Edge AI chip startup Blaize has shown its first boards and tools for low latency, low power industrial designs.
The architecture developed by Blaize, which has design centres in Leeds and Kings Langley, UK, is optimized for running AI at the edge of the network. The Blaize Graph Streaming Processor (GSP) architecture has been implemented in a fully programmable chip with 16 cores. This delivers 16TOPS of AI inference performance for a power consumption of 7W.
The GSP chip has been used for a system-on-module (SoM) board and a PCI express card. These are programmed by a software development kit called Picasso or a code-free tool called AI Studio. The latency for an AI framework running on the chip is under 20ms. The chip can run five frameworks simultaneously with a total latency under 100ms.
“Today’s edge solutions are either too small to compute the load or too costly and too hard to productize,” said Dinakar Munagala, Co-founder and CEO, Blaize. “Blaize AI edge computing products overcome these limitations of power, complexity and cost to unleash the adoption of AI at the edge, facilitating the migration of AI computing out of the data centre to the edge.”
The P1600 SoM, shown above by Santiago Fernandez-Gomez, VP of Platform Engineering, is aimed at embedded systems and includes an ARM processor with video encode and decode as well as camera and USB interfaces so that no host processor is needed.
The Xplorer PCIe 3.0 boards are intended for an edge AI server or appliance. The X1600E is a small form factor accelerator platform for small and power-constrained environments such as convenience stores or industrial sites. The X1600P is a standard PCIe-based accelerator in a half-height, half-width form factor that can scale to eight boards that provide up to 64TOPS of AI inference performance.
Both the Picasso software development kit (SDK) and AI Studio tool use Blaize Netdeploy with edge-aware algorithms to get the best accuracy and performance for edge deployments. These use standard frameworks such as TensorFlow or PyTorch converted vias NetDeploy.
“There’s a huge software gap going from research models to production AI apps – this can take months if not a year. The AI software suite reduces this to days and is built on open standards to break the proprietary GPU code block,” said Rajesh Anantharaman, head of product and strategy at Blaize.
Customer samples of both product lines are available now with full production expected starting in Q4 2020. The Blaize Xplorer X1600E is selling at $299 in volume, while the industrial grade Pathfinder P1600 SoM costs $399 in volume, and the Blaize Xplorer X1600P is available for $999 in volume.
The technical storage or access is strictly necessary for the legitimate purpose of enabling the use of a specific service explicitly requested by the subscriber or user, or for the sole purpose of carrying out the transmission of a communication over an electronic communications network.
The technical storage or access is necessary for the legitimate purpose of storing preferences that are not requested by the subscriber or user.
The technical storage or access that is used exclusively for statistical purposes.The technical storage or access that is used exclusively for anonymous statistical purposes. Without a subpoena, voluntary compliance on the part of your Internet Service Provider, or additional records from a third party, information stored or retrieved for this purpose alone cannot usually be used to identify you.
The technical storage or access is required to create user profiles to send advertising, or to track the user on a website or across several websites for similar marketing purposes.