Playstation 3 CPU Almost Finished?
dnxthx writes "According to this ZDNet article the design of the Playstation 3 chip is nearly complete. The PS3 chip will have near "supercomputer capabilities" --- including 1 TFLOP. Reportedly, this chip is being engineered with Linux in mind."
Cell's designers are engineering the chip to work with a wide range of operating systems, including Linux.
I don't see how that sentence translates to the statement by the submitter that the chip is designed with Linux in mind. Besides, shouldn't the OS adapt to the chip, not the reverse?
TheJapanese government realised that the computers in the PS2s were very powerful for the time and could be networked to create a crude missile guidance system.
you haven't seen anything yet.
In terms of scalability, the uber-parallel-processing-pipelined PS2 makes a lot of sense, and will continue to get more powerful in the future as its software improves. In terms of usability though, the PS2 has irked a lot of console developers because it's an entirely different beast and doesn't behave like a PC when you get down to performance bottlenecks.
The PS3 and beyond can only continue this trend. Sony hopefully won't make the same mistake ease-of-use wise, but the PS3 will be getting tantalizingly close to the "do everything you ever cared to do in a game" performance.
The future of this technology is hugely dependant on software capability to make sense of and utilize it. This will be the biggest hurdle, and clearly nothing like it really exists today.
One of the next big steps may be carbon-nanotube based computing, because it will enable architectures with massive hierarchical processing power and near limitless involatile stupidly fast memory, all embedded everywhere. Carbon (and other) nanotubes will be used for both logic and memory (as well as actual display surfaces), and ultimately be laid out more like a brain than a serial system.
I look foward having a complete system in a display where you push morphing procedures in one end which ultimately get streamed into content on the output side.
The networked aspect will be important too, but not how it's colored in this article. Your games will ineveitably run graphics processing on your local machine, with non-realtime and background tasks offloaded to others on the network. However, distributed simulation of gaming environments will only really make sense when players become the content producers and the worlds expand procedurally to simulate whatever ideas of interest their imaginations have conjured.
Then I just have to ask, when game consoles power the realization of our imaginations, whose world are we going to be living in? [hint: this is rhetorical, don't answer, just think about it]
At this rate, commercial production of Cell could come as soon as the end of 2004.
The article states they've merely got the pen and paper design almost complete. No working hardware, and it 'could' end up in the PS3
Toshiba and IBM have had more than their share of flops.
Remember the Toshiba MPACT chipset that was supposed to take over the 3D Graphics/Sound/Video market in the PC world?
Why didn't they just buy out transmeta? I know they just had a big round of layoffs, lost some big contracts, and can really use the cash right now.
The main benifit of course would be having linus. Throw in the transmeta technology after that.
The really scary thing about the whole sony/linux relationship is the parent company Sony is also Sony Records, one of the biggest supporters of DRM and the DMCA. It's kind of odd that they would support an open O/S that will never have DRM in it, makes me wonder why?
--toq
The cake is a pie
The terraflop statistic is a little hard for me to swallow.
The NERSC IBM SP RS/600 (the fifth most powerful computer in the world, according to top500.org) located in Berkeley consists of 2,944 processors. The processors are distributed among 184 compute nodes with 16 processors per node. Each node has a common pool of between 16 and 64 GBytes of memory.
This machine is a 3 terraflop system. Although, I guess three PS3's could do the same...
I'm having some trouble believing that in two years there will be a consumer chip 100 times as fast as the ones today. Moore's law would say that it will be twice as fast. I'd believe 5 times and maybe even 10. But not 100. ZDNet is way too gullible.
From the article:
:), along comes 'linked' cpus. Sure parallelization rocks for performance, but it's a head ache for game design & implementation. This is one thing the X-Box got right - port your PC game over in days, not months. Ok, enuf k'vitching.
;-)
"It's like a beehive -- cell components can also be ganged together," he said.
Just when I thought programming the PS3 couldn't be any *worse* the then PS2 (lots of fun debugging the EE, VU0, VU1, GS, SPU, IOP all running simulatenously on the PS2
How long do we have to wait for Gran Turismo to show-case the PS3 ?
The cell is a highly parallel chip, it is outside the bounds of Moore's "law" because it doesn't follow the same design methodology. If I designed an FPGA today that had 1000 FPU's, and a simple CPU to control them, I could easily best a P4 in FLOPS. Trivial. Sony has done/will do in hardware what I have suggested, and given that they've been working on it for a couple of years, I think there may be more than just a couple of extra FPU's.
All it takes is a little thought....
Simon
Physicists get Hadrons!
It used to be the first thing one did with a new processor or a new platform was port BSD to it. BSD is the most portable operating system in the world. Now NetBSD and FreeBSD are freely available in source code format, so it is ideal for porting to new platforms. It's more stable and fastert and more advanced than linux because it doesn't have to re-implement everything from scratch.
No one said ANYTHING about 1 trillion triangles being pushed out. This is pure micro-operations were talking about. I will bet you cant even add 1.0 + 1.0 a trillion times on this new processor. Triangles need to go down the graphics pipeline, and there are three points on a triangle. Since this is a processor, and not a GPU, the processing time will have to be divided up among all the application's functions. Not even a hype crazed marketing exec. would try and claim a trillion triangles per second.