Temperature Limit for CPU/GPU Throttling

Autonomous Vehicles DRIVE AGX Thor DRIVE AGX Thor General
1

DRIVE OS Version: 7.0.3

Issue Description: Temperature Limit for CPU/GPU Throttling

I’ve read the Thermal Design Guide for the Jetson Thor Modules and just wanted to verify that keeping Tj < 90 C will prevent any throttling (in other words it is the “recommended operating temperature threshold at which the module will operate without performance reduction”).

The table 2-1 is confusing since:

  • note 4 (The Thor SoC will shut down the Thor Module once this temperature limit is reached to protect the
    Thor SoC    ) is applied to both the 90 and 118 C lines
  • and note 5 (This is the temperature threshold below which the product will operate at the specified clock speeds.) is applied to the 90, 115, and 118 C lines

Thanks,
Rich

5

Dear @rich.frederiksen ,
Could you point to reference document related to this query.
If you want to know the recommended operating temperature of DRIVE AGX Thor devkit, please check Mechanical installation guide at NVIDIA DRIVE Documentation | NVIDIA Developer

6

Here’s a link to the document I was referring to: Jetson Thor Series Modules - Thermal Design Guide

7

We were thinking more about how to properly monitor a running Thor SoC to know when it’s approaching a temperature where CPU/GPU throttling might occur.

8

Dear @rich.frederiksen ,
Did you try temperature read_soc_temp in MCU console?

9

I tried the read_soc_temp command in the NvShell and it’s a little confusing. It gives the following:

NvShell>read_soc_temp
Info: Executing cmd: read_soc_temp, argc: 0, args:
Soc Instance: SOC 0
I:RptrID - 0x810C PsCd - 0x12 - TMON Read local temp pass
 INFO: I2C_IOHWABS: Transaction blocked for 2 ms
I:RptrID - 0x810C PsCd - 0x13 - TMON Read remote temp pass
 local temp = : 98.812500
remote temp = : 102.437500
reading tmp-451 temperature is successfull
Command Executed

Tegrastats outputs the following with temperatures ranging from 45.531 to 51.625 C ( 113.96 - 124.92 F).

07-07-2026 13:12:57 RAM 16792/60085MB (lfb 32x4MB) CPU [100%@2599,100%@2600,100%@2600,100%@2599,100%@2599,100%@2600,100%@2599,100%@2600,100%@2600,100%@2599,100%@2600,100%@2600,100%@2600,100%@2599] GR3D_FREQ 0% cpu@51.625C tj@51.625C soc012@51.468C gpu@45.531C soc345@46.093C

Do you know what the units are for read_soc_temp and what the local/remote temps are?

In either case though what would be the temperature at which throttling would start to occur? The 90 C limit from the doc I attached and the tj output from tegrastats seem like the best bet at the moment.

11

Dear @rich.frederiksen ,
DRIVE OS does not throttle. it will shutdown. To view temperature you can use cat /sys/class/thermal/thermal_zone*/type and cat /sys/class/thermal/thermal_zone*/temp . It is matching to tegrastats temp output for nodes.