How the GPU utilization of Jetson TX2 is measured?

Autonomous Machines Jetson & Embedded Systems Jetson TX2
1

Hello,

I have Jetson TX2, which has 256 GPU cores.
I am curious about how the GPU utilization of Jetson TX2 is measured.

The Jetson TX2 board has 6 CPU cores and has 6 CPU uilization values.
However, the Jetson TX2 board has 256 GPU cores, but only has one GPU uilization value.

Can you explain me how the GPU utilization of Jetson TX2 is measured or the exact definition of GPU utilization on TX2?

2

Have a reference to below link.

https://tegra-sw-opengrok.nvidia.com/source/xref/rel-32_l4t/core/utils/tegrastats/elements.c#39
https://tegra-sw-opengrok.nvidia.com/source/xref/rel-32_l4t/kernel/nvgpu/drivers/gpu/nvgpu/os/linux/sysfs.c#379

3

Hi ShaneCCC,

I can not open the links.

Can you check the links you gave?

4

Looks like the uilization is read from below node.

char *igpu_load_nodes = {
“/sys/devices/17000000.gp10b/load”, // t186 4.4/4.9
“/sys/devices/18000000.vgpu/load”, // virtualized config
“/sys/devices/17810000.vgpu/load”, // virtualized t194 config
“/sys/devices/platform/gpu.0/load”, // t210 3.1
“/sys/devices/gpu.0/load”, // t210 4.4
“/dev/gpu_load”, // t186/t194 qnx
};

5

Thank you for your answer.

However, what I want to know is:

GPU utilization of TX2 is:

  1. Maximum utilization of 256 GPU cores
  2. Average utilization of 256 GPU cores
  3. Other way, etc.

I am curious about that information because I measured GPU utilization running CNN inference, then found that GPU utilization is too higher than my expectations.

6

Hello,

I am also interested in what GPU load actually means on the TX2.

Thank you

7

It seems load is in thousands in /load file.

So 360 => 36%

openat(AT_FDCWD, "/sys/devices/17000000.gp10b/load", O_RDONLY) = 3
fstat(3, {st_mode=S_IFREG|0444, st_size=4096, ...}) = 0
read(3, "<b>360</b>\n", 4096)                  = 4
close(3)                                = 0
fstat(1, {st_mode=S_IFREG|0644, st_size=69183, ...}) = 0
write(1, "RAM 4440/7851MB (lfb 357x4MB) cp"..., 115RAM 4440/7851MB (lfb 357x4MB) cpu [0%@1266,0%@652,0%@652,0%@1266,0%@1266,0%@1268] EMC 4%@1866 APE 150 GR3D <b>36%</b>@420