Penguin Computing

Penguin Computing operated as a Linux server company until it acquired in 2003 Scyld Software, a leader in Beowulf cluster management software.[1] Penguin Computing is based in Fremont, California.[2] The company's early software solutions were offered under its Scyld brand and included Scyld ClusterWare for cluster provisioning and management, as well as the Scyld Cloud Manager for cloud-enabled HPC environments.

In 2015, Penguin Computing was awarded a contract with the U.S. Department of Energy’s National Nuclear Security Administration (NNSA) tri-laboratory Commodity Technology Systems program, or CTS-1.[3] Under the $39 million contract, Penguin Computing provided over 7 petaFLOPS of computing power at Los Alamos National Laboratory, Sandia National Laboratories, and Lawrence Livermore National Laboratory. Coincident with this contract win, Penguin Computing built up a division to focus specifically on the federal HPC market.[4]

The CTS-1 contract was one of the first and largest deployments of Intel's Omni-Path high-performance communications architecture.[5] This resulted in Penguin Computing being awarded with Intel's "Partner of the Year - HPC Technical Computing" award.[6]

Penguin Computing was an early contributor to the Open Compute Project (OCP), is a member of the project foundation, and one of a limited number of authorized OCP providers.[7] In November 2015 Penguin Computing announced the developed of its Tundra Extreme Scale (Tundra ES) product line with the intention of applying the findings of the OCP to high performance computing.[8] In 2020, Penguin Computing announced the new Tundra AP line of servers, which support Intel Server System compute modules in an OCP form factor.

In 2009 Penguin Computing launched Penguin Computing On-Demand (POD) which offers high performance computing (HPC) cloud computing. The POD cloud was one of the first remote HPC services offered on a pay-as-you-go monthly basis. Like its high performance clusters, the POD cloud is a bare-metal compute model to execute code, but each user is given virtualized login node. Penguin Computing offers users more than 150 pre-installed commercial and open source applications. POD computing nodes are connected via nonvirtualized 10 Gbit/s Ethernet or QDR InfiniBand networks. The POD Cloud data center has redundant Internet links and user connectivity ranging from 50 Mbit/s to 1 Gbit/s.[9]

At POD's launch Penguin Computing's CEO Charles Wuischpard contended that because of the performance overhead from virtualization, other clouds were not suited to HPC and that the computing nodes allocated to customers may be far apart, causing latency that impairs performance for some HPC programs.[10]

In 2018 Penguin Computing was bought by the publicly traded SMART Global Holdings for a $60 million purchase price, with a further $25 million due if Penguin Computing hits certain profit milestones. The $60 million included the assumption of Penguin Computing’s debts, the company has borrowed $33 million from Wells Fargo to fund a new manufacturing facility. In the first quarter of financial year 2018 Penguin Computing had a gross profits of $10.3 million on sales of $48.5 million. HPC computer clusters were the main source of revenue. Penguin owned 1.2 percent of the HPC server market and when the company was bought by SMART it had ten supercomputers in the TOP500.[11]

• Donald Becker