Validating an AI model on NUCLEO-N657X0-Q board in overdrive mode using STM32Cube AI Studio

[stes11]

Hello,

I’m using STM32Cube AI Studio to optimize and validate an ONNX model on a NUCLEO-N657X0-Q board. In general, everything works fine. However, I would like the board to be configured in overdrive mode (npu_freq = 1000MHz, mcu_freq = 800MHz) by Cube AI Studio for the validation runs on the target because I’m hoping to achieve a better inference time.

When executing new “runs” in Cube AI Studio, the device attributes are always as seen in the screenshot below.

Results overview of a run in Cube AI Studio

If I use the the ST developer cloud to optimize and benchmark my model with the STM32N6570-DK board (because the NUCLEO-N657X0-Q is not available in the DC), the device attributes are as follows:

Device attributes for the benchmark in the developer cloud

Since those two boards have the same STM32N657X0H3Q MCU, I expect a similar inference time on my NUCLEO-N657X0-Q board with Cube AI Studio if I could just activate the overdrive mode.

 

I didn’t find any settings in the Cube AI Studio user interface regarding the overdrive mode, so I tried two things:

1. Creating a new project from scratch in Cube AI Studio and edit the .ioc file in the root directory of the project to overdrive mode because I was hoping that Cube AI Studio takes this .ioc file as a template for the various runs. Unfortunately, this was not the case. The runs were executed in the “default configuration” as shown in the first screenshot.
2. Creating a new CubeMX project according to this article: 

   I left out the part with X-CUBE-AI because I understand that it is not recommended anymore to use it and to use Cube AI Studio instead. After that, I imported the CubeMX project (.ioc file with overdrive configuration) into Cube AI Studio and ran the optimization and validation. Unfortunately, mcu_freq = 600MHz and npu_freq = 800Mhz is still used. I would have expected that the configuration from the imported .ioc file is used. I’ve attached the file “FullBridge.ioc”.

 

To summarize, my question is if it’s somehow possible to let the NUCLEO-N657X0-Q board run in overdrive mode while performing the optimization and validation process with Cube AI Studio. If not, is there some workaround with which I can achieve this (e.g. without using Cube AI Studio) ? 

 

Thank you in advance!

[hamitiya]

Hello ​@stes11 

You are right, with STM32N6570-DK, STM32Cube AI Studio uses overdrive mode where with NUCLEO-N657X0-Q, it does not.

It is because the Out-Of-The-Box demo, which is delivered with STM32N6570-DK Discovery Kit, enables the overdrive mode by default (for the best performance), where the MCU stays without overdrive for the Nucleo. Since overdrive enablement is not reversible (it break fuses in the microcontroller), we don’t want to apply it by default.

To apply an overdrive mode / modify the clock tree with the NUCLEO, you can either modify your project after being done with STM32Cube AI Studio and execute ST Edge AI Core with the Command Line Interface, or I can create for you a set of settings to patch your current STM32Cube AI Studio installation.

 

I will still take the point since it can be interesting to give the user the ability to either use overdrive configuration or not.

 

Best regards,

Yanis

[stes11]

Hello ​@hamitiya ,

thank you for your answer. I tried to modify the project, rebuild it and flash it again. Unfortunately, this did not work as I wished it would. However, in order to be able to validate my model with STM32Cube AI Studio I would highly appreciate if you can provide the settings for running the NUCLEO in overdrive mode. I’m running version 1.2.0 of Cube AI Studio on Linux.

Thank you in advance!

Best regards,

Stefan

[hamitiya]

Hello ​@stes11,

 

First of all, to have best performance, you need to break OTP fuse to allow higher voltage for XSPI

This can be done through STM32CubeProgrammer. To do so, please execute the following:

• Open STM32CubePrg
• Connect to your NUCLEO-N657X0-Q
• Go to “Option Bytes” (OB) menu
• Scroll to HCONF1 / OTP124

• 

 

In the lines HSLV_VDDIO3 and HSLV_VDDIO2, ensure you have value “0x1”. If not, apply the following for both of them

• 1 (not 0x1, it sounds counterintuitive but 0x1 is not allowed by the GUI)

Press “Apply”

 

Afterwards, you need to “patch” a file in your STM32Cube AI Studio installation. 

• Go to </path/to/your/installation>/resources/stedgeai-desktop-backend/dependencies
• Replace “boardExternalMem.xml” with updated one in copy of this message
• Restart STM32Cube AI Studio if it was already opened

After this patch, you should see your runs with frequency up to 800MHz for the MCU, and 1GHz for ST Neural-ART Accelerator. XSPI (Memory bus) frequency raise from 50MHz to 200MHz

 

 

Best regards,

Yanis