Technology

Blaize Graph Streaming Processor
The New Compute Architecture

Blaize Graph Streaming Processor™ (GSP) architecture and graph native Picasso software are a fusion of breakthrough hardware and software technologies – each optimizing the other – to clear the barriers to widespread AI adoption. AI must be useable to be useful, and must fit into real-life budgets of energy, cost, size and complexity.

Rethinking everything from transistors to boards to software, Blaize achieves leaps in energy efficiency, AI training to inference integration, and broad usability.

New Class of Silicon

The fully programmable GSP architecture leverages task level parallelism and streaming execution processing in an intuitive manner, enabling developers to take advantage of extremely low energy consumption, high performance and superior scalability. In comparison, existing GPUs and FPGAs exert a jaw-droppingly higher energy price, while CPUs cost more and scale poorly, and all are subject to excessive latency due to their sequential execution processing architectures.

New Computing in Action: Blaize Graph Streaming Processor Architecture

The Blaize Graph Streaming Processor architecture busts through the limitations of older architectures to enable greater energy efficiency, lower latency, reduced size, and significant cost savings for AI workloads.

Get GSP whitepaper

New Class of Software

Blaize GSP and the Blaize Picasso software development platform enable AI applications developers to build entire applications optimized for edge deployment and run efficiently in a complete streaming fashion. We’re making the benefits of AI easier to deliver through innovations in the software tool chain that reduce the time and complexity of getting AI models running on hardware.

Blaize develops software for new generation computing, where simplicity rules, and the user experience is the focus. Software workflows that do not require customers to learn a new system in order to utilize the silicon. Tools designed to handle tremendous complexity without being complex to use.

"The path forward for computing in automotive and mobility applications should be implemented on a more unified and flexible platform like a Graph (Data Flow) compute-based architecture. Blaize graph-native GSP computing technology is the first practical, programmable and viable solution. Together with Blaize, NSITEXE would like to boost flexible Graph (Data Flow) compute technology based accelerator coverage."

-Hideki Sugimoto, CTO of NSITEXE

Fast, Flexible & Cost-effective Product Delivery

AI customer and market needs rarely stand still. Blaize architecture is equally dynamic—it is intrinsically process portable—ready to port to the most advanced manufacturing process technology as needs dictate. Process changes take weeks vs. months required for legacy chips architectures.

Contact