Nervana Systems

Nervana Systems was an artificial intelligence software and hardware company founded in 2014 and based in San Diego and Palo Alto, California. The company focused on deep learning, a form of machine learning built around multilayer neural networks, and sought to give businesses the tools and computing infrastructure needed to create custom AI applications. Its offering was designed as a full-stack platform. At the software level, Nervana developed neon, an open-source deep-learning framework intended to support high-performance model training. The framework emphasized assembler-level optimization, multi-GPU processing, and algorithmic techniques such as Winograd convolution, which can reduce the computational cost of operations commonly used in neural networks. Nervana positioned neon as a performance-oriented alternative to other frameworks available at the time, including Caffe, Theano, Torch, and TensorFlow. In February 2016, the company announced Nervana Cloud, a software-as-a-service environment based on neon. The service was intended to allow organizations to develop and run custom deep-learning software without building every part of the underlying infrastructure themselves. At launch, Nervana Cloud used Nvidia Titan X graphics processors. In parallel, Nervana was developing its own application-specific integrated circuit, known as the Nervana Engine, for deep-learning workloads. The company described the chip as a purpose-built alternative to general-purpose graphics processors, with a design that removed hardware elements primarily associated with graphics and concentrated computing resources on neural-network operations. Nervana claimed that the architecture could provide substantially greater performance than Nvidia's Maxwell-generation GPUs, although the cited performance expectation was a company claim rather than an independently established market result. Nervana was led by chief executive officer Naveen Rao, chief technology officer Amir Khosrowshahi, and vice president of algorithms Arjun Bansal. It raised approximately $28 million in disclosed funding, including a $20.5 million Series A round announced in June 2015. The round was led by Data Collective and included participation from several institutional and venture investors. The company remained relatively small, with contemporary estimates placing its workforce at about 48 employees when Intel acquired it. Intel announced the acquisition of Nervana Systems on August 9, 2016. The reported transaction value was approximately $408 million. The acquisition gave Intel access to Nervana's deep-learning software expertise, engineering talent, and specialized accelerator work as competition in artificial-intelligence computing intensified. After the acquisition, Nervana's technology and personnel were incorporated into Intel's AI efforts rather than continuing as an independent company. Intel later discontinued development under the Nervana name. In January 2020, Intel said it was shutting down Nervana development in favor of technology obtained through its acquisition of Habana Labs. As a result, Nervana Systems is best understood as a defunct AI technology brand whose principal legacy was its attempt to combine optimized deep-learning software, cloud delivery, and dedicated neural-network hardware within one platform.

Nervana Systems was established in 2014 by Naveen Rao, Amir Khosrowshahi, and Arjun Bansal. The founders built the company around the idea that deep-learning adoption required more than isolated algorithms: organizations also needed efficient software frameworks, scalable computing services, and hardware designed specifically for neural-network workloads. Nervana therefore pursued a full-stack model spanning software, cloud infrastructure, and custom silicon. The company's software work centered on neon, an open-source deep-learning framework. Nervana promoted neon as a high-performance system that could compete with contemporary frameworks such as Caffe, Theano, Torch, and TensorFlow. Its stated performance strategy included low-level assembler optimization, support for multiple GPUs, and the use of Winograd methods for convolution calculations. These methods were intended to improve the speed and efficiency of model training, particularly for workloads involving convolutional neural networks. Nervana expanded this software approach into a hosted service. Announced in February 2016, Nervana Cloud was a software-as-a-service platform based on neon. It was designed to help businesses develop custom deep-learning applications using hosted compute resources rather than assembling a complete machine-learning stack independently. The initial environment used Nvidia Titan X GPUs, while the company worked on a proprietary alternative for future deployments. That alternative was the Nervana Engine, an application-specific integrated circuit designed for deep learning. Unlike graphics processors developed for a broad range of visual and computational tasks, the proposed chip was intended to devote its resources to neural-network operations. Nervana argued that removing legacy graphics functions could increase compute density and improve efficiency. The company publicly stated that the Nervana Engine could deliver performance roughly ten times greater than Nvidia's Maxwell architecture, although this was a company estimate rather than a verified independent benchmark. The business raised approximately $28 million in funding. Its $20.5 million Series A financing, announced in June 2015, was led by Data Collective and supported by investors including Allen & Company, AME Cloud Ventures, Playground Global, CME Group, Draper Fisher Jurvetson, Fuel Capital, Lux Capital, and Omidyar Network. Despite its ambitious technology scope, Nervana remained a small organization; estimates placed its employee count at approximately 48 at the time of its acquisition. Intel acquired Nervana Systems on August 9, 2016, for a reported approximately $408 million. The transaction reflected the strategic importance of machine-learning acceleration and gave Intel an established AI team, a deep-learning software stack, and a specialized accelerator program. Nervana no longer operated as an independent company after the acquisition, and its capabilities were folded into Intel's broader artificial-intelligence initiatives. Intel ultimately moved away from continuing development under the Nervana identity. In January 2020, it shut down Nervana development in favor of technology from Habana Labs, which Intel had acquired. Nervana Systems consequently became a defunct brand, while its work remains part of the competitive history of early enterprise deep-learning platforms and dedicated AI accelerator design.

