Large Language Models: The Largeness, the Power and the ‘Zero Shot’ Mystery

Large language models fit the classic model of a red-hot technology in an early stage of commercial viability: there’s more talk about it than knowledge, and FOMO – the fear that your competitors are implementing it at your peril – is helping to drive explosive demand. There’s also an allure and mystery around LLMs: some of the awe-inspiring “zero shot” things they do surprise even the data scientists who trained the models (more on this below). Against this backdrop….

SambaNova: New AI Chip Runs 5 Trillion Parameter Models

Specialty AI chip maker SambaNova Systems today announced the SN40L processor, which the company said will power SambaNova’s full stack large language model (LLM) platform, the SambaNova Suite. Manufactured by TSMC, the SN40L can serve a 5 trillion parameter model, with 256k+ sequence length possible on a single system node, according to the company.

$100M Series B for Generative AI Platform Writer

SAN FRANCISCO — Writer, a generative AI platform for enterprises, announced its Series B funding round of $100 million today. The round is being led by ICONIQ Growth with participation from WndrCo, Balderton Capital and Insight Partners, who led the Series A, and Aspect Ventures, who led the seed round. This round includes participation from […]

HPC News Bytes 20230918: New AMD CPUs and Intel FPGAs, Arm’s IPO and Strategic Pivot, AI for Science

A happy mid-September morning to you. Here’s a hop (4:54) through recent HPC news, including: AMD launches EPYC 8004 CPUs for energy- and space-constrained workloads; Intel announces FPGAs going into its Innovation 2023 event; Arm’s successful IPO and strategic pivot; a report on AI for science….

UK AI Supercomputer, ‘One of the Most Powerful in Europe,’ to Be Installed at Univ. of Bristol

The University of Bristol will host the new AI Research Resource, dubbed Isambard-AI, the UK announced, part of a £900 million supercomputing initiative made public last March. The UK said the system will be one the most powerful in Europe. “The world-class AIRR cluster will vastly increase the UK’s compute….

SiMa.ai Debuts Palette Edgematic for Edge ML Applications

SAN JOSE, Sept. 12, 2023 — SiMa.ai, an embedded edge machine learning company, today launched Palette Edgematic, a free visual development environment designed for organizations to get started with ML at the edge. With Palette Edgematic, SiMa is delivering an onramp to AI and ML at the edge via a no-code approach to creating, evaluating […]

Former Google Cloud President Tariq Shaukat Joins Sonar as Co-CEO

AUSTIN and GENEVA – September 12, 2023 – Sonar, a Clean Code solution provider, announced that Tariq Shaukat has joined the company as co-CEO and a member of the Board of Directors. The company said Shaukat will lead the company with founder and CEO Olivier Gaudin. This model strengthens the leadership team and prepares the company […]

Enfabrica Raises $125M Series B for AI Infrastructure Networking Chips

MOUNTAIN VIEW, Calif. – Sept. 12, 2023 –  Enfabrica Corporation, a startup building converged networking and memory fabric silicon and software for AI and accelerated computing workloads, today announced its close of a $125 million Series B financing round. The oversubscribed round, which increases Enfabrica’s valuation more than 5X, was led by Atreides Management, with […]

HPC News Bytes 20230911: NVIDIA LLM Inferencing; Honeywell and Quantinuum; TSMC in Silicon Photonics; Microsoft Copilot AI Indemnification

As we reflect on the events of 9/11/2001 (it’s still living hell no matter how long ago it happened), let’s quickly (4:52) review last week’s HPC news highlights, including: NVIDIA TensorRT-LLM for faster AI inferencing; Honeywell integrates quantum-hardened encryption keys from Quantinuum; TSMC enters the silicon photonics arena; Microsoft to defend Copilot AI customers; Hyperion Research hosts HPC User Forum

At the HPC User Forum: 2 Full Days in Tucson of HPC and AI

This week saw HPC industry analyst firm Hyperion Research host the HPC User Forum, a supercomputing conference with an end-user emphasis held four times a year (two in the U.S. and two internationally) offering two intensive days of presentations and panels involving commercial and government users, along with hardware and software vendors.