Jetson Xavier NX benchmark score

[lisoulin]

HI
Can we get bechmark score on Jetson NX, such as ResNet-50 ,Inception V4 , VGG-19?
It supposed to be a table for these data.
Like this…

Jetson Benchmarks

It’s better if we also get the procedure , like this…

Deep Learning Inference Benchmarking Instructions

Hi all, below you will find the procedures to run the Jetson Nano deep learning inferencing benchmarks from this blog post with TensorRT. note: for updated JetPack 4.4 benchmarks, please use github.com/NVIDIA-AI-IOT/jetson_benchmarks [image] While using one of the recommended power supplies, make sure you Nano is in 10W performance mode (which is the default mode): $ sudo nvpmodel -m 0 $ sudo jetson_clocks Using other lower-capacity power supplies may lead to system instabilities or shutd…

[dusty_nv]

@lisoulin

Can we get bechmark score on Jetson NX, such as ResNet-50 ,Inception V4 , VGG-19?

Hi @lisoulin, please see here for preliminary benchmarks of Jetson Xavier NX: https://devblogs.nvidia.com/jetson-xavier-nx-the-worlds-smallest-ai-supercomputer/

[lisoulin]

Thanks for the reply.
Actually, I have seen this link which you post.

From the chart, the score for ResNet-50 of Jetson NX is over 1250 FPS,
And I also tried to run on my Jetson NX to find out best score , but unfortuantely my fine tune is about 850 FPS.
It has real gap with official data.

And I also find an interesting thing, my Jetson is 15W. And from the data https://developer.nvidia.com/embedded/jetson-agx-xavier-dl-inference-benchmarks

If the AGX is in 15W mode, the score of ResNet-50 is near to 850, it simliar with my result on NX 15W mode.
So could you help provide the test procedure that NX can reach over 1250 FPS on ResNet-50?
My command is “./trtexec --output=prob --deploy=…/data/googlenet/ResNet50_224x224.prototxt --int8 --batch=128”

[lisoulin]

[lisoulin]

[dusty_nv]

@lisoulin

My command is “./trtexec --output=prob --deploy=…/data/googlenet/ResNet50_224x224.prototxt --int8 --batch=128”

That is only using GPU, whereas the benchmark results are for GPU + 2xDLA. You can run it with this GitHub repo: GitHub - NVIDIA-AI-IOT/jetson_benchmarks: Jetson Benchmark

[lisoulin]

Thanks for the guideline
I follow the website procedure to run individual benchmark
But I still got the same level score about 800~850.

Interesting , the sameple output on this web sidte GitHub - NVIDIA-AI-IOT/jetson_benchmarks: Jetson Benchmark
ResNet50 score is also locate at 800~850.
Any idea?

[lisoulin]

[lisoulin]

[dusty_nv]

Hi @lisoulin, yes that is close to expected performance on JetPack 4.4 DP.

[lisoulin]

Hi Dusty:
Do you mean Resnet-50 score at 800-850 is expected result?
My question is the chart from the link, Resnet-50 of Jetson NX is over 1250 FPS.
Is it any mistake ? Because it’s chart ,instead of number, and looks like over 1250FPS, is it something wrong?
https://devblogs.nvidia.com/jetson-xavier-nx-the-worlds-smallest-ai-supercomputer/

[dusty_nv]

You can find the latest results from today’s blog:

https://devblogs.nvidia.com/bringing-cloud-native-agility-to-edge-ai-with-jetson-xavier-nx/

The preliminary results that were posted previously have been updated with these. There was a change in the benchmarking methodology in the GPU + 2xDLA case that allowed more time for the DLA’s to load the networks and begin running, so we have updated the performance here. The scripts on the jetson_benchmarks repo on GitHub also reflect these updates going forward.

[lisoulin]

Thank you very much!
My question is resolved.