Skip to content

[R-pkg-devel] Issue "Re-building vignettes had CPU time 4.1 times elapsed time"

2 messages · Nicholas Tierney, Ivan Krylov

#
Hi Ivan,

Thank you for investigating this so thoroughly! My apologies for not
replying sooner.

I think it is a good practice to assume up to 2 cores for most users.
Unfortunately I am unable to trigger this NOTE locally, even running:

```
_R_CHECK_VIGNETTE_TIMING_CPU_TO_ELAPSED_THRESHOLD_=1.1 \
 R CMD check --timings greta_0.5.1.tar.gz
```

I have tried setting default chains and cpu to be 2, but it seems, based on
your evidence perhaps that there is something inside tensorflow, or in
reticulate that might be triggering this. I will reach out to the
developers of tensorflow and reticulate to query this.

Thank you so much again for your help!

Nick
On Fri, 30 Jan 2026 at 01:26, Ivan Krylov <ikrylov at disroot.org> wrote:

            

  
  
#
? Wed, 11 Feb 2026 10:15:39 +1100
Nicholas Tierney <nicholas.tierney at gmail.com> ?????:
What's your operating system? How many logical processors does it have?
Could it be offloading Tensorflow operations to a GPU? This check
doesn't work on Windows, but should be very sensitive otherwise.
Sorry, my mistake. Setting the number of chains to 2 instead of 4
seemed to lower the average CPU load closer to 2 on my computer, but
the instantaneous CPU load still spikes to around 400%. This is
definitely happening in the threads created by Tensorflow, see the
attached plot of approximate CPU load versus DLL name and process
runtime.