MAMEWorld Forums - EmuChat - 2nd Generation i5 Sandy Bridge MAME benchmarks stock / overclocked

Reading taz-nz's post about benchmarking gave me the kick in the pants to benchmark the hardware I got at the beginning of this month.

ASUS P8Z77-M LGA 1155 Intel Z77 HDMI SATA 6Gb/s USB 3.0 Micro ATX Intel Motherboard
Intel Core i5-2500K Sandy Bridge 3.3GHz (3.7GHz Turbo Boost) LGA 1155 95W Quad-Core Desktop Processor BX80623I52500K
G.SKILL Ripjaws Series 8GB (2 x 4GB) 240-Pin DDR3 SDRAM DDR3 1600 (PC3 12800) Desktop Memory Model F3-12800CL9D-8GBRL

I used the official MAME 64-bit binaries from the MAMEDev.org site...

http://www.mamedev.org/downloader.php?file=releases/mame0141b_64bit.exe
http://www.mamedev.org/downloader.php?file=releases/mame0145b_64bit.exe
http://www.mamedev.org/downloader.php?file=releases/mame0148b_64bit.exe

I ran each benchmark multiple times, and averaged the results. Between each run, I deleted the cfg, snap, diff, and nvram directories, as well as any other temporary log files that were generated. The only exception to the nvram deletions were gradius4 and scud as they need them to get in-game.

One curious observation... The two games that created "diff" files (blitz and gauntleg) seemed to get lower results on the first run in a series of benchmark runs, even though I was deleting the diff before every run. I suspect some sort of hard drive or windows caching under the covers causes this behavior because I could "prime the pump" by manually running blitz and gauntleg just before running my benchmark script and then the first run wouldn't be markedly different from the rest.

I did the benchmarks first with stock CPU and memory settings. Then I ran the ASUS AI Tuner and let it auto-overclock the CPU. I plan to do some manual overclocking at some point, and if/when I do, I'll come back and add an additional set of benchmarks. My main concern with the auto overclocking is that I don't believe it is actually overclocking the memory, which should make a measurable difference.

I'd also like to try building a "Core AVX" optimized binary to see how much different the results are.
http://gcc.gnu.org/gcc-4.6/changes.html

Stock Intel Core i5-2500K Sandy Bridge 3.3GHz (3.7GHz Turbo Boost)

           0.141   0.145   0.148
blitz     219.84  219.37  225.14
dolphin    27.73   26.27   25.61
gauntleg  298.94  295.72  282.48
gradius4  164.59  169.88  160.71
propcycl  183.15  155.46  162.75
radikalb  149.28  143.18  153.02
scud      106.43  108.95  107.48
starsldr   61.62   87.89   96.59

Intel Core i5-2500K Sandy Bridge overclocked to 4.28GHz

           0.141   0.145   0.148
blitz     274.11  274.10  280.41
dolphin    34.10   32.45   31.67
gauntleg  372.93  369.93  352.33
gradius4  206.64  213.24  201.79
propcycl  227.80  192.60  201.67
radikalb  187.15  179.62  191.77
scud      132.35  135.29  136.12
starsldr   76.63  109.29  121.05

Intel Core i5-2500K Sandy Bridge overclocked to 4.8GHz
           0.141
blitz     300.30
dolphin    38.25
gauntleg  419.94
gradius4  222.66
propcycl  255.11
radikalb  202.51
scud      147.37
starsldr   86.08

GroovyMAME support forum on BYOAC

Edited by krick (06/03/13 05:49 AM)