Why SNES hardware is running faster than expected and why it's a problem | Hacker News Reader

Why SNES hardware is running faster than expected and why it's a problem

arstechnica.com

Emulation either via FPGA or software is really the only solution to preserving retro gaming and computing in the long term. Electronic hardware is very difficult to keep alive for decades.

Failing hardware isn't the problem. The consoles are showing nondeterministic behavior (slight variations in audio/video sync), and probably did from the factory.

The article repeatedly describes clock frequencies that drift further out of spec as the systems age. That's failing hardware.

The clock frequencies are not "out of spec" -- the manufacturing tolerance is 32,000 +/- 0.5%, and virtually all the measured consoles are well within that range. The manufacturer additionally specifies a maximum of 0.3% variation due to temperature and 0.2% due to aging.

There's also no evidence that the clocks are drifting further due to age. We don't have data for what APU frequencies used to be; we have a single measurement from over 20 years ago suggesting that one particular console had a DSP sample rate of 32,040 Hz, which is slightly lower than the current measured average of 32,075 but still well within the range of consoles we've looked at today.

Replace the resonator?

Indeed, there is some ongoing work towards creating a mod that replaces the APU clock with a circuit that derives a signal from the CPU master clock.

Sounds like a good idea but we’re talking about the TAS and speedrunning community specifically. Wouldn’t it count as a console mod and therefore delegitimise their runs?

It depends on the game and community; but generally speaking, console mods don't delegitimize a run if they don't change the behavior of the game in an illegitimate way. It's common for top-level runners to play on consoles with modded video encoding hardware for better picture quality, for example.

When it comes to TAS specifically, most SNES TASes cannot be replayed on console at all because the hardware is too nondeterministic -- the same exact input will not give you the same behavior. It's usually something trivial like an extra frame spent waiting for a sound effect, but that can cause you to (for example) enter the next room with different RNG state, causing enemies to behave completely differently, etc.

So being able to console-verify a TAS on a modded console, with a clearly-defined hardware mod that picks a single deterministic behavior from the set of observed behaviors on unmodded hardware, would be a strict improvement over the current status quo of only being able to replay these TASes in emulators.

Some speedrunning communities allow or even require certain mods to be installed as they ensure the same performance across different hardware

Replacing an aging part with the same part would hopefully be considered a repair, and in any case it's definitely less invasive than emulating the whole system instead.

Reply on news.ycombinator.com