According to https://all-things-andy-gavin.com/2011/02/06/making-crash-bandicoot-part-5/

Andy had given Kelly a rough idea of how we were getting so much detail through the system: spooling. Kelly asked Andy if he understood correctly that any move forward or backward in a level entailed loading in new data, a CD “hit.” Andy proudly stated that indeed it did. Kelly asked how many of these CD hits Andy thought a gamer that finished Crash would have. Andy did some thinking and off the top of his head said “Roughly 120,000.” Kelly became very silent for a moment and then quietly mumbled “the PlayStation CD drive is ‘rated’ for 70,000.”

Kelly thought some more and said “let’s not mention that to anyone” and went back to get Sony on board with Crash.

So what happened? This seems like much too big a deal to not follow up! Given that Crash and its sequels sold millions of copies, even if the rating was to 5% failure probability instead of 50%, it seems hard to imagine hundreds of thousands of broken CD drives being just forgotten.

Was the rating pessimistic by a particularly large margin? Did Sony switch to a more durable model of drive? Did they in fact eat the cost of replacing a bunch of drives without connecting it to Crash Bandicoot?

Did Crash get away with it because no one else thought of constantly reading the drive that way? Or is it the case that other developers tried, and Sony smacked them down? Did all the Crash sequels do the same thing, or did they figure out a less drastic solution for Crash 2?

  • 1
    Mind to add some information for the non initiated? What is a Crash Bandicoot? And in this context what is a forward and back? And what consists a hit? Just reading, or positioning? Equally important, what is the data used based on? Were do the percentage you mention come up? – Raffzahn Jan 30 '20 at 10:40
  • 1
    @Raffzahn Crash Bandicoot was one of the most popular games on the Sony PlayStation; more details are in the link I provided, which is a long story but excellent reading, highly recommended. It doesn't say what a hit consists of, but the mention of going either forward or backward through a level, together with the characteristics of disk drives in general, suggests it must be positioning plus reading. – rwallace Jan 30 '20 at 10:49
  • @Raffzahn The article doesn't say what the data is based on, or what failure of probability corresponds to the rating of 70,000 hits; I suggested two possible failure probability numbers to point out even quite a low number would still correspond to a lot of dead CD drives. – rwallace Jan 30 '20 at 10:50
  • The points were only in part for my information need (in know the game), but rather meant to improve the question text so it may be helpful to future readers as well. Pointing to a no further specified external link is maybe suboptimal. – Raffzahn Jan 30 '20 at 11:00

The data here is a bit unclear what the mentioned measurement of 'hits' is about and where the percentage of 5 or 50 comes from. The whole setup doesn't give much information.

Lets try to see what it could mean in relation to real world numbers.

CD-Drives are usually not rated in 'hits' but MTBF hours, like most machinery, often amplified by duty percentage. A typical early 90s drive, like a Mitsume 2x drive had a 5,000 hours MTBF at 100% duty, while a late 90s 12x drive, like the CRMC-FX240, was rated at 50,000 hours power on with 40% duty (equalling 20,000 hrs at 100%). After 2000 these times walked up to more than 100,000 hrs. For now it may be safe to assume that Mitsumi wasn't in the highest class of reliability :)

5,000 hours of usage are quite a lot for a single game. That's like playing for 8 hours a day, 5 days a week for 2.5 straight years. Unless a particular drive ends up in the 16-percentile of standard distribution this guarantees a full year of playtime.

Sounds not really plausible. So what about the mentioned 70,000 'hits'?

The cited text can be read as if this may mean each access of the drive, which happendes whenever the scene changes - like a new level or screen. Assuming that such a screen change will happen every 10 seconds, 120,000 hits would end up being like more than 300 hours of play - about a year by playing an hour a day, or still 8 weeks of daily 8 hours play if payed to do so :)) Did they really think it would take that long to finish that game?

Again, a bit strange, isn't it.

So, without some useful related data, I'd put these numbers into the distant memory category, with no real world meaning.

  • 3
    Minor, barely worth mentioning, but having played the game I think the blog poster is meaning that the drive was loading small chunks of geometry continuously — it’s a game in which you move essentially in a straight line — no need for a scene change. Or it may just have been loading per-polygon visibility flags per camera position (which follows an almost-fixed path, but at least is always pointing in a pre-determined direction and within a small distance of a fixed path). – Tommy Jan 30 '20 at 12:54
  • 3
    @Tommy Hmm. That would bring it again back in the territory of continuous reading wouldn't it? Something were even the cheapest mechanic should provide several thousend hours. The quote seams dubious. – Raffzahn Jan 30 '20 at 13:44
  • 2
    Probably the issue is seeks (that's the only real difference between a very "active" drive and one spun up doing nothing). But I agree the whole story is dubious. It might have a core of truth in it, but if so it's been lost to fuzzy memory or bad editing. – hobbs Jan 31 '20 at 3:45
  • 1
    Did the Sony Playstation have any kind of disk caching? If so then a "hit" whatever it is might not mean a physical disk access. – JeremyP Feb 1 '20 at 13:55
  • 1
    @Raffzahn I find the claim to be somewhat strange anyway. 70,000 individual reads doesn't seem very many to me. – JeremyP Feb 3 '20 at 10:01

Firstly, I haven't found any anecdotes of Crash Bandicoot causing premature drive failure in PlayStations. While this mostly answers the basic question, it leads to the obvious question of why. It certainly is true that Crash Bandicoot did rely on streaming data from the CD whenever the player moved through the level; the whole level simply wouldn't fit in RAM.

The key, I think, is that there are several very different forms of mechanical wear in a CD-ROM drive. Probably the most severe form is a start-stop cycle which stresses the spindle motor, its mountings and bearings, and the friction pad coupling it to the disc. Next in line is a seek where the head has to be moved across the disc surface, to find the correct place on the continuous spiral track that CDs use; the head transport mechanism is optimised for very fine movements and isn't very good at moving long distances. Finally, the head has to switch on the laser, servo onto the track, verify that it's in the right place, and read the correct blocks into the drive's buffer - something that an audio CD player has to do continuously for an hour and a quarter at a time.

In a more basic game engine design, where the whole level is loaded up-front, a "hit" would probably involve at least one instance of all of these types; the disc would spin up, then several seek and read operations would occur, then the drive would go dormant again. This is probably what the "70,000 rating" the Sony guy was thinking of referred to; remember, that's a number that has to correspond to the whole warranted life of the console.

But a streamed level behaves much more like playing music in one of those portable CD Walkmans, where the data has to be read frequently, but the disc spins continuously and the head doesn't move very much. Each "hit" in Crash Bandicoot only involved performing a read operation very close to where the previous one had left off. Indeed there was a high probability that the blocks needed were already in the drive's read-ahead buffer, requiring no explicit mechanical activity to supply. So even a drive of moderate quality would have been perfectly happy with that access pattern.

  • 1
    I don't believe that the PlayStation ever spins down the drive except when the power is turned off or the door is opened, so game behavior can't meaningfully influence start-stop cycles, only seeks. – hobbs Feb 1 '20 at 5:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.