你好，
我们使用agx orin开发套件(32G 12核版本）跑算法发现和官方宣称的有不小的差异，该开发套件模型耗时22ms，32G模组耗时也是22ms，64G版本模组耗时16.8ms
May I know how to do the test to get the result?
Have you set the system to max performance by NVPModel?
同一套程序，测的是模型的推理时间，开的模式都是0
This AGX Orin 32GB developer kit is officially described as having only 32GB memory, while sharing the same CPU/GPU configuration with the 64GB module. Therefore, it should theoretically deliver identical 275 TOPS AI computing performance. However, my actual testing shows significant deviations from the 64GB module’s performance. All testing was conducted under JetPack 5.1 with power mode set to MAXN.
Could there be undisclosed performance limitations specific to the 32GB developer kit?
Hi,
Could you share more info about the model, sample, and the command you test?
Thanks.
The model is yolo-v5-m.
Hi,
Do you use the Ultralytics sample?
Have you converted the engine into the TensorRT?
Would you mind sharing more info about your testing?
Thanks.