Why did NVIDIA acquire OmniML?

2023-07-04

NVIDIA's acquisition of edge optimization platform OmniML is likely to create ripples in the edge artificial intelligence ecosystem. NVIDIA, a leading player in the field of artificial intelligence, shows no signs of slowing down. The chip manufacturing giant has been actively investing in the AI ecosystem, providing funding to startups such as Inflection AI, RunwayML, and Cohere. It seems that the company is also building its own AI technology stack through acquisitions, as evidenced by its quiet acquisition of OmniML. OmniML is a company that focuses on compressing machine learning models for deployment on edge devices. Last year, it launched a platform called Omnimizer, which applies computationally intensive AI models to low-end hardware. This aligns with what NVIDIA desires, which is why they acquired this startup in February this year. NVIDIA has not yet announced this acquisition, leading to speculation about the reasons behind this move. Considering NVIDIA's position in the field of edge AI, especially in areas such as automotive, robotics, and drones, OmniML may be just what the green team needs. What does OmniML do? Founded in 2021 by Dr. Di Wu, a software engineer from Facebook's PyTorch division, and Dr. Huizi Mao, who previously worked in Google Research's Mobile Vision team, OmniML combines their expertise with Dr. Song Han's pioneering deep compression technology to create OmniML. OmniML's main product is Omnimizer, a platform designed for fast and easy large-scale AI optimization. Omnimizer eliminates the need for different platforms and products to deploy and optimize edge device models, providing a single platform for deployment, training, and measurement. Additionally, the platform optimizes models to run on devices with minimal power consumption. It achieves this by incorporating target hardware into the testing loop, making the models not only smaller and faster but also more suitable for the hardware they run on. The models are optimized using neural architecture search, a technique for finding the best architecture for a given neural network, taking into account variables such as latency and available power. Prior to the platform's launch, OmniML secured $10 million in seed funding, led by GSR Ventures, Foothill Ventures, and Qualcomm Ventures. This funding round demonstrates the strong demand for a platform that can automatically optimize edge models. Through OmniML, companies can easily fine-tune their models to run on devices such as drones, smart cameras, and cars. This effectively frees the models from relying on the cloud, democratizing the ability to run ML models quickly and easily. Considering the value this could bring in the coming years, it's no wonder NVIDIA acquired this company at an early stage. NVIDIA's Edge Dream NVIDIA has three main edge products: NVIDIA EGX for enterprise edge computing, IGX for industrial applications, and Jetson for autonomous machines and embedded edge use cases. The company has also optimized its software to run on these devices, and its customers can use the Fleet Command platform to manage their edge devices. The Omnimizer platform further strengthens this comprehensive product lineup, as it aligns perfectly with NVIDIA's edge strategy. By adding the automatic model optimization technology included in Omnimizer, NVIDIA can make edge models more efficient. This not only allows them to install larger models on smaller devices but also creates more capable models to meet the needs of edge deployment. As mentioned earlier, OmniML keeps the hardware in the loop during optimization. This means that NVIDIA can use OmniML's technology stack to create custom profiles for its hardware suite, ensuring that the deployed models make the most efficient use of edge hardware. The core of this strategy lies in the NGC (NVIDIA GPU Cloud) catalog for enterprises. This catalog provides GPU-optimized software containers for companies seeking a fast and easy way to deploy models on a large scale. The available edge models in the catalog will greatly benefit from OmniML's optimization. If these models are optimized using this technology, NVIDIA will gain significant performance and efficiency improvements in the models provided at the edge. Additionally, customers' hardware can also be included in the optimized software through the hosting platform, creating a well-rounded solution that fits any architecture. Like other acquisitions and investments, OmniML helps solidify NVIDIA's leadership position in the AI market. If OmniML continues to operate as an independent player, it will still bring value to NVIDIA's technology stack, albeit through second-order effects. However, as part of the green giant, OmniML provides unique added value to strengthen NVIDIA's position in the field of edge AI. With NVIDIA's continued investment in the AI ecosystem and acquisitions of startups like OmniML, the company's relentless pursuit in the AI race continues. OmniML's expertise in shrinking ML models for edge deployment perfectly complements NVIDIA's presence in autonomous machines and industrial applications. By integrating OmniML's technology into its edge products, NVIDIA can optimize models for efficient deployment on low-end hardware. This strategic move strengthens NVIDIA's position in the AI market and enhances its comprehensive AI strategy. Source: https://analyticsindiamag.com/why-nvidia-acquired-omniml/

RECENT AI NEWS

RECENT AI TOOLS