Cerebras Systems, a leading innovator in AI processor technology, has announced a groundbreaking collaboration with Qualcomm, a global leader in wireless technology. The partnership aims to redefine the landscape of AI processing with the launch of the CS3 system, the third iteration of Cerebras’ wafer-scale AI processor.
CS3 system: Revolutionizing AI processing
The CS3 system is built on cutting-edge TSMC 5nm process technology, offering unparalleled computing and memory density. Unlike traditional processors, the CS3 system leverages a complete wafer of chips, eliminating the need for slicing and reconnecting individual chips. With 900,000 AI cores and 44 GB of on-wafer memory, the CS3 system boasts four trillion transistors, enabling lightning-fast processing times for generative AI tasks.
Cerebras’ proprietary software stack ensures seamless scalability across CS3 clusters, significantly reducing the development effort required for distributed AI processing. This efficiency has earned Cerebras the support of renowned organizations such as the Mayo Clinic and GlaxoSmithKline.
To meet the demands of large-scale AI applications, Cerebras has introduced the MemoryX parameter server, complementing on-wafer SRAM with a 2.4 Petabyte appliance. This innovative approach outperforms traditional GPU-based clusters, offering faster and larger AI capabilities in a single rack.
Recognizing the limitations of wafer-scale processors for large-throughput inference processing, Cerebras has joined forces with Qualcomm to offer an end-to-end AI platform. Qualcomm’s Cloud AI100 appliance, optimized for energy efficiency, seamlessly integrates with Cerebras’ training stack to deliver inference-target-aware output, reducing inference costs by 10X.
Optimizing AI models for inference processing
Qualcomm AI Research’s expertise in optimizing AI models for mobile Snapdragon chips has enabled Cerebras to implement sparsity, speculative decoding, and MX6 compression techniques. These advancements ensure optimal performance on the Cloud AI100 Ultra platform, which has garnered support from industry giants such as AWS and HPE.
Cerebras and Qualcomm are poised to revolutionize AI processing efficiency by combining forces. With the CS3 system for training and the Cloud AI100 for inference, customers can now benefit from a seamless, high-performance AI workflow. This collaboration surpasses previous benchmarks and sets a new standard for AI processing innovation.
The partnership between Cerebras Systems and Qualcomm marks a significant milestone in the evolution of AI processing technology. With the launch of the CS3 system and the integration of Qualcomm’s Cloud AI100 appliance, the industry is witnessing a paradigm shift in AI processing efficiency. As organizations strive to harness the power of AI, Cerebras and Qualcomm, stand at the forefront, offering unparalleled solutions to meet future demands.
