Benchmarks

General discussion on MAME, MARP, or whatever else that doesn't belong in any of the other forums

Moderators: mahlemiut, seymour, QRS

User avatar
mahlemiut
Editor
Posts: 4117
Joined: Mon Feb 04, 2002 10:05 pm
Location: New Zealand
Contact:

Re: Benchmarks

Post by mahlemiut » Sat Apr 29, 2017 4:26 am

Well, here we go. All I'll say is, MAME seems slower - especially with older drivers that don't need huge power to run.

System:
3.7GHz Core i3-6100
Intel HD Graphics 530
8GB RAM
Arch Linux (kernel 4.10.11)
WolfMAME 0.185 (SDL, 64-bit self-compiled)

Commandline options are the same as before. All values are speed percentage.

Code: Select all

pacman      556.96
malzak      458.32
ddragon     425.57
invaders    453.07
tbyahhoo    270.49
stunrun     349.78
mrdrillr    204.53
cotton2     239.73
beastrzr    285.20
puzzloop    275.18
sfiiin      444.87
strider     498.24
rsgun       223.62
Will try with -video bgfx later, which offloads blitting to a separate thread, like the old -mt option did (only in a way that actually works properly) EDIT: scratch that idea, BGFX refuses to run faster than 60fps for some reason, so Pacman won't run faster than 99%)
- Barry Rodewald
MARP Assistant Web Maintainer
Image

User avatar
LordGaz
MARP Knight
MARP Knight
Posts: 219
Joined: Sat Mar 09, 2002 5:07 pm
Location: London, UK

Re: Benchmarks

Post by LordGaz » Sat Apr 29, 2017 4:22 pm

Thanks for that, Moore's law is at an end. We're still getting faster but at a slower and slower rate.

User avatar
mahlemiut
Editor
Posts: 4117
Joined: Mon Feb 04, 2002 10:05 pm
Location: New Zealand
Contact:

Re: Benchmarks

Post by mahlemiut » Sat Apr 29, 2017 6:51 pm

Will pull 0.126 from the old box eventually, and hope that it'll run (it should). Might give a better idea as to how much speed is lost simply through MAME shifting from C to C++14 and various emulation improvements over the past 9 years.
- Barry Rodewald
MARP Assistant Web Maintainer
Image

User avatar
mahlemiut
Editor
Posts: 4117
Joined: Mon Feb 04, 2002 10:05 pm
Location: New Zealand
Contact:

Re: Benchmarks

Post by mahlemiut » Sun Apr 30, 2017 4:43 am

Grabbed a 64-bit SDL WolfMAME 0.127, and oddly enough, I still get 580% on pacman. Modern OSes use up this much more CPU time? Seems weird.

EDIT: Ok, adding -nomaximize makes it run much faster. Intel must have crappy texture stretching or something.
- Barry Rodewald
MARP Assistant Web Maintainer
Image

User avatar
mahlemiut
Editor
Posts: 4117
Joined: Mon Feb 04, 2002 10:05 pm
Location: New Zealand
Contact:

Re: Benchmarks

Post by mahlemiut » Sun Apr 30, 2017 4:57 am

Updated to use -window -nomaximize (removes some video hardware slowness). Still 64-bit 0.185.

Code: Select all

pacman     3999.63
malzak      541.17
ddragon     962.72
invaders   2621.93
tbyahhoo    337.64
stunrun     556.88
mrdrillr    228.16
cotton2     302.80
beastrzr    331.38
puzzloop    350.09
sfiiin      919.39
strider    1357.95
rsgun       267.01
- Barry Rodewald
MARP Assistant Web Maintainer
Image

User avatar
LordGaz
MARP Knight
MARP Knight
Posts: 219
Joined: Sat Mar 09, 2002 5:07 pm
Location: London, UK

Re: Benchmarks

Post by LordGaz » Sat Jan 27, 2018 10:56 am

It's that time again. Got meself an i5-8600K and naturally the first thing one does with a new system is benchmark it with MAME!

Mame now has a -bench switch so I decided to use that. From the MAME docs this is what -bench does.

Code: Select all

-bench [n]

    Benchmark for [n] number of emulated seconds; implies the command string:

    -str [n] -video none -sound none -nothrottle
So for example you can benchmark pacman for 60 ingame seconds in the command window of your choice with:

Code: Select all

.\mame64.exe -bench 60 pacman
Results using -bench 60.

Code: Select all

pacman     15122%
malzak       717%
ddragon     1490%
invaders    8735%
tbyahhoo     411%
stunrun      848%
mrdrillr     281%
tekken       577%
cotton2      385%
beastrzr     450%
puzzloop     450%
sfiiin      1637%
galaga      5681%
strider     2866%
rsgun       355%
Obviously you can't relate these results exactly to previous because it doesn't test video and I decided to run each for 60 secs so they go deeper into the game but at least we can benchmark on the same playing field.

Post Reply