DLA usage percentage and frequency reporting in tegrastats tool

[chiauee85]

Hi,

I run the tegrastats tool to collect performance statistics while my deepstream gst pipeline is running on Jetson AGX Orin . The pipeline has the inferencing enabled to run with DLA. When i checked the the tegrastats output, i noticed it has metric about DLA (as below). The metric mentioned DLA frequency but it seems to me that the DLA usage percentage is missing before the “@” symbol. Is this expected? i’m looking for way to check the usage percentage of DLA similar to the GPU metric reported by tegrastats tool.

“NVDLA0_FREQ @1536”

snippet from my tegrastats log:
06-01-2022 16:33:06 RAM 6327/30623MB (lfb 5561x4MB) SWAP 0/15311MB (cached 0MB) CPU [11%@2199,5%@2200,3%@2198,11%@2199,3%@2198,1%@2195,0%@2198,1%@2199,1%@2198,1%@2199,1%@2102,3%@2197] EMC_FREQ 28%@3199 GR3D_FREQ 22%@1300 GR3D2_FREQ 22%@1300 NVENC 998 NVDEC 998 NVJPG1 729 VIC_FREQ 39%@729 NVDLA0_FREQ @1536 APE 233 CV0@56.375C CPU@58.937C Tdiode@47C SOC2@55.218C SOC0@56.312C CV1@55.593C GPU@53.875C SOC1@55.187C CV2@51.875C VDD_GPU_SOC 13178mW/13178mW VDD_CPU_CV 3993mW/3993mW VIN_SYS_5V0 7975mW/7975mW NC 0mW/0mW VDDQ_VDD2_1V8AO 2120mW/2120mW NC 0mW/0mW

[AastaLLL]

Hi,

Which JetPack version do you use?
We don’t get the NVDLA0_FREQ from tegrastat on Orin with JetPack 5.0.1.

$ sudo tegrastats
06-02-2022 09:40:53 RAM 2872/30623MB (lfb 5315x4MB) SWAP 0/15311MB (cached 0MB) CPU [0%@729,0%@730,1%@718,0%@731,0%@730,0%@727,0%@723,0%@729,0%@729,0%@728,0%@730,0%@751] EMC_FREQ 0%@204 GR3D_FREQ 0%@114 GR3D2_FREQ 0%@114 NVJPG1 729 VIC_FREQ 729 APE 233 CV0@-256C CPU@49.468C Tdiode@39.75C SOC2@46.031C SOC0@46.781C CV1@-256C GPU@44.781C SOC1@45.5C CV2@-256C VDD_GPU_SOC 4592mW/4592mW VDD_CPU_CV 765mW/765mW VIN_SYS_5V0 3931mW/3931mW NC 0mW/0mW VDDQ_VDD2_1V8AO 704mW/704mW NC 0mW/0mW

Thanks.

[Splendor027]

As far as I realized, the power consumption of DLA is reported at the end of the tegrastats, which corresponds to NC 0mW/0mW in your output.

I suggest you run tegrastats just a few seconds before running the DLA part in your execution so that you can see the difference in the power consumption.

[chiauee85]

Yeap…i’m using JetPack 5.0.1 DP on my AGX Orin 32GB HW.

Attached are my main DS config file and gie config file for your reference.
deepstream-main-config-DLA.txt (1.7 KB)
config_infer_primary_gie-dla.txt (1.0 KB)

i enabled the following property in my gie config file.
enable-dla=1
use-dla-core=0
Also, i used deepstream-app (with Yolo custom plugin, weight and config from GitHub - marcoslucianops/DeepStream-Yolo: NVIDIA DeepStream SDK 6.1 / 6.0.1 / 6.0 configuration for YOLO models) to generate the TRT engine model for my Yolo v3 model to be run with DLA and then set the “model-engine-file” property in the gie config file to point to the generated engine mode.

[AastaLLL]

Hi,

You can get the DLA status with the following command (active/suspended).
Does this meet your requirement?

$ watch -n 1 "cat /sys/devices/platform/host1x/15880000.nvdla0/power/runtime_status"
$ watch -n 1 "cat /sys/devices/platform/host1x/158c0000.nvdla1/power/runtime_status"

Thanks.

[chiauee85]

That’s not what we are looking for. We are looking for utilization percentage reporting for DLA like how tegrastats reports the GPU utilization (eg: GR3D 22%@1300).
So does that mean the NVDLA0_FREQ reported by tegrastat which i have observed earlier on is unexpected and DLA metric reporting is not officially supported by tegrastat?

[AastaLLL]

Hi,

DLA only has the active/suspended status value.
Thanks.