MAMEWorld Forums - EmuChat

Methodology:

Copied mame64.exe used from previous benchmark to "intel.exe" and "nvidia.exe" then set the Optimus profile for each exe accordingly in the nVidia control panel.

The benchmark consists of a batch file that does the following for each game on the list:

nvidia.exe -noautoframeskip -frameskip 0 -seconds_to_run 90 -nothrottle -nosleep -video d3d -skip_gameinfo -effect none -nowaitvsync gamename intel.exe -noautoframeskip -frameskip 0 -seconds_to_run 90 -nothrottle -nosleep -video d3d -skip_gameinfo -effect none -nowaitvsync gamename pause

Results are below:

gradius4 Gradius 4: Fukkatsu
nVidia: 85.19%
Intel: 85.19%

slrasslt Solar Assault (ver UAA)
nVidia: 143.92%
Intel: 144.86%

sf2ceua Street Fighter II': Champion Edition (USA 920313)
nVidia: 1165.62%
Intel: 956.05%

pacman Pac-Man (Midway)
nVidia: 9672.65%
Intel: 8343.55%

kinst Killer Instinct (v1.5d)
nVidia: 270.46%
Intel: 485.21%

hsf2 Hyper Street Fighter 2: The Anniversary Edition (USA 040202)
nVidia: 1119.29%
Intel: 1148.92%

salmndr2 Salamander 2 (ver JAA)
nVidia: 188.83%
Intel: 191.01%

mk Mortal Kombat (rev 5.0 T-Unit 03/19/93)
nVidia: 757.54%
Intel: 814.83%

mkla3 Mortal Kombat (rev 3.0 08/31/92)
nVidia: 852.34%
Intel: 858.26%

Result summary:

Optimus doesn't seem to affect performance at all. Every result seemed to be within the margin of error of running the benchmark five times under the same config; results were never consistent across multiple runs outside of being within a certain spread. Older games that would naturally run faster have wider spread, of course.

I only remembered a key point after starting the benchmarks: Optimus works by dumping the nVidia framebuffer to the Intel one for output. Therefore there's actually a very slight bottleneck that wouldn't be there if the nVidia was outputting to its own port.

For normal 3D apps, Optimus would give a performance boost by offsetting the bottleneck with just how much faster the nVidia chip actually is. MAME doesn't do anything more than two textured triangles under the most basic of circumstances and wouldn't be a great use of Optimus.

All of this is pretty obvious in hindsight, but it never hurts to have it all down on paper.

Aside:
I ran MK and MKLA3 also for a comparison of how much harder T-Unit was going to be on the rig compared to the older hardware. Wasn't expecting it to be _unplayable_ on T-Unit, just was curious at the difference in numbers