[RC5] Client 2.8010 faster than later ones, x86 core #6 change ?

Frédéric Kayser f.kayser at free.fr
Sat Jun 8 00:05:41 EDT 2002


I've found that on my system (AMD Duron 963, FreeBSD 4.2), client version
2.8010 outperforms newer clients (8015 and 8018).
2.8010 has perhaps a slightly different #6 core called "RG/HB ath" whereas
it's called "RG/HB re-pair II" later.
Benchmarks show that "RG/HB re-pair II" is a very little faster than "RG/HB
ath".
In realworld terms it's somewhat different, after a few hours average speed
is 3,394,400 k/s (2.8010) compared to 3,387,100 and 3,386,600.
I've used the same .ini file, and it's repeatable.

Perhaps the new clients have introduced an overhead that globally slows the
cruncher, or the new core #6 is not really faster.


Have you noticed slowdowns between 2.8010 and later clients ?



Benchmarked speeds in keys/sec (best out of 3 runs):
                    #6          #3          #2          #4          #1
2.8010-463   3,432,997   2,980,191   2,963,697   2,821,431   2,735,029
2.8015-469   3,434,040   2,977,933   2,961,175   2,822,590   2,733,956
2.8018-472   3,433,885   2,979,382   2,978,925   2,839,458   2,740,711

Realworld speeds in keys/sec (average after a few hours):
                    #6
2.8010-463   3,394,400
2.8015-469   3,386,600
2.8018-472   3,387,100

Cores:
#0 (RG/BRF class 5)
#1 (RG class 3/4)
#2 (RG class 6)
#3 (RG re-pair I) or (RG Cx re-pair)
#4 (RG RISC-rotate I)
#5 (RG RISC-rotate II)
#6 (RG/HB re-pair II) or (RG/HB ath)
#7 (RG/BRF self-mod)
#8 (AK class 7) or (NB class 7)
#9 (jasonp P5/MMX) is #0 on v2.8010

-- 
Frédéric Kayser

--
To unsubscribe, send 'unsubscribe rc5' to majordomo at lists.distributed.net
rc5-digest subscribers replace rc5 with rc5-digest



More information about the rc5 mailing list