unforum

[astrovanman]

[PS3] Playstation 3 Discussion

[SuperJerms]

Well, if any of you read my extensive posts in the Xbox speculation thread, remember that the big deal was supposed to be the chip's ability to do parallel processing. Waters is sceptical that the chip can do this effectively, but I linked a really techincal article that exained the cell architecture patent.

So, everyone is freakin out about how powerful the cell chip is. Yes, the thing is powerful, but that's missing the point. A more powerful chip (that overcomes the problems in silicon processors) is not the biggest coup Sony could pull off. The biggest coup would be to create systems that allow decentralized processing.

Like has already been said, the cell is not new technology. From what I understand, it's basically a PowerPC with no cache and faster read abilities. It's somewhat similar to current graphics processors in terms of setup. Mac uses all PowerPC (G5's), and Texas Instruments has a processor set up similarly.

Notably, this makes the instruction basis for the two systems vastly different. That could have a big effect on gaming, since games won't be very simple to port back-and-fourth between Xbox and PS3. PS3 has a theoretical maximum of 2 teraflops processing. Key is that this is the theoretical. The actual performance will not be that high, though I understand that the more vectorized instructions are sent to the cell, the closer to the theoretical max perforance becomes.

Again, if it's not the speed/power that matters, it would be the decentralized processing. There is really no question about it, the future of computing will revolve around a decentralized model. It's not a question of if, it's a question of when. So, if Sony is the one who brings the technology, Sony will rule the world (provided they can populate the technology in the wild). Pre-E3 spec was that 4 cell processors would ship in the PS3. As I said before, that's a ridiculous amount of power for a game system. One cell makes more sense, but consider...

Assuming Sony really did make a chip that can effectively share processing power with other chips (without being on the same circutry) how do the get the public to adapt. Four cells per PS3 (times a couple of million units) would have definitely posed a threat to the Wintel world. But only one comes with each box. Cost-wise, good sense. I'm sure four cells would cost $800 by themselves, not to mention the other parts of the PS3.

How now, brown cow? One cell per consumer device...hardly gonna be a blip on Intel's radar, let alone a contender for the PC world. And that's still assuming they figured out a way to transmit processing data wirelessly (another big "if").

Water was right to pose that latter "if." Wireless speeds are nowhere near circuit speeds. Still, the patent does break down a decentralized system. Basically, software divides processes up into cells of instruction sets. The cells can process any given instruction set, and the technology is set up to put everything together properly after processing. Thus, reconfiguring the hardware doesn't hose the system...one step closer to simple parallel processing.

Obviously, the data still has to be transmitted somehow, introducing the bottlecap of wireless speeds Waters mentioned. Still, the hardware problem has always been more pressing, so Sony's solution could still be a big deal. With only one cell per system, it kinda feels like a moot point. Nobody will have the chance to create a multicell system until there is a software interface that allows for parallel processing (by processing via software cells instead of a stream) and applications that involve more than one cell.

Will that ever happen? I imagine it absolutely HAS to. Sony is taking a huge gamble having developed the cell. If it doesn't do the decentralized thing, they wasted a ton of money.

[Phaedra]

You're all so cute when you geek out.

[SuperJerms]

[weephun]

[SuperJerms]

A speculative analysis of Sony's earliest cell patents and overview of how cell processing would probably work based on said patents and Cray supercomputer designs.

Waters posted this in response, though I havn't read it yet. It includes spec based on recent mobos.

[water10]

The anandtech article is very complex. It goes very deep on the details of the Cell. I can't really say I understood everything ... The reason I like Anandtech is because since I've been building my own computers (last 5 years), I've been reading several web sites and forums, including tomshardware, anandtech, ambmb and others. Anandtech seems to be the most reliable. It's one of the most common web-sites among geeks! They are not biased towards Intel or AMD and on this case, Sony.

