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?"

Comments

Well.

[Pres. Ronald Reagan]

This seems like a fairly expensive way to make a cluster. $200 for each PS2 and $200 for each Linux kit? That comes out to $26,000. You could buy computers with more RAM and faster processors (than a 400MHz MIPS) for about the same amount.

Playstation 2 Linux

[Anonymous Coward]

Apparantly this runs on Sony's own version of Linux

See more about it here: http://playstation2-linux.com/

Maybe an XBox port in the future? :)

Linear price declines, exponetial performance gain

[SpikeSpiff]

The issue with using game hardware, especially "older" platforms, is that the price can not come down as fast as overall computing performance can increase.

This means that the cheapness of stable platforms can not compete with innovative platforms.

The real question is whether the administration and maintentance benefits of a homogenous and stable platform outweigh the higher cost of processing power.

I suspect that we will see a step function between rapidly and smoothly improving Dell boxes and occassional huge leaps on game platforms.

We've investigated GameCube clusters too.

[Samir Gupta]

At Nintendo, we've been excited about clustering applications of game consoles as well, and we've pursued a active program of research within our company.

Internally, we've experimented with large clusters of GameCubes to handle applications such as online games where various game entities in the universe can be logically decomposed into discrete units and processes running on each node of the cluster. This provides a more natural and robust organization to the traditinal setup of a few massive servers, since if one server crashes, it may bring down large parts of the game universe. In our setup, if a node fails, it might affect one NPC at worst, which another node will take over in due time.

While our investigation has targetted the needs of games in mind, I'm excited about using them for sheer computation, since the cost/MIPS of a game console is far less than traditional mainframe, supercomputing, or even PC platforms, and we are in preliminary talks with some large Japanese universities to experiment with using the GameCube as a compute unit.

While I must admit I'm sort of biased :-), we believe that our GameCube makes a superior clustering platform compared to the PS2, computationally (higher CPU speed), physically (its smaller size and form factor, less heat dissapation) and financially (lower unit cost).

Our future game consoles will likely support clustering "out of the box", with expansion as easy as hooking them together, allowing games, such as FPSes, or AI-heavy games like the Sim* series, to seamlessly evolve with the greater "virtual" CPU and memory resources that a cluster provides.

Also worth considering: the Xbox

[mkro]

A Xbox is cheaper than the PS2 (An Xbox is about $150, according to OSDN Pricewatch), comes with twice the amount of memory, ethernet, and instead of buying a $200 Linux kit, you pick up a flashable, legal* mod chip for $25-$50. How the Emotion Engine compares to the Xbox P733 I have no idea, but I can't imagine the EE is that much faster.
Both The Xbox-Linux Project and Gentoox can provide you with a distro. For free.

Even if you're not planning a cluster, this is a good deal for a low-performance work station, or just a "media box", using Xbox Media Player, which plays most (all?) popular media formats, both music and video.

It's been repeated countles times that Microsoft are losing money on the console itself, and depend on the games to cover their expenses. Therefore, paying up for a Xbox and giving your money to MS isn't immoral as long as you don't buy any games.

See, it's a win-win situation :)

* I lost track of the current situation in the U.S., but in the free world (Read: Europe) at least the chips not using MS code is legal.

It's the trend of the future

[Ryu2]

Since 3-D graphics essentially is comprised manupulating vectors very quickly, the Graphics processors found not only in the PS2, but the latest PC graphics cards are now essentially very fast stream-based vector processors, and can be readily harnessed for general-purpose scientific computation other than graphics: particle, cloth, fluid simulations. The GPU replaces the CPU for computation, and texture or other video memory, with its much higher bandwidth and lower latency than system ram, is used as a backing store for data.

A lot of the GDC and SIGGRAPH 2003 papers focus not on graphics directly, but on scientific computations using the CPU. It's very cool, and if nVidia and ATI the like ever want to expand into a new market, they should build cards with multiple GPUs each, and sell them to the scientific community, or to non-realtime CG places like Pixar to accelerate their offline rendering.

This page has a good summary of the current research going on to make GPUs do stuff other than graphics. http://wwwx.cs.unc.edu/~harrism/gpgpu/index.shtml

Timex Sinclair Clusters

[leejor]

Reminds me of a nut case I met at a garage sale some 10 years ago. He was scavenging ZX80 Timex Sinclair in a effort to prove that clustered computing built around very cheap systems was the wave of the future.

I also complained about how he had been an EE for IBM who was not appreciated for his genius. He was very worried that once he released his ZX80 FrankenCluster, IBM would steal it from him due to his old employment contract.

Lee Joramo

Re:More info

[jean-guy69]

ahem... not exactly if you read carefully:

45MFLOPS, which is better than the 36MFLOPS obtained for the Intel machine using assembler that has been heavily optimized. If anything, this performance estimate is conservative in favor of the PS2, because our primary goal was a working assembly dot product in macromode and we have made no attempt to optimize the code. For example, our code uses only three of the VPU registers, and a speedup of up to 4x (the latency of the floating point multiply instruction) may be obtained by using all of the available registers.

so, using *unoptimized* ASM on PS2, PS2 is 25% faster that the intel machine using *heavily optimized* ASM.. and optimizing code would probably earn BIG performance gains (400% !?) on the PS2.

taking the sentence the the letter there is a potential of 500 % the speed of the PIII 600 on the PS2 for this particular calculation.