As the demand for AI models continues to grow, companies are seeking more efficient and powerful solutions beyond traditional GPUs. One such alternative is the neural processing unit (NPU) developed by Kneron, a silicon vendor at the forefront of edge AI inference and fine-tuning. Kneron recently unveiled its next generation of silicon and server technology at the Computex conference in Taiwan, showcasing its commitment to advancing AI at the edge.
Kneron’s KL830 NPU, set to debut in 2023, aims to address the global shortage of GPUs. With Qualcomm and Sequoia Capital as investors, Kneron is well-positioned to make significant strides in the industry. In addition to the KL830, Kneron also provided a glimpse into its future offering, the KL1140, slated for release in 2025. These NPUs represent a growing trend among vendors like Groq and SambaNova, who are exploring alternatives to GPUs to improve power efficiency in AI workloads.
One of the key features of Kneron’s update is the introduction of private GPT servers powered by NPUs. These servers can run on-premises, eliminating the need for organizations to rely on large systems with cloud connectivity. The Kneron KNEO 330 system, integrating multiple KL830 edge AI chips, offers an affordable solution for on-premises GPT deployments. Already in use by prestigious organizations like Stanford University, the predecessor KNEO 300 system has proven its value.
While hardware is Kneron’s primary focus, software also plays a crucial role in their technology stack. The company has developed multiple capabilities for training and fine-tuning models on top of their hardware. By combining open models and fine-tuning them to run on NPUs, Kneron offers a comprehensive solution for AI development. Additionally, Kneron supports the transfer of trained models onto their chips through a neural compiler, making it easy for users to utilize models trained with popular frameworks.
One of Kneron’s key advantages is its low power consumption. The KL830 NPU boasts a peak power consumption of only 2 watts while providing consolidated calculation power (CCP) of up to 10eTOPS@8bit. This remarkable efficiency allows Kneron’s chips to be integrated into various devices, including PCs, without the need for additional cooling solutions. The low power consumption and high performance make Kneron an attractive choice for organizations looking to optimize their AI workflows.
To learn more about Kneron’s groundbreaking technology and explore strategies for auditing AI models, don’t miss the AI Impact Tour on June 5th. This exclusive invite-only event in NYC will bring together top executive leaders to discuss the future of AI and its impact on organizations. Request an invite now and secure your attendance at this incredible opportunity to stay ahead in the world of AI.
