That should not happen, I switched to using "std::mt19937" because "random()" on an Apple computer of a friend did work differently than "random()" on Intel Ubuntu (which did same as Pi5).Without computing D it takes 86 seconds total for the initial greedy pass on a similar vintage Xeon E5-2620v3 using my own C code. The run leads to a slightly different total path length, but hopefully that's due to the random number generator.Not precomputing D might also save memory so the computation can be performed on a Pi.
I will definitely look into that when you posted.For comparison the C++ code downloaded from the above link running on the same system takes 52 seconds to create D and 921 seconds for the greedy pass. The memory on that computer is running at 1866 MT/s, which in my opinion caused too much meowing from all three of the kittens.
I'll clean up my C code before posting it. I wonder if it will run on the Pi 4.
I will take whatever solution is fastest, until now I thought that would be distance matrix ...
I completed putting identical 16GB ram modules into my 8-socket server, now with 8×4×16GB=512GB:
Code:
hermann@x3950-X6:~$ sudo dmidecode -t memory | grep Manufacturer | uniq -c 32 Manufacturer: Samsunghermann@x3950-X6:~$ sudo dmidecode -t memory | grep "Part Number" | uniq -c 32 Part Number: M393A2G40DB0-CPB hermann@x3950-X6:~$ sudo dmidecode -t memory | grep "Size: [0-9]" | uniq -c 32 Size: 16 GBhermann@x3950-X6:~$ free -h total used free shared buff/cache availableMem: 503Gi 8.5Gi 494Gi 44Mi 3.2Gi 495GiSwap: 8.0Gi 0B 8.0Gihermann@x3950-X6:~$ The I did run passmark test again, and CPUmark raised from 75,582 before to now 80,593(!);
https://www.passmark.com/baselines/V11/ ... 0191142032
You know the sound of such servers.
I created new short "Jet sound in basement, 8-socket server starting
" for those who do not:https://www.youtube.com/shorts/KBpVzTYbr-g
Statistics: Posted by HermannSW — Thu Aug 28, 2025 4:36 pm