A complete, rack-integrated NVIDIA A100 deployment sold as a single lot.
Each of the 96 compute nodes is an 8-GPU A100 80GB SXM4 system with NVSwitch and eight 200Gb/s ConnectX-6 NICs, built on ZT Systems hardware and installed four per rack across 24 racks — 768 A100 GPUs and 61.4TB of combined HBM2e in total.
Technical specifications
Cluster
- Compute nodes
- 96
- Racks
- 24 (4 nodes per rack)
- Total GPUs
- 768× NVIDIA A100 80GB SXM4
- Total GPU memory
- 61.4TB HBM2e
- Platform
- ZT Systems
Compute — per node
- GPU model
- 8× NVIDIA A100 80GB SXM4
- GPU interconnect
- NVSwitch
- CPU model
- 2× AMD EPYC Rome, 64-core
- Accelerator
- 1× FPGA
Memory and storage — per node
- Memory
- 2TB DDR4-2933
- NVMe
- 16× M.2 NVMe slots via 4× PCIe carrier cards
Networking and power — per node
- Fabric networking
- 8× NVIDIA ConnectX-6 200Gb/s
- Power supplies
- 6×
Notes
- No Redfish or BMC web interface — management is via SSH and serial-over-LAN. The original owner’s centralized management platform is not included.
- Firmware updates from ZT Systems are not included or guaranteed.
- Sold as one complete lot. Freight, rigging and site logistics are quoted per destination.
Compliance & Important Information