The Crux of Sony's Strategy Problem

[TheBigFatJ]

Sqrl said:

smallflyingtaco said:

Sqrl said:
Actually I will take my conclusion a step further and say that the architecture is actually quite ideal for the sorts of scientific endeavors that we have heard about. For a research team the PS3 represents a powerful computer that facilitates the sort of workload breakups that are common in this sort of research. The result is that workloads can be prioritized and a workload designated to have a higher probability of interesting results can be processed more quickly.

As for price, looking at the normal cost to purchase time on a supercomputer, the PS3 is actually a cheaper alternative when you factor in the need to schedule time on a supercomputer and the need to pay a fee for every such use. Where the alternative is a one time purchase price and constant uninterrupted accessibility.

Just to avoid any backlash, I am not suggesting that the PS3 be relegated exclusively to such tasks, only that the price point and performance of the unit is more in line with this sort of work and that there is a disparity in its intended purpose and practical purpose.

This is not true for all scientific applications. There are problems that are more about memory than processing power or cases where breaking your problem down to fit within a PS3s memory constraints would increase your processing requirements beyond the advantage you get from your PS3. Outside those constraints your right, the PS3 is a great deal but if Sony is lossing money on the hardware or making a subpar profit on them then they do not want to sell their ultracheap workstations to people who will never buy software for them because they are becoming part of clusters.

 

Very few things are absolute, but I feel confident in saying my point was true for a reasonably large percentage of uses. But thats really not the central issue here, the big thing is that the architecture is not practical for gaming purposes. Expense was spared in places (as you pointed out) like memory where it could have done a lot of good and it wasn't spared in other places where they were recieving diminishing returns.

As far as Sony not wanting scientists to buy the console, I disagree. They absolutely love the PR generated by these sorts of stories as it perpetuates the idea that the PS3 is a supercomputer even though that just simply isn't the case, its a very robust system no doubt, but it is truly several orders of magnitude away from being in the same league as current supercomputers. Anyways, as I was saying this sort of PR is great for them and even if we assume they are taking losses on the 8 or 9 consoles they sell to these guys they easily recoup that value in free advertising....but can you imagine if someone were to actually cure a disease or make truly important discovery on a PS3? That would be the PR equivalent of winning the $320 million jackpot.

When it comes to everyday folks there really isn't a lot they can do to stop you from purchasing the console simply for a cheap workstation. Not like they can stand at each store and screen folks. But what they can do is pack-in some games and hope it intrigues folks into buying more.

I happen to be an HPC admin.  A minority of jobs run on my clusters can be done in less than 512MB/memory per job, and that alone wouldn't fit within the PS3's system memory (256MB).  In fact, in 2004 we purchased our first 192GB memory machine so we could run models that consume hundreds of gigs of memory.  And recently I upgraded one of our clusters to 16GB of memory per node.  Necessities for performance.

What's more, I/O is becoming a bigger and bigger necessity.  A single gigabit connection isn't enough anymore.  The PS3 could be used on small scale for very specific tasks, but the users would quickly discover that they're limited to a tiny problem space whereas much of the interesting stuff requires significantly more memory.   Hell, I just added 250TB to our data center to hold data, and we already had significantly more spinning disk than that.