Description
Triage
- efficiency cores - not the issue as I turned off hyper threading and e cores on another 14900K from 32 vCores to 8 (turned off 16 ecores) - performance was worse
Data
64 bit long (only went up to 32 bit start and 63 bit max)
native long 64 bit = 42 sec for 13900K 128g
m0: 1410123943 p: 770 m: 7125885122794452160 ms: 632 dur: 13
native long 64 bit = 50 sec for M4 Max 48g
m0: 1410123943 p: 770 m: 7125885122794452160 ms: 866 dur: 17
native long 64 bit = 80 sec for M4 Pro 8p4e 24g
m0: 1410123943 p: 770 m: 7125885122794452160 ms: 1291 dur: 25
BigInteger:
BigInteger 64 bit = 460 sec for M4 Max 48g (9x slower)
m0: 1410123943 p: 770 m: 7125885122794452160 ms: 7312 dur: 143
BigInteger 64 bit = 80 sec for M4 Pro 8p4e 24g (9x slower)
m0: 1410123943 p: 770 m: 7125885122794452160 ms: 11710 dur: 236
BigInteger 64 bit = sec for 13900K 128g 8p16e (56x slower)
m0: 1410123943 p: 770 m: 7125885122794452160 ms: 35617 dur: 705
Test results between runs of
long
#19
https://github.com/ObrienlabsDev/performance/blob/main/cpu/virtual/multithreaded/128bit/java-benchmark-128-cli/performance-nbi/src/main/java/dev/obrienlabs/performance/nbi/Collatz.java
BigInteger
#10
https://github.com/ObrienlabsDev/performance/blob/main/cpu/virtual/multithreaded/128bit/java-benchmark-128-cli/performance-nbi/src/main/java/dev/obrienlabs/performance/nbi/CollatzBigInteger.java
CPU and RAM profiles are the same at 100% and 6g heap
Intel
Apple
Will check PCore vs Ecore distribution
Chips are
M4 Max 16 thread (12p/4e) - 75:25
13900K 32 thread (8p/16e - 16p(ht)/16e) - 50/50
- the intel chip has a 50:50 distribution as opposed to the M4 Max which has a 75:25 distribution
M1 Max (8p/2e) - 80:20
M4 Pro (8p/4e) - 66:33
M2 Pro (6p/2e) - 75:25
14900K (ht off) (8p/16e) - 33:66
check thermal throttling