Unable to match published Benchmarks of Jetson AGX Xavier

Autonomous Machines Jetson & Embedded Systems Jetson AGX Xavier
1

Hi,
I have executed Jetson Benchmarks in Xavier AGX Xavier devkit.
The results didn’t matched the max throughput published here
Results I got matches, 97% with Limited Latency and 84% with Max throughput

Command used:
sudo python3 benchmark.py --all --csv_file_path benchmark_csv/xavier-benchmarks.csv --model_dir $(pwd)/models --jetson_devkit xavier --gpu_freq 1377000000 --dla_freq 1395200000 --power_mode 0

MAX-N mode Published Results
Jetson 4.4.1		 % Achieved in published KPI’s
Jetson 4.5.1
Model name Devices BatchSizeGPU BatchSizeDLA Avg FPS
jetson 4.5.1		 FPS(Limited
latency)		 FPS(Max
throughput)		 FPS(Limited
latency)		 FPS(Max
throughput)
0 inception_v4 3 4 1 513.388317 528 704 97.2326358% 72.92447685%
1 vgg19_N2 1 4 0 270.036357 276 432 97.83925978% 62.50841597%
2 super_resolution_bsd 500 1 4 0 277.767551 281 302 98.84966228% 91.97601026%
3 unet-segmentation 1 2 0 236.369876 240 251 98.48744833% 94.17126534%
4 pose_estimation 1 4 0 427.127834 439 484 97.29563417% 88.24955248%
5 yolov3-tiny-416 1 16 0 1072.215892 1100 1127 97.474172% 95.13894339%
6 ResNet50_224x224 3 16 4 1807.485788 1946 2109 92.88210627% 85.7034513%
7 ssd-mobilenet-v1 3 16 2 1607.719843 1602 1919 100.3570439% 83.77904341%
Avg 97.55224531% 84.30639488%

Nvidia AGX Xavier details:

  • NVIDIA Jetson AGX Xavier [32GB]
    • Jetpack 4.5.1 [L4T 32.5.1]
    • NV Power Mode: MAXN - Type: 0
    • jetson_stats.service: active
  • Board info:
    • Type: AGX Xavier [16GB]
    • SOC Family: tegra194 - ID:25
    • Module: P2888-0001 - Board: P2822-0000
    • Code Name: galen
    • CUDA GPU architecture (ARCH_BIN): 7.2
  • Libraries:
    • CUDA: 10.2.89
    • cuDNN: 8.0.0.180
    • TensorRT: 7.1.3.0
    • Visionworks: 1.6.0.501
    • OpenCV: 4.1.1 compiled CUDA: NO
    • VPI: ii libnvvpi1 1.0.15 arm64 NVIDIA Vision Programming Interface library
    • Vulkan: 1.2.70
  • jetson-stats:
    • Version 3.1.1
    • Works on Python 3.6.9

Could you help me to reproduce the published benchmarks.
Thanks in advance,

3

Hi @sairamalavuru, the jetson-benchmarks code on GitHub is configured to replicated the limited latency scenario. We are checking with the perf team about the max throughput scenario.

1 Like