• 2020

  Nervana development is discontinued

  [1]Nervana Systemsen.wikipedia.org

  Intel ends development under the Nervana program and favors technology from Habana Labs.

• 2016

  Nervana Cloud is announced

  [1]Nervana Systemsen.wikipedia.org

  The company introduces a cloud-based deep-learning development service built on its neon framework.

• 2016

  Intel acquisition

  [1]Nervana Systemsen.wikipedia.org

  Intel acquires Nervana Systems for an estimated $408 million.

• 2015

  Series A financing

  [1]Nervana Systemsen.wikipedia.org

  Nervana raises a $20.5 million Series A round led by Data Collective, contributing to approximately $28 million in total reported funding.

• 2014

  Nervana Systems is founded

  [1]Nervana Systemsen.wikipedia.org

  Naveen Rao, Amir Khosrowshahi, and Arjun Bansal establish the company to develop deep-learning software, cloud services, and specialized hardware.

A performance-focused, full-stack artificial intelligence company combining optimized deep-learning software, cloud infrastructure, and purpose-built neural-network hardware.

• Open-source deep-learning framework

  neon

  neon was Nervana's open-source framework for developing and training deep-learning models. It was designed around performance, using low-level optimization, multi-GPU support, and Winograd-based convolution techniques. Nervana presented it as a competitive alternative to frameworks including Caffe, Theano, Torch, and TensorFlow. The framework was later discontinued.

• Artificial intelligence software-as-a-service platform · 2016

  Nervana Cloud

  Nervana Cloud was a hosted platform announced in February 2016 for developing custom deep-learning software. Based on the neon framework, it provided access to cloud computing resources and initially ran on Nvidia Titan X GPUs. It represented Nervana's effort to package its software and infrastructure into a service for business users.

• Deep-learning application-specific integrated circuit

  Nervana Engine

  The Nervana Engine was a purpose-built accelerator under development for deep-learning workloads. Its architecture was intended to omit general-purpose graphics features and concentrate hardware resources on neural-network computation. Nervana claimed that it could substantially outperform Nvidia's Maxwell-generation GPUs, but the cited performance expectation was a company claim rather than an independently verified result.

• 2020

  Intel shifts away from Nervana developmentStrategy

  Background

  Intel had acquired Habana Labs and evaluated its technology alongside the assets obtained from Nervana.

  Action taken

  Intel shut down development associated with Nervana and chose to prioritize Habana Labs technology for its AI accelerator strategy.

  Aftermath

  The Nervana brand and its independent product-development program were discontinued.

  Source

• 2016

  Intel acquires Nervana SystemsM&A

  Background

  Nervana had developed a combined deep-learning software, cloud, and accelerator technology stack while AI computing was becoming strategically important to semiconductor companies.

  Action taken

  Intel acquired the company on August 9, 2016, integrating its personnel and technology into Intel's artificial-intelligence efforts.

  Aftermath

  Nervana ceased operating as an independent company. Intel later discontinued development under the Nervana name in favor of technology from Habana Labs.

  Estimated acquisition value. $408 million (August 9, 2016)

  Source

• 2020

  Intel ends Nervana development in favor of Habana Labs technology[1]Nervana Systemsen.wikipedia.org

  Other

  Intel discontinued development associated with Nervana and redirected its AI accelerator strategy toward technology from Habana Labs, another Intel acquisition.

• 2016

  Nervana announces Nervana Cloud[1]Nervana Systemsen.wikipedia.org

  Product launch

  The company introduced a cloud-based service built around its neon framework for developing custom deep-learning applications.

• 2016

  Intel acquires Nervana Systems[1]Nervana Systemsen.wikipedia.org

  M&A

  Intel acquired Nervana Systems, bringing its deep-learning software, engineering team, and neural-network accelerator program into Intel's artificial-intelligence business.

• 2015

  Nervana Systems raises a $20.5 million Series A round[1]Nervana Systemsen.wikipedia.org

  Other

  Nervana announced a Series A financing led by Data Collective, with participation from a group of venture and strategic investors, to support its deep-learning software, cloud, and hardware work.

1. 1Nervana Systemsen.wikipedia.org