Nvidia DGX H100 User Manual

NVIDIA DGX H100 User Guide
echo "{\"Targets\":[\"∕redfish∕v1∕UpdateService∕FirmwareInventory∕HGX_0\
,→"]}" > parameters.json
curl -k -u <bmc-user>:<password> -H 'Expect:' --location --request POST�
,→https:∕∕<bmc-ip-address>∕redfish∕v1∕UpdateService∕upload -F
,→'UpdateParameters=@parameters.json;type=application∕json' -F�
,→UpdateFile=@<fw_bundle>
Make sure to specify the nvfw_DGX-HGX-H100x8_0002_xxxxxx.x.
x_prod-signed.fwpkg rmware le.
▶ Update DGX HGX H100 components
To update DGX H100 system components, you need to specify the component name
as a target in a JSON le. The following example updates the host BMC:
echo "{\"Targets\":[\"∕redfish∕v1∕UpdateService∕FirmwareInventory∕HostBMC_
,→0\"]}" > parameters.json
curl -k -u <bmc-user>:<password> -H 'Expect:' --location --request POST�
,→https:∕∕<bmc-ip-address>∕redfish∕v1∕UpdateService∕upload -F
,→'UpdateParameters=@parameters.json;type=application∕json' -F�
,→UpdateFile=@<fw_bundle>
The following targets are available:
▶ HostBMC_0 — This is the DGX H100 BMC.
▶ HostBIOS_0 — This is the DGX H100 BIOS.
▶ EROT_BMC_0` — This is the external root of trust for the host BMC.
▶ EROT_BIOS_0 — This is the external root of trust for the host BIOS.
▶ CPLDMID_0` — This is the midplane CPLD.
▶ CPLDMB_0 — This is the CPU tray CPLD.
▶ PSU_0 to PSU_5 — These are the PSUs.
▶ PCIeSwitch_0 and PCIeSwitch_1 — These are the Gen5 PCIe switches on the
CPU tray.
▶ PCIeRetimer_0 and PCIeRetimer_1 — These are the PCIe retimers on the CPU
tray.
To update a target, change the path ∕redfish∕v1∕UpdateService∕
FirmwareInventory∕HostBMC_0 in the preceding example. For example, for CPU
tray CPLD, specify ∕redfish∕v1∕UpdateService∕FirmwareInventory∕CPLDMB_0.
Make sure to specify the nvfw_DGX-HGX-H100x8_0002_xxxxxx.x.
x_prod-signed.fwpkg rmware le.
▶ Forced Update
The DGX H100 system components rmware is only updated if the incoming rmware
version is newer than the existing version. To override this behavior and ash the com-
ponent anyway, specify the ForceUpdate eld and set it to true.
curl -k -u <bmc-user>:<password> --request PATCH 'https:∕∕<bmc-ip-address>
,→∕redfish∕v1∕UpdateService' --header 'If-Match: *' --header 'Content-
,→Type: application∕json' --data-raw '{"HttpPushUriOptions" : {
,→"ForceUpdate": true}}'
58 Chapter 8. Redsh APIs Support

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