Off topic discussion • Re: A bit of Raspberry with 144C/288T

I stopped the code you asked me to execute after 2h(!) after all entries for cores 0-17 have been computed.
I did not want to run for 7x2 more hours to complete.
I hope you can find what you wanted to see in attached output.

Thank you! I see core-to-core latency within a socket that looks similar to what I get for a Xeon E5-2620v3. However, the socket to socket latency is about 43 percent slower on the eight-socket Xeon compared to the dual socket. That's not too surprising and, in my opinion, pretty good.

Some published timings appear at

https://chipsandcheese.com/p/core-to-co ... ge-systems

Note the entries in the published tables are obtained from those which appear during the test divided by two. Thus, the linein your output is like having 40.2 in the 0,1 slot of the table. I don't know why the values appearing in the table as well as the ones they talk about have been divided by two. Maybe this is traditional.

Do you have to remove the compute books to deactivate them? Is that mostly to save electricity?

Statistics: Posted by ejolson — Sat Aug 16, 2025 3:32 am

---