Earlier this week, Sambanova ai, a San Francisco based company that offers Dataflow as a Service, raised a $676 million Series 5B round of funding led by Softbank. The company will use the funds to expand its team and continue its mission to provide cloud-based solutions for big data. The company’s products, which include SN30 and Flow, help customers analyze, visualize, and manage data on a massive scale.
DataScale SN30
Powered by a Cardinal SN30 RDU, SambaNova claims its DataScale SN30 system can deliver a GPT training rate six times faster than an eight-socket Nvidia DGX A100 system. It also provides 12.8TB of memory, compared to the DGX’s 3TB.
SambaNova is a startup with more than 500 employees and a valuation of more than $5 billion. SambaNova is betting its AI-centric systems will provide the requisite infrastructure for the next-generation of AI-accelerator machines. Its new offerings include a state-of-the-art DataScale SN30 system and an AI-centric cloud service.
Earlier this year, SambaNova announced it had raised $1 billion in funding. The company is now positioned as the “AI-enabler,” with plans to bring its data-centric cloud to the masses. It has already signed deals with OTP Bank and the Argonne Leadership Computing Facility (ALCF), two institutions that are renowned for their deep learning expertise. SambaNova’s co-founder, Christopher Re, founded data company Lattice, which was acquired by Apple in 2017.
SambaNova also announced that its AI-centric systems will be making their grand debut at SC22 in Dallas in November. One of the first companies to test the new offerings is Lawrence Livermore National Laboratory, which is a well-known early adopter of HPC technologies. In addition, the company is touting the fact that more companies will be announced at the end of the year.
Dataflow-as-a-Service
Despite its name, SambaNova is a service company, not a hardware company. They have re-invented themselves as a provider of machine learning services. They offer a range of applications, including Computer Vision, Natural Language Processing, and Recommender Systems. Their models include pre-trained deep learning models.
SambaNova’s Dataflow-as-a-Service offers a scalable, extensible AI services platform. It delivers Foundation Models for Natural Language Processing, Computer Vision, and Recommender Systems. It can be co-located on premises or accessed through the cloud. It also includes a library of foundational, industry-specific data, and pre-trained deep learning models. It enables enterprises to tap into SambaNova’s AI system, rather than have to hire their own data scientists. This saves time and resources.
SambaNova’s algorithms analyze billions of data points and provide results within minutes. They also allow for high image fidelity and defect detection. This allows for increased model quality and efficiency.
SambaNova’s DataScale has already set world records in AI performance. It delivers high performance computing and efficiency across applications. Its tensor streaming processors, which are based on a simpler architecture than CPUs, offer six petaops of performance and 3.3kW of power. They are available on demand through SambaNova’s Nimbix Cloud.
SambaNova has raised more than $676 million in funding, led by SoftBank Vision Fund 2 and Temasek. The funding will help SambaNova to continue to disrupt the computational limits of AI software and hardware. Its new DataScale system is currently being tested by the Department of Energy, and several biopharma companies have signed up for SambaNova’s services.
Funding round
Having just closed a $676 million funding round, SambaNova Systems is looking to disrupt the AI hardware market in a big way. Founded by Stanford Professors Kunle Olukotun and Chris Re, the startup has taken an integrated approach to building the world’s most powerful AI hardware.
With its custom AI chips, SambaNova is developing an AI system that analyzes billions of data points and provides results within minutes. Its flagship product is called Dataflow-as-a-Service, which is a subscription-based AI services platform that enables enterprises to use SambaNova’s algorithms without needing to install and maintain a large AI system. This allows organizations to focus on business goals, rather than the technical details of running an AI system.
In December, SambaNova introduced its new hardware, known as DataScale. Its chip uses a reconfigurable “dataflow” architecture that enables applications to drive optimized hardware configurations.
Conclusion
SambaNova is aiming to take on Nvidia Corp. and Cerebras Systems in the artificial intelligence space. The company’s flagship chip is a seven-nanometer, 40-billion-transistor machine that’s designed for artificial intelligence training and computation workloads. SambaNova also launched its Nimbix Cloud, which allows users to run its dataflow-as-a-service. This is a logical step for the company, which wants to make AI systems more accessible.
