Home

Welcome to the OMPEval wiki!

Performance comparison of 7 card poker hand evaluators

The following tests were run on Intel 3770k @4.28Ghz using a single thread. Results are in millions of evaluations per second. Three different benchmarks were run:

Seq: Sequential evaluation (aka enumeration) performance.

Rand: Evaluation from a pregenerated array of random hands (7 x uint8_t per hand).

Rand2: Evaluation from an array of random Hand objects.

	TDM-GCC 5.1 64bit (-O3 -msse4.1)			TDM-GCC 5.1 32bit (-O3 -msse4.1)			MSVC2013 64bit (/O2 /GS- /Oy /GL /arch:AVX)
	Seq	Rand	Rand2	Seq	Rand	Rand2	Seq	Rand	Rand2
OMPEval	775	272	520	724	234	471	691	262	529
OMPEval7 (*)	776	284	594	725	241	505	692	271	597
SKPokerEval	223	146		134	99		204	148
2+2 Evaluator	1588	19	(**)	1122	19		1544	19
Plexiq	290	156	232	43	38	46	183	131	174
ACE (decompr.)	80	43	87	75	38	62	69	39	72
HoldemShowdown	93	41	68	(***)			81	42	63
PokerSource	134	77	103	(***)			118	79	102
PokerStove	58	27	70	50	21	57	92	69	87
Senzee 7 card	31	21	32	29	13	23	24	20	27

(*) Compiled with lookup table that has no support for hands smaller than 7 cards.

(**) For 2+2 evaluator this benchmark is a bit meaningless because after 7 cards the partial "Hand object" becomes the final rank, so there's no work to be done.

(***) It is probably possible to build these but there are some linking problems with TDMGCC5.1.

OMPEval performance with different compiler settings

	Seq	Rand	Rand2
TDM-GCC 5.1 64bit (-O3 -msse4.1)	780	269	544
TDM-GCC 5.1 64bit (-O3)	771	237	392
TDM-GCC 5.1 64bit (-O3 -D OMP_SSE2=0)	781	199	567
TDM-GCC 5.1 32bit (-O3 -msse4.1)	724	234	471
TDM-GCC 5.1 32bit (-O3 -msse2)	647	192	343
TDM-GCC 5.1 32bit (-O3 -D OMP_SSE2=0)	498	91	482
MSVC2013 64bit (/O2 /GS- /Oy /GL /arch:AVX)	648	257	525
MSVC2013 32bit (/O2 /GS- /Oy /GL /arch:AVX)	368	169	445
MSVC2013 32bit (/O2 /GS- /Oy /GL /arch:SSE2)	320	174	381
MSVC2013 32bit (/O2 /GS- /Oy /GL /arch:IA32)	235	107	422