512 H200

GPU Nodes (x64): ThinkSystem SR680a V3, Dual Intel Xeon Platinum 8558, 2.1GHz (48 core each), 3 TB TruDDR5 5600MHz RAM, 6.4TB NVMe storage + dual 1GB NVMe system storage, 8 x NVIDIA H200 GPUs each with 141GB HMB3e GPU memory, NVIDIA ConnectX 7 400Gbps InfiniBand Networking, additional 400Gbps InfiniBand for network storage, and fully redundant power.

 

Network:

NVIDIA Quantum-2 InfiniBand networking, with up to 3,200Gbps of aggregate bandwidth per node.

All GPU nodes also have 400Gbps ROCE to storage.

All switches have redundant power.

 

Storage (2x)

ThinkSystem SR655 Dual EPYC 9227 2.5 GHz (48 total cores), 256GB RAM, 215TB NVMe Data Island, 10Gbps + 400Gbps ROCE, redundant power. Additional storage available on-demand.

 

Reliability and Security You Can Trust: End-to-end SOC2 certified. Enjoy redundancy, availability, and reliability across all deliverable facets including power, cooling, network, fire suppression, and security in a U.S. East Coast tier 3 data center

 

Fast, Flexible & Scalable Connectivity: Enjoy a reliable 10 Gb/s internet connection through redundant carriers for seamless connectivity.

Cloud Peering - Easily peer with major cloud providers.

 

Premium Support and Packaging

24x7x365 Support - Our dedicated team is available around the clock to assist with any technical needs via our ticket-based system. We feature proactive monitoring, spare inventory, and next-business-day OEM hardware replacement.

 

Pricing: Discounts available based on number of servers reserved. Your access is guaranteed for at least 30 days. Optionally, you may extend your term. Either party may terminate upon 2 weeks notice (with the earliest termination being the end of month 1).

 

Please see https://www.corvex.ai/ for more information or reach out via email.