MAMEWorld Forums - News - Re: 64-bit Windows build now available.

> I can confirm that the 64 bits version runs way faster. Not sure about fps's but for
> example:

One thing you can do is disable frame limiting, allowing Supermodel to run as fast it can. Then compare. On very fast systems, it might become difficult to control the games, so I recommend creating a save state just before a scene that is known to slow the emulator down

> Do we have any kind of "way" to "know" how much mhz did these games used originally
> in each model 3 board?
>
> I mean, I know that:
> -Step 1.0: 32bits RISC PowerPC 603 66Mhz
> -Step 1.5: 32bits RISC PowerPC 603 100Mhz
> -Step 2.0: 32bits RISC PowerPC 603ev 166Mhz (Same for 2.1)

I'm not sure I understand the question. The clock frequencies you listed are what the games actually run at (although last I heard, there was some skepticism about Step 1.5 being 100MHz).

This notion of games being underclocked in Supermodel is somewhat artificial and really only relevant within the context of the emulator. Emulators are rarely built to be cycle-accurate simulators at the hardware level, certainly not with "complicated" architectures like the Model 3.

The PowerPC instruction timing is not necessarily correct and, more importantly, operations that would consume real time on the actual hardware, such as DMA transfers and rendering, do not eat up PowerPC cycles in the emulator. When the PowerPC issues a DMA transfer command, for example, the emulator does the entire transfer right away and then goes on to the next instruction as if arbitrary amounts of memory could be magically moved in 1 cycle. In reality, the DMA transfer would take some amount of time to complete and in the meantime, the PowerPC would continue running ("consuming" cycles, if you want to think of it that way, but without doing any useful work).

Basically, there are a lot of wasted cycles each frame where the PowerPC is waiting on other hardware to finish something. I don't need to run the PowerPC if it's just going to sit around and do nothing for millions of cycles. Better to underclock it. For the most part, the games are going to be timed off of IRQ events anyway (with some exceptions), so as long as enough cycles are executed per frame (and this is what you're asking about, I think), the result will be indistinguishable from running at the full 66 or 100 MHz.

What constitutes "enough" is virtually impossible to tell except through experimentation but all the games are doing similar things each frame, so there's probably some acceptable threshold we can identify. Your findings suggest that it may be 50 MHz, particularly for Step 1.5 and 1.0. Assuming the correct clock frequency is 66 MHz, a very crude estimate would suggest that the CPU spends about a quarter of the time doing nothing particularly useful.

Bart