The integration of validated designs is becoming increasingly crucial for establishing trust and compliance in artificial intelligence (AI). These designs enhance transparency and accountability, facilitating adherence to regulations and ethical standards. By transforming AI from a potentially risky experiment into a dependable tool, validated designs are accelerating innovation in the sector.
According to Zeus Kerravala, founder and principal analyst of ZK Research Inc., the collaboration between major companies like Cisco and Nvidia, as well as partnerships such as Pure Storage and Arista, exemplifies the shift towards validated designs. Kerravala stated, “These turnkey systems to me are the right way to go for an enterprise because when you start to deploy these things, there’s a lot of dials you can turn and levers you can pull in order to tweak the systems. That’s why the validated designs are important.”
His comments came during an interview with Dave Vellante from theCUBE at the NYSE Wired: AI Factories – Data Centers of the Future event in March 2024. They discussed how validated designs are not only driving innovation but also accelerating the adoption of AI technologies.
Key Components for Effective AI Systems
Validated designs thrive on a combination of fast storage, high-speed networks, and powerful processors. Kerravala emphasized the importance of these elements, stating, “You need three things to make sure your AI works, fast storage, fast network and fast processor.” He explained that if network speeds do not align with the data access speeds on the storage side, the entire system may fail.
The synergy between these components is crucial. For instance, InfiniBand technology offers ultra-low latency and high-speed connectivity, making it well-suited for demanding AI workloads such as distributed deep learning. Meanwhile, Ethernet provides a cost-effective solution for AI inference and data ingestion. This combination enables the creation of AI infrastructure that is both high-performance and adaptable, capable of managing larger models and expanding datasets.
Kerravala pointed out the historical context of these technologies, stating, “If you roll back the clock 30 years, there were all these different protocols and Ethernet beat them all.” He noted that while InfiniBand remains a strong option, Ethernet’s familiarity makes it a preferred choice for many users.
The Rise of Neocloud Providers
The emergence of neocloud providers is reshaping the landscape of cloud computing, driven by the growing demands of AI workloads. These providers focus on regional data residency and introduce a new tier in cloud architecture specifically designed for AI computation. Kerravala believes this evolution will necessitate enterprises to reconsider their data center architectures.
He noted, “If you’re a Global 2000, I do think that you do need to think about how to re-architect your data centers.” He emphasized the need for a balanced approach, where organizations not only focus on scaling out but also on ensuring vertically integrated performance within their systems.
The conversation between Kerravala and Vellante illustrates the critical role of validated designs in enabling effective AI systems. As the industry continues to evolve, the integration of these designs will be essential for organizations aiming to maintain compliance while fostering innovation in their AI initiatives.
For those interested, the complete video interview is available as part of SiliconANGLE and theCUBE’s coverage of the NYSE Wired event.