MAMEWorld Forums - News - Re: MAME 0.176 on Raspberry Pi benchmarks

> > > Well it's interesting to see somebody has done this, however, benchmarks with the
> > > video and sound disabled aren't really representative of a real world use case...
> > >
> > > > ChoccyHobNob has released some RasPi benchmarks today with current MAME,
> showing
> > > the
> > > > performance increases across the generations of Raspberry Pi's. Be sure to
> check
> > it
> > > > out!
> > > >
> > > > Says ChoccyHobNob:
> > > > There are a lot of accusations about MAME being slow, about how it is a
> > > documentation
> > > > project, not an exercise in performance tuning. Most people seem to think that
> a
> > 20
> > > > year old version of MAME is the best you can hope for out of a Raspberry Pi.
> I’m
> > > here
> > > > to say that’s not the case. If you set your expectations at a reasonable level,
> > you
> > > > tweak your settings in a sensible way, and you overclock wherever you can, you
> > will
> > > > probably be pleasantly surprised.
> > > > ...
> > > > MAME benchmarks are run without using the display or sound devices, this keeps
> > the
> > > > benchmarks purely about MAME’s performance, and tries to take any bottlenecks
> > with
> > > > the host system out of the loop. In the case of the Raspberry Pi, the GPU gives
> > the
> > > > performance a pretty good beating, meaning that games need to run quite a bit
> > > faster
> > > > than 100% to be playable.
> > > >
> > > > http://choccyhobnob.com/articles/benchmarks-for-mame-on-raspberry-pi/
> >
> > Sound being disabled has no bearing on emulation speed - it is still emulated. The
> > video is the same, however, may the effect may range from a small bit to a large
> > amount depending on how complex video is. The benchmarks are largely to gauge CPU
> > worthiness anyway, I believe.
>
> Yes, I'm aware of that.
>
> What I'm saying is that (and the post even mentions it) is that the simply process of
> outputting the video to the screen is very costly on the Pi and drags performance
> down quite a lot.
>
> In terms of real world benchmarks the only thing anybody is going to care about is
> the end result, and, even with the warning in place, it's likely these benchmarks
> will end up making the hardware look more capable than it actually is.

No.. no.. I get you. If he is willing to do it all again, he could do simply "-str 90 -nothrottle" and allow the sound/video to be rendered and get a more accurate "real" performance indication. Historically, though, MAME benchmarks have always focused most on CPU performance than rendering.