Urgent.News

What's breaking now, across thousands of outlets.

Tech

AMD buys chip startup that hardwires AI models into its silicon

Taalas' current chip runs a small version of Meta's Llama 3.1, though the company is working on chips for bigger and more advanced models.

Advanced Micro Devices (AMD) has acquired Taalas, a Toronto-based startup specializing in chips designed for AI model inference. This strategic move comes as the demand for AI chips continues to surge, outpacing the supply of advanced AI chips available in the market. Taalas' technology, which involves custom, or hard-wired chips for a single AI model, promises to deliver output thousands of times faster than traditional GPUs for specific models.

The acquisition, though AMD has declined to disclose the exact purchase price, follows closely on the heels of Nvidia's $20 billion acquisition of Groq, the highest-value transaction in Nvidia's history. Taalas, which has raised a total of $219 million in venture funding since its founding in 2023, currently runs a small version of Meta's Llama 3.1 model.

The company is developing chips capable of supporting even larger and more advanced models. Its current chips utilize an older TSMC process, with speedy SRAM memory integrated directly onto the chip. Taalas' CEO, Ljubisa Bajic, asserts that the startup is capable of transforming any AI model into custom silicon, a process that can be realized in hardware within just two months of receiving a previously unseen model.

AMD CEO Lisa Su emphasizes that while GPUs remain a cornerstone of AMD's AI strategy, acknowledging their flexibility to accommodate newly developed AI models, the acquisition underlines the company's commitment to offering integrated systems with various components and chips, aiming to provide a more comprehensive solution to the evolving AI landscape.

Written by urgent.news from CNBC's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

This story

This is one outlet's version. Read the fullest account.

Read the original at cnbc.com →

More in Tech

7 JWT Security Mistakes I See in Almost Every Auth Implementation

Most JWT security problems don't come from a broken library. They come from a handful of small decisions made once during setup and never revisited: which algorithm to trust, where to store the token…

  • Trusting the algorithm header can lead to algorithm confusion attacks like "alg: none" bypass.
  • Storing tokens in localStorage exposes them to unauthorized access via scripts.
  • Skipping claim validation beyond signature checks leaves tokens unchecked for expiration and issuer.

The Synth Sounded Fine When Nobody Was Listening

This is my entry for DEV's Summer Bug Smash — Smash Stories. There is a specific kind of bug that makes you doubt your own ears. I had a browser-based audio engine that played fine in export and sounded broken in real time. Hit play, and the output was thin, muffled, strangled — like someone had thrown a blanket over the speakers.

  • Browser-based audio engine muffled in real-time playback
  • Live playback path ran on scheduler, offline render had none
  • Fix increased backward-seek threshold to prevent node cancellation

More from Thursday 6 August →