¿Qué aspecto técnico de los videojuegos os ha sorprendido más por su complejidad?

issus escribió:Yo tengo un amigo, que hizo su testis doctoral sobre paralelización de algoritmos de raytracing. Eso fue a finales del siglo pasado, principios del actual, hace unos 20 años. Ahora corremos juegos con capacidades de raytracing en tiempo real, que aunque evidentemente no son tan complejas como las que él ejecutaba en sus pruebas, le llevaban minutos en ordenadores potentes de la época. Me flipa, por tanto, el RT actual, aunque sea limitado.

Saludos

It's captured at 10 bits per sample, 40 million samples per second. This was found to be a good balance in terms of USB3 bandwidth, and the lower sample resolution isn't an issue as the most important thing is frequency resolution. WAV (i.e. RIFF WAVE) is just a container format like AVI, though the most commonly-used data within the container is linear PCM at 44.1 thousand samples per second and 16 bits of resolution. You're not wrong that the resulting file is a series of samples, but it's like referring to any given video file as an "MP4".

The resulting captures (.LDS files) clock in at around 100 gigs for a single side of an hour-long disc, but can be compressed with FLAC for roughly 40-50% size savings, the resulting files bearing the extension .LDF. End-to-end, going from a single raw .LDF to a HuffYUV-encoded, AVI-containered video stream takes about 8 hours per 30 minutes of video on my 4.1GHz Zen 2 3950X (which has 16 cores and 32 threads).

Once you get past the laser pickup, LaserDiscs are almost completely different in terms of how they're read - or at least converted into data - compared to a CD. The pits and lands on a CD-DA disc translate directly into a self-clocked stream of bits. On a LaserDisc, it's not the pattern but the length of the pits and lands that matters. For all intents and purposes, it's an analog optical format.

Funny enough, audio CDs and LaserDiscs do have one point of overlap, which is that one of the two audio channels on the LaserDisc can contain data which has been EFM-encoded and then modulated onto the overall carrier. Prior to that modulation step before it's mastered onto the disc, the EFM data broadly resembles what you'd find on an audio CD.

The above point is one of the main reasons why the classic "just run the disc through a capture card" plan is flawed: You just plain can't capture the digital data. In the case of the BBC Domesday 1986 project, that means you can't actually get at the underlying databases, map survey data, and other things that the software needs to run. In the case of LaserActive games, it means you can't get at the relevant program code and data for the various MegaDrive and PC-Engine titles.

Having what amounts to everything on the disc, and doing the processing in software later, throws the door open to a wide array of methods for getting a theoretically-perfect capture. More and more LaserDiscs are succumbing to gradual "laser rot", where the metal layer stops adhering to the acrylic disc itself. This prevents the laser from maintaining tracking for brief moments, and results in little sparkles or brief portions of incorrectly-decoded video. This loss of tracking is usually accompanied by the RF signal from the laser pickup swinging well outside a nominal range, and so this can be detected by the ld-decode pipeline. It can then look at the next line, previous line, left/right on the same line, or even forward/back one field in order to find the best fit to fill it in.

For the ultimate capture quality, if you have 3 or more of the same disc from the same mastering run, and you've taken the time to get .LDS or .LDF captures of each one, then the ld-decode toolchain can perform a median function for every single Y/C sample of each capture. Statistically, the vast majority of dropouts are going to be on different spots in each disc, so given enough input captures from different discs, it should be possible to reconstruct a capture that is 100% dropout-free.