Highlights:

  • According to Modular, their Mojo language offers a memory safety technique like Python’s but without a noticeable performance hit.
  • The business also offers a software platform called AI Engine along with the programming language.

Modular Inc., the company developing programming languages designed specifically for building artificial intelligence applications, disclosed securing fresh finance worth USD 100 million.

General Catalyst spearheaded the investment. The GV startup fund of Alphabet Inc. and several other institutional funders also participated. Modular will use the funding to improve its AI programming language and its other software product, the AI Engine, which claims to speed up neural networks for businesses.

AI models are typically created by developers using the Python programming language. Its syntax is generally short and straightforward, making it easier to design neural networks than in other languages. But the performance suffers because of its simplicity: Python-written programs sometimes run slowly.

Mojo, a programming language created by Modular, is promoted as a quicker alternative. The syntax of the language is quite similar to that of Python, making it reasonably simple to use. According to Modular, the main distinction is that Mojo is up to 35,000 times faster.

Python’s so-called memory safety mechanism is one of the factors causing its sluggish performance. Because of this approach, programmers can prevent typical bugs like buffer overflows, which happen when a server’s RAM is misused by an application. While automating that process allows developers to save time, the code becomes slower as a result.

According to Modular, its Mojo language offers a memory safety technique like Python’s but without a noticeable performance hit. The language promises to make the work of developers easier in other ways as well. Mojo lessens the manual code necessary to guarantee that an AI model functions appropriately on various chips.

The business also offers a software platform called AI Engine along with the programming language. The platform is intended to accelerate existing AI models used by the companies, including those not created in Mojo. The AI Engine, according to Modular, may boost neural networks’ inference performance by a factor of more than seven without requiring any code modifications.

The platform accelerates AI models by converting their source code to the Mojo programming language. It also employs several other improvements that can enhance the hardware effectiveness of a neural network.

An AI program uses computations involving data snippets known as constants to perform inference. Such computations can be made by Modular’s AI Engine during compile time when programmers convert an AI’s source code into a usable application. Processing is sped up because there is no longer a need to repeat the calculation when the neural network makes an inference.

Additionally, the platform takes advantage of operator fusion, a performance enhancement method. It includes integrating two AI parts into one more effective one that runs on less hardware overall.

Following a USD 30 million fundraising round, Modular unveiled the AI Engine and Mojo in June. More than 120,000 developers, according to the business, have shown interest in the two technologies since then. While Mojo will launch early next month, The AI Engine is presently accessible through an early access program.