Blaize releases edge AI platform with 16 TOPS performance at 7w

News Highlights:

1> Blaize announced that the embedded and accelerated platforms built on the GSP architecture have a 60x improvement in sampling efficiency, a 10x reduction in latency, a 50x reduction in memory bandwidth, 16TOPS processing power at 7W power consumption, and are 100% programmable.

2> Blaize software tools provide higher accuracy and performance, faster deployment, and the industry's first code-free AI development option at the edge.

3> Blaize's early customers are looking to deploy previously infeasible real-time AI edge vision use cases in industrial, autonomous transportation, smart retail, subway and factory applications.

Blaize recently released the company's first artificial intelligence computing hardware and software products, designed to overcome the unmet requirements of current artificial intelligence application computing and productization. Compared with traditional GPU/CPU solutions, the Blaize Pathfinder and Xplorer platforms combined with the Blaize AI software suite enable developers to create a more practical and commercially viable artificial intelligence product in a wide range of edge uses and industries.

“Today’s edge solutions are either too small to compute the load or too expensive to be productive,” said Dinakar Munagala, co-founder and CEO of Blaize. “Blaize AI edge computing products overcome performance, complexity, and cost limitations to unleash AI applications at the edge and facilitate the migration of AI computing from the data center to the edge.”

As the demand for AI at the edge changes entire industries and AI capabilities become a "must-have" requirement for new products, the edge AI market is expected to grow rapidly. Omdia Research predicts that the edge chip market will reach $51 billion in 2025.

Blaize focuses on system and energy efficiency, low latency, and flexibility. Blaize embedded platforms and accelerator platforms are built on the Blaize Graphics Streaming Processor (GSP) architecture, which is designed for the needs of edge computing. With 16 GSP cores and 16 AI inference performance at only 7W of power consumption, GSP provides up to 60 times the system-level efficiency for edge AI applications compared to GPU/CPU. In addition, GSP reduces memory bandwidth by 50 times and reduces latency by 10 times. This efficiency is achieved without sacrificing flexibility - the Blaize GSP is a 100% programmable processor - and has advanced features including multi-threading and pipeline processing.

The specific products are as follows:

Blaize Pathfinder P1600 Embedded SOM

The P1600 embedded system-on-module (SOM) brings the programmability and efficiency benefits of GSP to embedded edge AI applications deployed at the sensor edge or network edge, without the need for an additional host processor.

Blaze Xplorer Accelerator Platform

The Blaze Xplorer platform is designed to accelerate enterprise-grade edge AI applications by simply plugging into a PCIe slot in a host server or device. The X1600E is a small form factor accelerator platform for small and power-constrained environments such as convenience stores or industrial sites. Due to the high efficiency of GSP, which is Blaze's unique accelerator form factor, it can be easily added to AI applications in industrial PCs or card slots in small 1U servers.

The X1600P is a standard PCIe-based accelerator with a half-height, half-width form factor. The X1600P can easily replace power-hungry desktop GPUs in edge servers and deliver 16-64 TOPS AI inference performance within a low power range.

Blaize AI Software Suite

The Blaize AI software suite is built on open standards and is easily accessible, freeing product developers from being locked into proprietary development environments. The software suite, consisting of the Picasso Software Development Kit (SDK) and AI Studio, a completely code-free visual interface, provides development tools for traditional developers and non-programmers. In addition, both toolsets leverage Blaize Netdeploy, an innovation with edge-aware algorithms for optimal edge deployment accuracy and performance.

Pioneering use cases with customers

Blaize customers are leveraging Blaize’s unique product features to make real-time AI use cases at the edge a reality. Example applications include:

Industrial: Runs five independent neural networks at 50FPS with less than 100ms total latency, monitoring human and robot pose detection and position, product ID and counting, and product quality.

Smart City: Three independent neural networks running at 50FPS monitor human detection, posture, location, automatic detection, and intersection safety and security.

Sensor Fusion: High-resolution FHD video and LiDAR/Radar sensor fusion running as a complete graphical native application on the Blaize GSP.

Last mile delivery: Sensor fusion of HD cameras and LiDAR/radar at 30 fps, less than 100 ms latency, and less than 15 watts of power.

Retail: Four independent neural networks, running at 50FPS, monitor human detection, location, masks, and theft for store safety and security, as well as product inspection.

Blaize Pathfinder and Xplorer Pricing

Customer samples of both product lines are available now, with full production expected in the fourth quarter of 2020. The Blaize Xplorer X1600E is priced at $299 in volume, the Pathfinder P1600 SOM is priced at $399 in volume, and the Blaize Xplorer X1600P is priced at $999 in volume.

About Blaize

Blaize provides transformative edge computing solutions for AI data collection and processing at the edge of the network, focusing on intelligent vision applications including automotive, retail, security, industrial and metro. To date, Blaize's strategic investors include DENSO, Daimler, SPARX Group, Magna, Samsung Catalyst Fund, Temasek, GGV Capital, Wavemaker and SGInnovate.