Description:

The NVIDIA DGX-2 is a rack-mount AI supercomputer built for large-scale deep learning and HPC workloads — high-definition video training, natural language processing (NLP), and complex model/data parallelism. Its 16 fully NVSwitch-connected Tesla V100 GPUs act as a single unified-memory system, delivering 2.4 terabytes per second of bisection bandwidth — a 24X increase over prior-generation interconnects — and enabling training of models too large for standard multi-GPU servers. nvidia

This unit (Serial: 0574118000520) is fully populated with all 16 Tesla V100-SXM2-32GB GPUs and includes the complete set of 8 ConnectX-5 EDR/100GbE network cards, matching NVIDIA's original full system configuration for networking and GPU count. System memory (DIMMs) has been removed from this unit and is not included — the CPU trays and all DIMM slots are empty, as shown in photos. Buyers should factor in the cost of sourcing DDR4 RDIMM memory separately to bring the system to a bootable/operational state. Storage (2x Samsung 960GB NVMe SSDs) and mounting rails are included.

What's included:

• NVIDIA DGX-2 chassis
• (16) NVIDIA Tesla V100-SXM2-32GB GPUs
• (2) Intel Xeon Platinum 8168 CPUs, 2.70GHz
• (8) ConnectX-5 EDR/100GbE 1-Port CX555A network cards
• (2) Samsung MZ-1LW9600 960GB NVMe SSDs
• Rack mounting rails
  (If an item isn't pictured or explicitly listed above, assume it is not included.)

Specifications:

• Manufacturer: NVIDIA
• Model: DGX-2
• Serial Number: 0574118000520
• GPUs: 16x Tesla V100-SXM2-32GB (512GB total GPU memory)
• CPUs: 2x Intel Xeon Platinum 8168, 2.70GHz, 24-core
• Networking: 8x ConnectX-5 EDR/100GbE
• Storage: 2x 960GB NVMe SSD
• System Memory: Not included — sold without DIMMs installed
• Form Factor: 10U rackmount, rails included

Condition:
This unit was removed from an active data center environment. System memory has been pulled and is not included. All 16 GPUs, both CPUs, all 8 network cards, and both NVMe SSDs are physically present per photos. Please review all photos closely and message us with any questions before purchasing.