Tenstorrent To Offer AI Workstation For Developers
By Sally Ward-Foxton, EETimes (June 25, 2024)
SANTA CLARA, Calif. — Tenstorrent’s first generation AI chips will soon come in a workstation format for AI developers, the company’s CEO Jim Keller told EE Times.
The new Tenstorrent-powered workstation, internally dubbed the “Quiet Box” because of its silent liquid cooling system, uses eight Wormhole (first generation) AI chips across four PCIe cards. It will be aimed at AI developers. Keller is quick to point out that there is no Nvidia-powered AI workstation equivalent on the market.
Giving EE Times a sneak preview of the new workstation, developed with an OEM partner, Keller said Tenstorrent is working hard on getting the most important five AI models running on the Quiet Box at full speed. Part of this work has involved rewriting the models—twice—to reduce their size. All five are now under 600 lines of code, Keller said, with Tenstorrent’s version of StableDiffusion under 500.
The Quiet Box could be on the market as early as August.
To read the full article, click here
Related Semiconductor IP
- NPU IP Core for Mobile
- NPU IP Core for Edge
- Specialized Video Processing NPU IP
- HYPERBUS™ Memory Controller
- AV1 Video Encoder IP
Related News
- Rambus Cryptography Research Unveils Latest Release of DPA Workstation Analysis Platform
- Tenstorrent Achieves First-Pass Silicon Success for High-Performance AI Processor SoC Using Synopsys' Broad DesignWare IP Portfolio
- Jim Keller joins Tenstorrent as President and CTO.
- Tenstorrent Selects SiFive Intelligence X280 for Next-Generation AI Processors
Latest News
- Jim Keller: ‘Whatever Nvidia Does, We’ll Do The Opposite’
- FlexGen Streamlines NoC Design as AI Demands Grow
- IntoPIX Presents Its New Titanium Software Suite: Empowering AV-Over-IP Workflows With Speed, Quality & Interoperability
- Global Semiconductor Sales Increase 2.5% Month-to-Month in April
- Speedata Raises $44M to Launch First-Ever Chip Designed Specifically for Accelerating Big Data Analytics - Compute's Second Largest Workload