Rather the opposite. I think it ended up as part of several compute clusters because the console was sold for less than it cost to produce, something normally offset by game sales. So suddenly Sony had a large market that only cost it money.
Rather the opposite. I think it ended up as part of several compute clusters because the console was sold for less than it cost to produce, something normally offset by game sales. So suddenly Sony had a large market that only cost it money.
I'm sure internally it was pitched as a neat idea why don't we see what happens (god can you imagine the corporate politics in a decision like that?), but practically it allowed Sony get get special import status.
I'm sure eventually they realized they didn't care about that enough it outweigh allowing those filthy users to possibly do something interesting.
Sony throws a lot of stuff at the wall. There was a good chunk of hardware cut out of later PS3s, not to mention continued attempts and silent canceling of things like those move lightguns, etc.
[0]: https://www.wpafb.af.mil/News/Article-Display/Article/399987...
Wasn't more an issue that games didn't specifically optimize for cell processors? They had to run on dozens of platforms and the Cell processor was the odd one out.
> to get useful compute out of it during the length of time that option was arounnd.
There were at least three largish cluster setups, the biggest being a super computer with almost 2000 PS3s. Of course you are right, that is barely a blip on their sales.
That’s what being exotic hardware means.
The major use case for these was for number crunching (training neural nets, running physics simulations, etc). And these things were something crazy, like 100x faster than a similarly priced PC server. There were plenty of libraries out there for doing linear algebra.
A quick google search suggests that this was extraordinarily common in academia. I found materials from lots of different universities that were close to what I remember my peers talking about.
This was also a little bit before CUDA hit. So there really wasn't a mainstream offering for doing massively parallel, SIMD calculations.
Edit: I went to Wright State Uni and most of my CS professors did research associated with the Air Force, which as forgotpwd16 notes, had at the time a super computer made from PS3s. This might explain why the PS3 was so popular among our department.
I think it's an urban legend fueled by marketing, I've never seen any hard proof of that, except some BOM calculations from independant analyse, but I don't think mass production work that way.