Cerebras Systems has shattered barriers in artificial intelligence (AI) technology with the introduction of its groundbreaking Wafer Scale Engine 3 (WSE-3). This latest innovation boasts an extraordinary 4 trillion transistors, each a mere 5nm in size, propelling it to the forefront of AI chip design.
Positioned as the pinnacle of performance, the WSE-3 is purpose-built to accommodate AI models up to ten times larger than the industry-leading GPT-4 and Gemini.
The powerhouse behind Cerebras’ CS3 supercomputer
At the helm of this technological marvel is Andrew Feldman, CEO and co-founder of Cerebras Systems, who has spearheaded the company’s relentless pursuit of innovation over the past eight years.
Feldman expressed profound pride in the realization of the WSE-3, a feat once dismissed as a mere “pipe dream” by skeptics. He proclaimed the WSE-3 as the epitome of speed and efficiency, tailored to address the most pressing challenges in cutting-edge AI research and development.
Cerebras’ commitment to pushing the boundaries of AI extends beyond chip design. The WSE-3 serves as the beating heart of the CS3 supercomputer, the third iteration of the only commercial wafer-scale AI processor.
Feldman highlighted the WSE-3’s remarkable advancements, promising twice the performance of its predecessor, WSE-2, while maintaining identical power consumption and cost—a true testament to Moore’s Law.
Powered by Taiwan Semiconductor Manufacturing Co’s state-of-the-art 5nm manufacturing process, the WSE-3 fuels the CS3 supercomputer with unparalleled computational might. With an impressive 125 petaflops and 900,000 AI-optimized compute cores on a single processor, the CS3 stands as a testament to Cerebras’ unwavering commitment to innovation and performance.
Pioneering AI capabilities for the future
The advent of the CS3 supercomputer marks a pivotal moment in AI research and development. Designed to train the next generation of large language models (LLMs), the CS3 delivers unprecedented compute performance while consuming less power and space—a game-changer for enterprises and hyperscalers alike.
With the ability to train trillion-parameter models with ease, Cerebras empowers organizations to unlock new frontiers in AI applications, from healthcare to scientific exploration.
As businesses grapple with the challenges of rising energy costs and sustainability concerns, Cerebras’ WSE-3 offers a beacon of hope. By delivering enhanced performance without a corresponding increase in power consumption, Cerebras addresses the urgent need for energy-efficient AI solutions.
Collaborating with Qualcomm, Cerebras further optimizes the CS3’s output, slashing inference costs through innovative technologies such as sparsity, speculative decoding, and MX6 compression. This strategic partnership ensures that Cerebras remains at the forefront of AI innovation, providing customers with unparalleled performance and efficiency.
Meeting market demand with unprecedented success
Feldman revealed that Cerebras has already amassed a substantial backlog of orders for the CS3 across a diverse range of sectors, including enterprise, government, and international clouds.
This overwhelming demand underscores the industry’s recognition of Cerebras’ unparalleled technological prowess and the transformative potential of the WSE-3.
Cerebras Systems’ unveiling of the WSE-3 represents a quantum leap forward in AI chip design, setting new standards for performance, efficiency, and scalability. As the world embraces the possibilities of AI-driven innovation, Cerebras stands poised to lead the charge, revolutionizing industries and shaping the future of technology.
