Shulou(Shulou.com)11/24 Report--

Thanks to CTOnews.com netizen OC_Formula for the clue delivery! The share price soared 185%, with a market capitalization of more than $1 trillion.

When it comes to who is the biggest winner in the AI circle this year, I believe Nvidia must be on the list.

Although it has won, Nvidia is still not willing to be a GPU and is now targeting another market.

This time the goal is the edge computing chip.

Nvidia secretly acquired artificial intelligence startup OmniML in February, according to The information.

MIT Han Song is one of the co-founders of OmniML, a start-up focused on edge AI computing.

Although in January this year, OmniML announced the establishment of a strategic partnership with Intel, but it did not delay Nvidia to enter the Bureau to engage in harvesting and acquisition immediately in February.

At present, the OmniML official LinkedIn account has been shown to belong to Nvidia, and Google searches the official website https://omniml.ai/, which is also directly redirected to Nvidia's official website after clicking.

Although there is not much news, judging from these signs, the acquisition should be solid.

In May this year, Nvidia reported revenue of $7.19 billion in the first quarter of fiscal 2024 and forecast revenue of $11 billion in the second quarter of this year.

Nvidia, with chips and money, invested in three generative AI unicorns, including Inflection AI, Runway and Cohere, in June alone.

In announcing the financing, Inflection AI said it was working with Nvidia to build one of the world's largest GPU clusters to train AI models.

Data show that Nvidia's annual PC GPU shipments reached 30.34 million yuan in 2022, nearly 4.5 times that of AMD. As of the fourth quarter of 2022, in the independent GPU market, Nvidia has 84% of the market share, and the market capitalization has reached the trillion mark.

However, although it has become the overlord of GPU, Nvidia still faces some awkward situations in the edge computing chip.

For example, in recent MLPerf tests, especially edge computing, Nvidia's AGX Orin is inferior to startup SiMa.AI in ResNet power efficiency.

According to the test results, in the single stream, the energy consumption per frame of the SiMa.AI is 1.45 times that of the Nvidia AGX Orin (the lower the better), and the delay of the SiMa is 27% faster. On the multi-stream side, the gap is 1.39 times, and the delay is 22% faster.

SiMa.AI, a start-up specializing in developing chips for devices from robots to cars to cameras, was founded in 2018 and has so far raised $200m.

According to insiders, the Nvidia acquisition is intended to use OmniML technology to help customers develop AI models faster to improve machine learning accuracy and reduce delays.

Previously, OmniML has publicly said that it is working with customers in areas such as self-driving and smart cameras to build a computer vision edge algorithm optimization platform based on artificial intelligence to improve security and real-time perception.

In addition, in recent years, Nvidia's business has gradually expanded to the field of automotive AI chips.

Edge computing can provide low delay, high security and offline operation ability for industrial robots, autopilot and other fields.

These actions show that Nvidia is very direct this time, that is, to engage in edge computing AI chips.

MIT Han Song Lianchuang, a leading edge computing OmniML founded in 2021 and headquartered in California, received a $10 million seed wheel in March last year.

Omniizer, released by the company last September, is a platform that simplifies and accelerates machine learning operations (MLOps) by bridging the gap between machine learning models and edge hardware.

OmniML has three Chinese co-founders, namely, Dr. Wu Di, CEO of the company, Professor Han Song of the Department of Electronic Engineering and computer Science of MIT, and Dr. Huizi Mao, Chief Technology Officer.

Among them, Han Song is an assistant professor of MIT EECS and former co-founder of Shenzhen Science and Technology, graduated from Tsinghua University and graduated from Stanford University under Professor Bill Dally, chief scientist of Nvidia. He is mainly involved in deep learning and computer architecture.

Prior to this, Professor Han Song's team of MIT proposed an algorithm-system collaborative design framework, which only uses the memory of 256KB and 1MB to achieve in-device training, and the cost is less than 1000 of PyTorch and TensorFlow.

Despite its strong scientific research background, Nvidia acquired OmniML, on the one hand, it must hope to accelerate the layout of the AI market. On the other hand, it may also have something to do with Han Song having studied under Professor Bill Dally, the chief scientist of Nvidia.

Since they all used to be a family, this acquisition makes more sense.

Reference link:

[1] https://www.theinformation.com/articles/nvidia-acquired-ai-startup-that-shrinks-machine-learning-models?rc=riq8lb/

[2] https://www.eetimes.com/mlperf-inference-startups-beat-nvidia-on-power-efficiency/

[3] https://arxiv.org/abs/2206.15472

This article is from the official account of Wechat: qubit (ID:QbitAI), by Sean.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.