Cisco N9364F-SG3-L | Nexus 9300 64-Port 1.6T OSFP Switch, 102.4Tbps, Liquid-Cooled

Q: What type of cooling infrastructure is required for the N9364F-SG3-L?
A: The N9364F-SG3-L is a 100% liquid-cooled switch that integrates into standard data center racks with facility-side liquid cooling infrastructure. The design requires no bulky tanks or major overhauls, using direct-to-chip liquid cooling for the Silicon One G300 ASIC and supporting components to achieve 70% better energy efficiency than equivalent air-cooled solutions.

Q: Can this switch support lower speed optics, or is it limited to 1.6T?
A: While designed for 1.6T OSFP optics, the N9364F-SG3-L supports multiple port speeds including 800G, 400G, 200G, and 100G through port breakout configurations. This flexibility allows deployment in mixed-speed environments and provides investment protection as you scale from current 400G/800G infrastructure to 1.6T.

Q: What are the primary use cases for the N9364F-SG3-L?
A: This switch is purpose-built for high-density AI cluster back-end networking, including GPU-to-GPU training fabrics, large-scale inference deployments, and hyperscale data center spine layers. The 252MB shared buffer and hardware-based congestion management make it ideal for handling the synchronized microburst traffic patterns common in distributed AI workloads.

Q: Which network operating systems are compatible with this switch?
A: The N9364F-SG3-L supports Cisco NX-OS for traditional data center deployments, Cisco ACI for policy-driven automation, and SONiC for open networking environments. This multi-OS flexibility allows integration into existing Nexus 9000 fabrics or new disaggregated architectures.

Q: How does the Silicon One G300 improve AI workload performance?
A: The G300 ASIC delivers up to 2.5x increased burst absorption through its 252MB fully shared buffer, hardware-based congestion management that reacts 100,000x faster than software alternatives, and rich programmable diagnostics. These capabilities result in up to 28% reduction in job completion time for AI training workloads by eliminating packet loss during traffic bursts.