Home Computing Lemurian Labs is building a new compute paradigm to reduce cost of running AI models

Lemurian Labs is building a new compute paradigm to reduce cost of running AI models

by Amelia Ramiro

Lemurian Labs, an early-stage startup founded by alumni from Google, Intel, and Nvidia, has announced a $9 million seed investment to develop a new chip and software that aims to make processing AI workloads more accessible, efficient, cheaper, and environmentally friendly. The company’s goal is to reimagine accelerated computing by flipping the traditional approach of making data travel to compute resources. Instead, Lemurian Labs wants the compute to move to the data, minimizing the distance data needs to travel.

In an interview with TechCrunch, Jay Dawani, co-founder and CEO at Lemurian, explained that computing comes down to three things: math, memory, and movement. Traditionally, data gets stored in memory, moves through an interconnect to a math unit, gets manipulated, and then gets written back in memory. Lemurian’s approach aims to change this by moving the compute resources around the data, rather than making the data travel.

The company plans to change the math on the chip to achieve this vision, which is no small task. In the early days of chip development, engineers opted for a floating-point approach because a logarithmic approach couldn’t be made to work. However, Lemurian claims to have solved this problem by developing a log number system that turns expensive multiplies and divides into adds and subtractions, leading to savings in area, energy, and speed.

Lemurian Labs will release the software part of the stack first, which is expected to be generally available in Q3 next year. Developing the hardware will be more challenging and will require time, money, and testing in production. The company currently has 24 employees, mostly highly skilled technical engineers with a background in this field. They plan to hire six more people in the coming months and, if successful in securing a Series A funding round, another 35 in the next year.

The $9 million seed investment was led by Oval Park Capital, with participation from Good Growth Capital, Raptor Group, Alumni Ventures, and others.

While building and bringing a chip to market is a significant and expensive challenge, Lemurian Labs believes that if they can deliver on their vision, it could substantially reduce the cost and improve the efficiency of building generative AI models and future technologies.

You may also like