Slashdot Mirror
Kutaragi Admits Sony Hardware In Decline
An anonymous reader writes "In a surprising admission, Sony Computer Entertainment President Ken Kutaragi acknowledged that Sony's strength in game hardware might be in decline. BetaNews has the article, which reports on Sony's PS3 struggle for the holiday season." From the article: "In an extraordinary public statement of regret and despair over having to postpone his company's PlayStation 3 debut in Europe and Australia until March, and to limit availability elsewhere to only 500,000 units come November, Sony Computer Entertainment President Ken Kutaragi is quoted by Reuters as having told reporters, 'If you asked me if Sony's strength in hardware was in decline, right now I guess I would have to say that might be true.'"
Cell is a pretty sweet piece of chip for video games
Actually, I'd personally argue that the Cell is pretty crappy for videogames; its a very powerful processor that can do a massive number of floating point operations but doesn't really address the fundamental performance problems that exist in Videogames.
In videogames (like most applications) 5% of the code takes up about 90% of the processing power; take scene graph management as an example, most of the effort to manage your scene graph (to determine which objects to be rendered) is doing matrix-matrix multiplications, vector-matrix multiplications, and vector cross/dot products to determine whether the bounding sphere is within an arbitrary box, segmenting the box, and then testing whether the bounding sphere is within the 8 new arbitrary boxes. On the Cell processor each matrix multiplication will result in 64 floating point multiplications and 48 floating point additions (with an aditional 128 integer multiplications and 64 integer additions if the programmer is dumb), each vector matrix multiplication results in 16 floating point multiplications and 12 floating point additions (with an additional 32 integer multiplications and 16 integer additions if the programmer is dumb) and so on; if you were truly designing a "games processor" you would include far more vector instructions (with the possiblility of a hardware based matrix multiplication) to greatly reduce the ammount cycles needed to handle these types of operations.
The reason Sony didn't do this is that this would reduce the clock speed that they could run the processor at (in theory, a matrix multiplication could do 64 floating point multiplications at the same time, which would use a lot of energy, which would produce a lot of heat, which would mean that you couldn't run it at nearly the same speed) and thus reduce the speed at which you could run generic code; the fact is that Sony and Toshiba have already said that the Cell was designed to be included in their entire electronics line, the only reason it is the CPU in the PS3 is that they had to spend the money to develop a CPU for the PS3 anyways and they could use that money to develop a generic multimedia processor instead.
Trust me, Sony could have developed a much more powerful CPU (for gaming) at a fraction of the cost if they wanted to; PS3 early adopters are paying extra money so that Sony can use the Cell in future TVs to upscale a 480p image to 1080p.
I have yet to see any quoted source on the insane complexity of the devkits. Nowhere. The closest thing I've seen actually quoted is that it's difficult, but not near the shittyness that was the PS2 Emotion Engine.
The big issue is that Sony seems unable to manufacture the full system.
Disclaimer: I'm somehow biased since I work mostly on PS3. But things I have to complain about are on the software side, not hardware...
The problem with the PS3 devkit (as of SDK 092) is that the software just isn't good enough to let developers work efficiently with them. For example to update the flash ROM of the kit (one with each new SDK version) you have to copy it to a usb storage device, power down the kit, reboot the kit and insert the key *at the right moment*, that's just stupid (both 360, PS2 and previous versions of the PS3 devkit would let you do this by network). The Sony dev forums are full of posts from developers trying to recover from bricked kits.
And speaking of the dev forums, developers are getting really pissed off, the tone as been changing lately. These forums are usually really polite and politically correct, but we're starting to see some really snarky comments from veteran developers like David Etherton (who's doing both PS3& 360 development and had done a lot of PS2 dev). There's even a guy from Insomniac (a Sony 1st party) who basically declaring Sony dead by their own incompetence. Basically what most of the devs are saying is that the tools and the doc suck and everybody is on his own trying to get something working, and Sony doesn't seem eager to help. For example, IBM is working on some compiler technology to enable arbitrary code to run on the SPUs, Sony's answer to that is that it's not the most efficient way to use the SPUs so they flat out refuse to integrate any of that in the PS3 tools.
The Cell is an interesting processor, but in a multiplatform context it's just isn't what the devs need. It used to be that on PS2 some of the low-level engine guys and the graphic guys would have to get really intimate with the machine to get it working correctly, that's only a few people. But right now with the PS3 if your game is planning on using the SPUs for anything other than graphic stuff you'll have to train your entire team to use it since none of the C/C++ code you'd usually write will work well (if it event works) on it. This cannot happen, there's not enough low-level guys in this industry. The PS3 also has 8 cells, 1 is disabled to improve yields, that leaves 7, 1 is reserved by Sony that leaves you with 6, and 1 other can be requested by the OS at any time. They also lock 96MB of memory for the system (32 in VRAM, 64 in main RAM), that's like 19% of the memory. That leaves you with what good be a good machine but the architecture is really stream oriented, and most game code just doesn't fit that paradigm.
And the RSX is getting humiliated by the 360's chip, it just isn't funny how much slower that chip is. On paper they look a bit similar but the reality is that the 360's chip is faster at context switching and can pipeline state changes this gives it a a real big advantage with real work game scenes (compared to useless tech demos and benchmarks). And since it uses unified shaders engine the performance is pretty much always balanced, quite unlike the the RSX. And we won't even start talking about the moronic RAMDACs in the PS3, we've compared the output quality of the PS3 and 360 with the same image side by side with identical TVs, the PS3 has an enormous ammount of ringing arround edges in 720p, the 360 has a much better image quality.
And back to tools, on one side you have Microsoft with Visual Studio 2005 with a custom compiler full of special 360 extensions for better performance, a ton of optimized libraries, great documentation and PIX. PIX is like to coolest tool ever invented to debug and profile graphic performance, the PC version is not so good because it can't exposed the low-level stuff but the 360 version is completely out of this world. None of our graphic programmer does it's primary work on PC because of it.
On the other side you have Sony with a custom version of GCC and GDB as your debugger, custom only the fact that it targets Cell because otherwise it's pretty much the same thing, and the SPU version has tons of problems with dead code removal (which due to the SPU limited memory size, is frigging important). Now I like free software but GCC just isn