-
Notifications
You must be signed in to change notification settings - Fork 22
Commit 20ade4a
hw-mgmt: sensors: Add scales for current and power measurement for GB300 NVLink systems
The outputs of current and power of hotswap devices are not normalized
to the shunt sense resistor on the GB300 switch but set to a generic
1mOhm value.
To properly fix this, the current and power measurements, need to be
multiplied by 3.401 before being sent, (a calculated factor based on
the effective shunt resistor's resistance) to get the correct result.
Note that this factor is only true for GB300 switches, different
layout, and shunt resistor choice, will lead to a different
normalization factor.
When values are normalized, output will be like:
Pin =290.4x3.401 = 987.71
For GB200 and VR systems - correct scaling should be supplied by HW
team.
Tehn, scales are to be added to:
usr/etc/hw-management-sensors/n51xxld_sensors.conf
usr/etc/hw-management-sensors/n61xxld_sensors.conf
There is related NVBug
5449609: Incorrect Hotswap Controller Current Measurements GB300M
Switch
Note, this fix allows correct output by standard sensors utilities.
Additional work is required to support scaling for NOS show command.
This work includes:
(1)
Adding scales in label file, like:
+ "labels_scale_HIxxx_alternatives" : {
+ "pdb_hotswap1_power1_input" : "3.401"
}
}
(2)
Extending archive ui_tree_HIxxx.tar.gz with scale for hotswap power
entries.
(3)
Modify hw_management_parse_labels.py to extend parsing over
"labels_scale_{sku}_alternatives_array".
Bugs #5449609
FR #4380879
Signed-off-by: Vadim Pasternak <[email protected]>1 parent 7af1dfa commit 20ade4aCopy full SHA for 20ade4a
File tree
Expand file treeCollapse file tree
2 files changed
+589
-0
lines changedOpen diff view settings
Filter options
- usr
- etc/hw-management-sensors
- usr/bin
Expand file treeCollapse file tree
2 files changed
+589
-0
lines changedOpen diff view settings
0 commit comments