Performance Summary

QCQuad-Core Solver
S2Solver2
R5Ryzen 5 (spec below)
R7Ryzen 7
THPTransparent Hugepages


Optimal Solves
1000 Random Cubes
Solver Plat
form
THP Dist4
(GB)
Average
Solve
Longest
Solve
Percent
Change
Log
File
QCR5N1.814.62142.7741.67view
QCR5Y1.810.32101.0229.41view
QCR7N1.84.5344.2114.97view
QCR7Y1.83.9438.2813.02view
S2R5N1.813.26123.1943.51view
S2R5Y1.89.2483.7630.32view
S2R5N3.610.0295.0848.89view
S2R5Y3.66.7362.2932.83view
S2R7N1.84.0436.7313.17view
S2R7Y1.83.5731.7311.63view
S2R7N3.63.1227.9818.18view
S2R7Y3.62.6423.5115.38view
S2R7N7.12.4922.3628.35view
S2R7Y7.11.9417.8022.09view

The test platforms are both HP Pavilion desktop computers with processors from Advanced Micro Devices (AMD). All tests were done using Google Chrome on the Fedora operating system. The hardware specs for the two platforms are:

Ryzen 5: 3400G, 3.7 Ghz 4-Core/8-Thread, 16 GB RAM (DDR4 2667 MHz, dual-channel)

Ryzen 7: 5700G, 3.8 Ghz 8-Core/16-Thread, 64 GB RAM (DDR4 3200 MHz, dual-channel)

CPU cache sizes are here, RAM timings are here and here.




Notes:
A frequency distribution for the 1000 random cubes tested is here.

Url parameter "conc" was used as shown:
Ryzen 5: conc=9
Ryzen 7: conc=18
This allows the 18 cube moves to be distributed evenly across either 9 or 18 processes for better load balancing which yields a small performance gain.

New parameter use_stl=0 was used to eliminate use of the Date.now function in the main search loop. This was done because Date.now, which is a Javascript built-in function, has recently become significantly slower on some platforms. This eliminates the search time limit which is not needed for these tests.

The 3.6 GB Dist4 is the largest that Chrome can load on the Ryzen 5, likewise for the 7.1 GB Dist4 on Ryzen 7.

A table showing the average & longest solve for each solution length has been appended to the log for each test.

Some additional information about transparent hugepages as it relates to these tests is here. For each test with hugepages, a listing of allocations for the search arrays has been appended to the log.

The previous version of this performance summary which has some additional tests is here.