A company led by the legendary chip architect Jim Keller, the brilliant mind behind AMD’s Zen chips and Tesla’s original self-driving processor, has finally unveiled its first hardware product after years of development. The firm, Tenstorrent, has launched “Grayskull” – a powerful new processor based on the RISC-V architecture that aims to be a more programmable and scalable alternative to traditional GPU accelerators for AI workloads.
Grayskull is specially designed to excel at handling sparse and dynamic computation patterns that are common in modern AI models. This allows it to be more efficient than GPUs for certain AI tasks.
Along with the Grayskull chip itself, Tenstorrent has also released two “DevKit” products for AI developers – the Grayskull e75 and the higher-end e150 model. These are self-contained accelerator cards that slot into a PC over PCIe and are meant for running and optimizing AI models. They come bundled with Tenstorrent’s software tools for easily deploying pre-trained models or customizing new ones.
The launch of Grayskull is a major milestone for the Santa Clara company. But the timing is especially fortuitous, as just days ago Tenstorrent was selected by Japan’s Leading-edge Semiconductor Technology Center (LSTC) to provide the core RISC-V and chiplet designs for a revolutionary 2nm AI accelerator chip. This ambitious project aims to push the boundaries of AI performance in Japan using Tenstorrent’s innovative architecture.
For developers eager to get started with Tenstorrent’s new hardware, the Grayskull e75 DevKit is available now for $599, while the higher-powered e150 model is $799. And with the might of Grayskull’s performance behind them, these could be the chips that help unlock the next wave of AI innovation.
