Slashdot Mirror
Playstation 2 Linux Cluster at NCSA
Mr. Spock writes "The National Center for Supercomputing Applications is looking at scientific computing on the Sony Playstation 2. They've set up a cluster with 65 compute nodes. They're running Linux for Playstation 2. What will they think of next?"
More info on the processing power of the PS2 as applied to computational chemistry.
Basically, this study shows the PS2 has roughly the computational linear algebra power of a PIII-600 (the then fastest processor on the market).
http://www.us.playstation.com/hardware/more/SCP
~$200
-= Why can't I add 'Anonymous Coward' to my list of Foes? =-
320 MFlops on matrix ops. Not great, but they say it's capable of 900 if they can feed the VPU fast enough. They think they can use additional existing hardware in the CPU to increase memory performance.
2. Why would you bother when you could use current commodity hardware for much less? I mean, a P3-600 is interesting, but you could probably drop some Duron 1.4s with a basic mobo and 256MB RAM for less out the door than a PS2.
Maybe. Wal-Mart sells Durons (with Lindows) for $199 complete, sans monitor.
But, apparently PS2's are under $100 according to another poster.
ISTM that they may be spending a lot of time figuring out how to optimize code for the PS2 though.
Nothing to see here; Move along.
The big problem with using PC graphics cards is that the memory bandwidth on the AGP bus leading back to the system is abysmal. I know several groups looked into this when the Geforce 3 came out and suddenly we had a high speed low cost programmable vector processor, the result was that unless your application could return relativly small datasets you weren't going to get much performance out of them. I think a PS2 would be similarly hampered by the small amount of ram available, not many interesting datasets would fit in it.
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
Not a chance. Playstation 2 linux is designed for the MIPS architecture; the Xbox has a good old-fashioned x86 under the hood.
Also, Sony's linux distribution for the PS2 is based on a 2.2.1 kernel (old old old!), XFree 3.3.6 (again, quite old), gcc 2.95 (somewhat out of date, though plenty of systems still use it). The FAQ here says that the software is only slightly more recent than the software included with Redhat 6.2.
In short, there's nothing worth porting when you can get all of the Debian goodness for so much less work. I personally don't know why I'd want to shell out that kind of money for some second-rate hardware and a profoundly old linux distro with the price of commodity hardware being as low as it is these days.
Or were you just kidding?
At least not on the internet.
Nein for google.
One EXTREMELY LAME hit from deja. Surprisingly, the sig is identical.
Until you show some credentials (as in a link to nintendo's site, with a page with AT LEAST your name on it), you don't exist.
In fact, it appears your department doesn't exist.
Heck, where's your thesis, at least?
I find it neat, though, that you went from being Head of New Technology Research at SEGA straight to being Head of New Technology Research at Nintendo. More amazing, though, is that both companies have exactly the same departments!
More interesting:
<sgupta@research.sega.jp>:
Sorry, I couldn't find any host named research.sega.jp. (#5.1.2)
Look, provide me a page at nintendo.co.jp with your name on it, and everything will be sorted out.
Otherwise, this is:
Bill Gates,
Microsoft Founder
Redmond
Signing off.
If you could be told what you can see or read, then it follows that you could be told what to say or think - BoC
IIRC (I own the kit, btw)
Sony put everything they themselves created on the first disc. Their bootloader, PS2 technical documentation (covered by a NDA), and thier software development libraries.
Basically the Linux packages are on the second disc.
Bork Bork Bork!!
How else do you propose that people run arbitrary scientific programs on the PS2? Right now Linux is the only solution, nobody saying it's the end-all be-all of computing.
Some people might need to use SSL while connecting to the playstation2-linux site, or you'll get a blank page:
https://playstation2-linux.com
Someone does this in William Gibson's "Pattern Recognition" as an art project, using Timex Sinclairs (or whatever the British equivalent was called).
The vector units on the PS2 are far and away more powerful than the graphics card on the XBox. They're fully general-purpose vector processors, not a graphics chip co-pro. I *do* know a little about this topic - I've written a VU simulator...
... The PS2 has (for its' time) enormous bandwidth between on-chip modules, local cache RAM on all the processors, a general-purpose MIPS chip to keep things ticking over (run this one, wait,... run this one, wait...), and the VU's to do the heavy lifting from frame to frame.
Go to the Sony PS2 demo section on the PS2 linux site, and look at the VU demos there - or at least read about them. There are examples of marionette models being manipulated in response to the user input (x,y,z,buttons for impulses, etc.) on the controller. The physics is correct. The entire program runs on VU1...
These aren't graphics pipelines with programmable filters (Cg, for example), they're general-purpose CPU's with float and integer registers, maths operations (obviously), local single-cycle RAM (for programs and data), dma channels, interrupts, the works.
I'll start to take note of the Xbox graphics when you can download programs to it, tell it to execute them (until an event happens), register inbound/outbound dma data-transfers so it doesn't run out of data, and then let the main CPU get on with doing its own stuff. What's that, you say ? It can't ? Oh well. Shame.
The PS2 was an experiment in a new computing architecture - one that almost cost it dearly, since it's a pain in the *rse to program if you don't adapt to its' strengths rather than force it to use your own. It's a dataflow architecture - you download programs to the CPUs, then stream the data (vertices, colours, textures, etc.) through the programs from RAM using DMA and onto the graphics rasteriser.
The thinking is that most transformation programs can be expressed relatively succinctly, and that there's always more data than program anyway, so your algorithm for the bubbling ripple effect is (say) 6k, with (say) 2M of vertex,texture,lookup, etc. You feed in the same (initial conditions) data every frame, and a time clock, letting the VU program calculate the vertex manipulations... Because the data is always going to be much larger than the code, the overhead in switching VU programs during a frame is negligible, so do it whenever you need to...
When you think of the problem that the PS2 engineers were trying to solve, the architecture is very neat & very elegant
In fairness, the more recent PC cards are almost getting there - Cg2 will be better (it'll have loops!) I think the maximum program length is currently 128 instructions as well . Whoosh. But at least it's getting better.
To try and compare the two is laughable at best. Yes, they both produce similar games, but they do it very differently under the hood, and this thread was about using the VU's in a cluster to form a supercomputer, after all... (I think that's misguided, myself, even using BSP you'd be hard pressed to cope with the latencies involved, but that's another story!)
Simon
Physicists get Hadrons!