Lightmatter got a big boost in its effort to develop photonic AI chips instead today when it announced a Series B round of $80 million. The money will help the company bring its first round of products to market, Lightmatter’s CEO said.
Lightmatter is one of a handful of companies aiming to build processors that uses light instead of electrons. In addition to using less energy than traditional electronic processors, adherents say photonic processors also aren’t bound by Moore’s Law, which theoretically gives them more room to grow.
The company, which emerged from the Massachusetts Institute of Technology four years ago, has already developed its first chip, called Envise, which uses a Mach Zehnder interferometer used to encode data with light, as HPCwire’s George Leopold wrote last year.
After data is encoded into light, it’s processed by the 256 photonic tensor cores that sit on each Envise chip, which also contains a photonic controller, a graph processor, 500MB of RAM, and a network interface. Lightmatter says the chips, which it’s positioning as AI accelerators, are capable of performing common functions like ReLU, GELU, sigmoid, and tanh, and support INT8, INT16, and bfloat16 numeric operations.
The company also sells systems equipped with Envise chips, such as the Envise blade, which packs 16 chips into a 4U server. Lightmatter claims that its Envise blade is capable of performing seven times the work of an GPU-based NVIDIA DGX-A100 system at the equivalent wattage.
Lightmatter is developing a photonic AI accelerator
AI and machine learning, which is exploding in popularity, are the primary workloads targeted by Lightmatter with its photonic chips. Nick Harris, the company’s co-founder and CEO, says the $80 million investment will help the enable customers to embrace AI and machine learning workloads without the environment costs.
“This investment is a strong endorsement of our approach to delivering higher performance, lower environmental impact AI compute solutions to the global technology community,” Harris stated in a press release. “Funds will be used to accelerate production and go-to-market of our first generation roadmap products, and build out our sales and operations teams.”
The Series B round was led by Viking Global Investors, with participation from GV (formerly Google Ventures), Hewlett Packard Enterprise (HPE), Lockheed Martin, Matrix Partners, SIP Global Partners, Spark Capital, and others. Lightmatter has raised $113 million total to date.
Paul Glaser, vice president and head of Hewlett Packard Pathfinder, said Lightmatter’s offerings are a “clear differentiator.” “Lightmatter unlocks performance improvements over current offerings while delivering significant power savings,” he said.
Related Items:
AI Chip Startup Syntiant Scales Production
Chip Startup Merges AI, 5G at the Edge
New AI Chips to Give GPUs a Run for Deep Learning Money
The Link LonkMay 07, 2021 at 04:13AM
https://ift.tt/3f5gsvE
Lightmatter Raises $80M for Photonic AI Chips - Datanami
https://ift.tt/2RGyUAH
Chips
No comments:
Post a Comment