monday: the provider of the day, new pages on EveryNeocloud, and the conferences and anniversaries of the week ahead.