I have four invalids on my Ryzen 3600 running Ubuntu 20.04.2.
https://milkyway.cs.rpi.edu/milkyway/results.php?hostid=892961&offset=0&show_names=0&state=5&appid=

That is not a large number, and I am not concerned about the amount. But they all validated against other Windows machines.

I also have a number on my Ubuntu machine that validated against Windows machines, so it is not exactly a bug. Maybe it is just a preference?

I should note that I am running the 1.80 N-Body using one core per work unit, as limited by an app_config.xml file.
All available cores are running, reserving one or two cores to support a GPU.

On the Linux machine I now have 19 invalids, and all have been validated by two Windows machines.
So I think it is clear by now that Windows machines prefer to validate against each other.

Also, I can compare Windows v. Linux on two Ryzen 3600s.
On the Linux (Ubuntu 20.04.2) machine, I see a 9% invalid rate, and a 21% error rate.
On the Win10 machine, I see a 0% invalid rate and also a 21% error rate.

So I will run them on Windows.

This is common on projects with multiple app platforms. The applications produce slightly different results and the validators don't have sufficient tolerance settings to allow validating different platforms against one another.

---

Yes, that is so. I was just a little surprised that it did not occur the other way, but now I see that it has.
That is, when I get an invalid on the Windows machine, they are validated on two other Linux machines.
https://milkyway.cs.rpi.edu/milkyway/results.php?hostid=880752&offset=0&show_names=0&state=5&appid=

So it works both ways. But given the preponderance of Windows PCs, I am a little better off running it that way and getting fewer invalids.
The speed seems to be about the same on either, though I have not done really long-term tests to find out.

Thanks for your input.