MAMEWorld Forums - EmuChat - Re: Midway Thunder Titles

> Please correct me if I am wrong but I would imagine an overclocked i9 processor is
> the king.
> That will likely change next month when the latest Ryzen chip becomes available.
> Cores help with dimishing returns per core. Clock speed remains the most important.
>

We'll see, so far no AMD has been able to match the Intel chips for top-end MAME performance, not even been close. If that happens it will be the first time in about 20 years.

> Thoughts:
> Could Mame run faster if optimized for multicore processing? As in more than 2-3
> cores?

The more you thread the less reliable / accurate you become, or the more overhead you have on thread management and copying of states around for the threads (which then becomes the bottleneck instead) Emulators for the more modern platforms get away with it because most of the games are built to run on consoles which are doing tons of background tasks anyway, so are tolerant (to a degree) of things happening at different times and differing from run to run. (even then there are games that aren't tho, there are PS3 games that will glitch / crash on original hardware due to timing, eg. if a trophy award/popup background task takes too long, which happens with large trophy libraries)

>
> Does Mame take advantage of large amounts of ram when available?
>

It will use the RAM it needs, no more, no less. RAM isn't something you magically take advantage of.

> There use to be versions of Mame specifically for the AMD cpu.
> Any point revisiting?
>

All those optimized versions gave marginal differences anyway, usually with extra bugs because the optimization stage of a compiler is often where bugs get introduced, and those CPU specific codepaths are less tested.

> Has there been any attempts more recently to utilize the GPU instead by Mame or other
> emulator projects?
> HLSL uses the GPU, right?

HLSL uses the GPU. Keep in mind that in general purpose application GPUs are only suited to certain tasks, essentially they provide a LOT of cores, doing many tiny parallel tasks. Emulation is inherently a serial task, so you're back to the above problem of there not being much suitable to execute in parallel; you can't emulate CPUs that way, and those are often the main bottleneck. There is talk of doing software rendering on the GPUs at some point which should free up some CPU cores.