TEST Cuda pcl，the result semms error

[tust1314zhou]

Hi, I test the Cuda pcl(icp and ndt)，
when I test, the cuda ndt result as follow:

It seems cpu faster than gpu， So I want to konw is there any problem here?

[tust1314zhou]

this is my computer:

[AastaLLL]

Hi,

Could you share which source and the testing command with us?
We want to give it a try to get more info.

Thanks.

[tust1314zhou]

cuPCL/cuNDT at main · NVIDIA-AI-IOT/cuPCL · GitHub , this is the source code。“I am testing the code according to the instructions on GitHub.”

[AastaLLL]

Hi,

Have you maximized the device’s performance?
This can be done via the below command:

$ sudo nvpmodel -m 0
$ sudo jetson_clocks

Thanks.

[tust1314zhou]

ok， I try it and then try you the result

[tust1314zhou]

Hi , I run the
$ sudo nvpmodel -m 0
$ sudo jetson_clocks，

the result still seems not right.
first is cuda ndt：

[tust1314zhou]

second is cudaicp,

[AastaLLL]

Hi,

We can reproduce a similar behavior.
Need to check with our internal team first. Will update more with you later.

Thanks.

[tust1314zhou]

Hello，Is there any new progress here？

[AastaLLL]

Hi,

Please try to adjust the parameter to make the fitness score of OSS-PCL is better (lower) than or equal to that of cuPCL.
We can get a better performance on GPU when the fitness_scores are close.

...
Loaded 7000 data points for Q with the following fields: x y z
Target rigid transformation : cloud_P -> cloud_Q
Rotation matrix :
    | 0.923880 -0.382683 0.000000 |
R = | 0.382683 0.923880 0.000000 |
    | 0.000000 0.000000 1.000000 |
Translation vector :
t = < 0.000000, 0.000000, 0.200000 >------------checking PCL NDT(CPU)----------------
PCL align Time: 79.5729 ms.
Normal Distributions Transform has converged: 1 score: 0.541405
Rotation matrix :
    | 0.999255 0.009666 0.037361 |
R = | -0.008477 0.999457 -0.031832 |
    | -0.037648 0.031492 0.998795 |
Translation vector :
t = < 0.037921, 0.105031, 0.180639 >------------checking CUDA NDT(GPU)----------------
CUDA NDT by Time: 73.6361 ms.
CUDA NDT fitness_score: 0.538531
Rotation matrix :
    | 0.999171 0.010360 0.039369 |
R = | -0.009026 0.999384 -0.033907 |
    | -0.039696 0.033523 0.998649 |
Translation vector :
t = < 0.056860, 0.143136, 0.188665 >

Thanks.

[tust1314zhou]

Why is this different from the results on github, which show that cuda gpu versions are much faster and score less