Nvidia DGX H100 Service Manual

NVIDIA DGX H100 Service Manual
After you rule out external connectivity issues, contact NVIDIA Enterprise Support to receive a replace-
ment card.
When you receive the card, begin the replacement by performing the following actions:
▶ Power o the system.
▶ Open the motherboard tray IO door to access the rear section of the motherboard. Refer to
Motherboard Tray - Opening and Closing the IO door for more information.
11.3. Remove the Non-Functional Card
First, turn the locking mechanism 90 degrees so the card can be extracted from the PCI slot:
1. Conrm the motherboard tray service lid is open and loosen the thumb screw for the PCI card
locking mechanism next to slots 1 and 3:
2. Rotate the locking mechanism to release the PCI cards, as shown in the following gure:
3. Pull the PCI Ethernet card from the slot in the riser:
68 Chapter 11. Network Interface Card Replacement

Nvidia DGX H100 Specifications

General

GPU	8x NVIDIA H100 Tensor Core GPUs
GPU Memory	640 GB HBM3 (80GB per GPU)
System Memory	2 TB DDR5
Form Factor	6U rackmount
Storage	30TB NVMe SSD
Power Supply	10kW
Interconnect	NVLink 4.0