The whole thing about distributed computing is just too far ahead! Just now, Intel introduced hyper-threading, with a processor being capable of multi-tasking more efficiently. Now, Intel and AMD is going multi-core. Sony is also going multi-core with the Cell. We're talking different CPU's processing and exchanging information at VERY fast speeds! It's not only wireless speeds that are not up to the task. Gigabit ethernet is not enough, neither is USB2 or Firewire. You just don't have enough bandwidth to have distributed computing like that article suggests. Yes, it will happen eventually, but it's not there right now. After reading both articles, I'm with anandtech. They pretty much get to the conclusion that the Cell can work pretty well on consoles, but there's jut not compatibility to bring them to the PC world. It can be done, but it's not likely ...

Now, considering Cell on the PS3, I like what I'm hearing/seeing. But it's just too damn early to say if Sony is being over optimistic (in other words, flat out lieing) or if the PS3 can really output those projected specs and have real-time games like the demos they showed! I'll be waiting to see. As I made clear in other times, I think Sony has a very good marketing department that not always delivers what's promissed ...

[SuperJerms]

Just finished the anontech article (I understood everything but the transistor section, which I just skipped over).

Really, the two articles are saying pretty much the same thing...but the anontech one only talks about internal data parallelization. And, like waters just said, neither article really tells how they are going to overcome the wireless bandwidth problem. Modern graphics processors read/write at 32 GB/s, and the anontech article says cell SPE processors will have 76.8 GB/s memory bandwidth through the cell Flex I/O interface! I too am curious how they could overcome the transfer bottleneck.

There is one (weak) in-order PPE processor, which can handle the general computing stuff and has a L2 cache attached (never saw a size for it, though). Also, cells have 8 SPE processors that handle in-order instructions, have no cache, but do have super fast 256K local memory. The local memory handles any data dependencies, so memory latency is reduced to 6 cycles for the processor (as opposed to 10 for cached data or 200+ for a cache miss!). Both articles say that parallelization will work really well, but the workflow will depend on how the complier and/or software is written to send instructions.

Since the cell is in-order, general computing will suffer if there are any cache misses in the PPE processor. In a console this doesn't matter, but in a non-game or media PC it would matter. That's where the viewpoint differs between the two articles. The first article is really enthusiastic about general cell computing, the second is more focused on console-specific tasks. They really only disagree on that one point...their analysis is roughly the same.

What anontech is saying is that the cell is limited by the PPE processor being weak. In the face of Intel's deep user base, millions of cells would have to be sold in order to pose a treat. The first article says the same thing, but it expected four cells per PS3, plus a host of consumer devices to fill out that population requirement. So, one cell per PS3 is a pretty big deal.

Both articles also point out that Intel has said that arcitecture like the cell is the future of computing. Intel will make cell processors. It's just that they don't plan to start for another five years after the PS3.

Sony has spent a lot of money making the cell. Can it afford for that investment NOT to catch on into a computing revolution?

[Scarr]

I think you are all missing the point … this is a pissing contest. Microsoft hasn't made any money of the Xbox they are in it for bragging rights. They are selling the consoles so you buy the games and hook up your computer to your entertainment center and upgrade your software and get the gold Xbox live etc. Basic marketing the next gen consoles are there to get you to spend more money.


Think iPod or early IBM. Yeah you get an iPod shuffle for a c-note but after a life time how many songs do you down load off of iTunes? Do you remember that Microsoft got the licensing rights from IBM for their software because IBM (and most of the world) had no idea how versatile a computer could be….

Turing look what your legacy has wrought*….

*Even Turing knew that "software" was the way to go it is to bad the gave him hormone shots to "cure" his homosexuality which drove him to commit suicide by pulling a "Snow White"

[water10]

[Scarr]

I didn't mean to come off as harsh i just was saying you should take into account their marketing into your conversation. Though i agree with you the more i hear the more i want both of them

[water10]

[Grifter_7]

I am an admitted Xbox fanboy... but I must say that the PS3 outshines the Xbox 360 so far.
Go here to see a side-by-side comparison that IGN did.

[SuperJerms]

[jyemuzu]