tt_tdp_limit_w

Name

Prometheus Metric Name

tt_tdp_limit_w

Metric Path (tt-telemetry)

Schema:

{hostname}/tray{tray}/chip{chip}/tt_tdp_limit_w

Example path:

bh-glx-c09u02/tray1/chip2/tt_tdp_limit_w

Description

The Thermal Design Power (TDP) limit in watts that the ARC firmware throttler currently enforces. When power consumption (tt_tdp_w) approaches this limit, firmware throttles the AI clock to keep the device within it.

This is distinct from tt_board_power_limit_w. The board power limit is the static maximum the board is configured to draw, whereas this metric reports the limit the throttler is actively enforcing, which can be reprogrammed at runtime. The limit applies per ASIC, is not arbitrated between callers (the last write wins), and reverts to the board default on chip reset. Firmware accepts limits in the range 50 W to 500 W, and refuses any limit above the board’s configured maximum.

Because the limit can be lowered out from under a running workload, comparing this metric against tt_tdp_w is the way to tell whether a device is being power-capped below its board maximum.

This metric is only created if the device’s firmware reports the TDP limit. It requires Wormhole firmware 18.4 or newer, or Blackhole firmware 19.8 or newer; older firmware does not publish the value and no metric is created.

Values

Type: Unsigned Integer

Units: Watts (W)

Allowable values: A non-negative integer representing the enforced TDP limit in watts. For example, 300 means firmware throttles the device to stay within 300 watts. A value of 0 means the board never configured a TDP limit; the value firmware reports is published as-is, so consumers that treat the limit as a ceiling should filter out 0.

Related Metrics

Metric	Description
tt_tdp_w	The current power consumption. This is the value that `tt_tdp_limit_w` throttles against.
tt_board_power_limit_w	The board power limit. The static maximum the board is configured to draw, which the enforced TDP limit cannot exceed.
tt_ai_clock_mhz	The AI clock frequency, which firmware reduces when the TDP limit is reached.

Prometheus Labels

Label Name	Value
hostname	The host from which the metric was collected.
hall	The datacenter hall where the host is located. Sourced from the Factory System Descriptor (FSD).
aisle	The datacenter aisle where the host is located. Sourced from the Factory System Descriptor (FSD).
rack	The rack number where the host is located. Sourced from the Factory System Descriptor (FSD).
shelf_u	The shelf U position in the rack where the host is located. Sourced from the Factory System Descriptor (FSD).
tray	The tray (UBB) that the device is located on.
chip	The ASIC location within the tray.
unit	The unit of measurement. Always `"W"` for this metric.