Hi Brian,
Thanks for trying a few more things. Let me make sure that I understand correctly. You boot the board and run the notebook and find that the overall acceleration shown is low. However, after running the notebook a few times, the execution times become unpredictable. Is that correct?
In the post I made yesterday, I ran the notebook immediately after the board booted, so I’m afraid that I’m not able to recreate the same problem you are facing on my Z2.
I wanted to check the performance variation you were seeing, so I made a script to run the code from the notebook multiple times on both v2.7 and v3.0.1 and gathered the results. Each script ran the code 50 times, measuring the hw execution time and sw execution time from a freshly booted board. In each case, Jupyter lab was open with two terminals running, one to launch the experiment script and one to monitor ram usage with htop.
Below is a screenshot of the summary results; however, I have attached an excel sheet with the full results. I’m also happy to share my experiment scripts if you wish:
The results show some increase in variation in the execution time between v2.7 and v3.0.1. However, to a much lesser extent than you were seeing. And again, I wonder if this is the best benchmark to be used as the data-transfer size is relatively small and will be sensitive to lots of noise from the OS scheduler.
One thing we have noticed with v3.0.1 is that the RAM usage is higher than in previous versions. If we have many notebooks open, we can use up memory resources, and things can become quite sluggish. Could this explain the significant variation you are seeing in execution time?
All the best,
Shane
- ps: Sorry I had to upload the results as .txt files but they are actually .csv as I was not permitted to upload excel worksbooks or csv files.
summary_results.csv.txt (434 Bytes)
v2.7_res.csv.txt (2.1 KB)
v3.0.1_res.csv.txt (1.2 KB)