Marvell Technology, Inc. introduced the groundbreaking Marvell Teralynx T100 switch silicon. This new product represents the first 102.4 Tbps switch silicon designed specifically for artificial intelligence. The hardware removes legacy elements that normally increase power requirements. Therefore, the device achieves up to 25% lower power than alternative market solutions.
Modern GPU racks face extreme thermal limits. Systems are currently approaching 120KW per rack. This reality forces operators to adopt complex cooling mechanisms. Networking components typically consume 15% to 25% of total rack power. The Marvell Teralynx T100 operates at under 1000W typical power. This efficiency allows operators to install more accelerators within existing power footprints.
Advanced Architecture Benefits AI Clusters
The innovation uses a monolithic device design. Marvell manufactures this component on advanced 3nm process technology. The hardware supports a 512-port scale-out radix. This setup helps reduce necessary network tiers and optical links. Consequently, the simplified architecture effectively drops the total cost of ownership.
“As AI workloads evolve and scale exponentially, hyperscalers require network architectures that optimize latency, power, and scalability simultaneously. The Teralynx T100 was purpose-built for AI, designed without the legacy baggage that inflates power, and engineered to deliver the deterministic performance and efficiency required to scale next-generation data center infrastructure,” said Rishi Chugh, vice president and general manager, Data Center Switch Business Unit, at Marvell.
The flexible pipeline architecture supports diverse interconnect standards. It complies with the Ethernet Scale-Up Networking protocol. The platform also meets the latest Ultra Ethernet Consortium guidelines.
“As hyperscalers expand AI clusters to tens of thousands of accelerators, data center infrastructure becomes a defining factor in network efficiency and performance. The Teralynx T100 architecture delivers significant advantages in latency, power efficiency, radix scalability, and overall TCO, advantages that stem directly from its purpose-built AI design approach and are required for the continued scaling of data center infrastructure,” said Alan Weckel, co-founder and technology analyst at 650 Group.
Operators can select multiple packaging options. These options include ball grid array and co-packaged optics. The system features integrated telemetry and congestion control. Furthermore, the product software includes a toolkit that supports the SONiC operating system. The platform also integrates with the Switch Abstraction Interface from the Open Compute Project. Customer sampling for the switch will begin during the current quarter.
Explore IT Tech News for the latest advancements in Information Technology & insightful updates from industry experts!
News Source: Businesswire.com
