The essentials of my processor are like this:

CPU type 	GenuineIntel
Intel(R) Xeon(R) W-2245 CPU @ 3.90GHz [Family 6 Model 85 Stepping 7]
Number of processors 	16 [8 hyperthreaded ones]
Operating System 	Red Hat Enterprise Linux 8.4 (Ootpa) [4.18.0-305.el8.x86_64|libc 2.28 (GNU libc)]
Memory 	62.4 GB
Cache 	16896 KB

It is sometimes stated that about 4 megabytes of L3 cache are needed to run each N216 task, but it seems to me that there is a possibility that this is too pessimistic, even if the 4 megabyte number is the correct number for a single such task. The reason I say this is that there are two kinds of data going through that cache for N216 problems: program code, and program data. If that is all the machine is running, it does not require separate copies of the program code in memory since the program code might as well be in read-only memory. This assumes the cache is large enough to contain the entire working set of the program, and presumably it is. Only the program data need be duplicated.

This may be splitting hairs because I do not know the working set size of the program code (which is probably essentially the same for each instance of the program) the and the working set size of the data (that is surely different for each instance of the program). Since the start-up code does not take long, nor does the finish-up code, that part of the code can drop from the working set almost all the time and not waste L3 space.

On my machine, I allow only 8 Boinc processes at a time because otherwise I must boost the fan speeds in my machine to prevent overheating, and running more than 8 requires fan speeds so fast as to be intolerably annoying.

---

Intel has made its VTUNE Analyzer available for free with Forum support. You can get some interesting information about what the program is doing.

https://software.intel.com/content/www/us/en/develop/tools/oneapi/components/vtune-profiler.html#gs.2mf7da

---