From PlayStation to Supercomputer for $50,000

New York Times ^ | 2003-05-26 | By JOHN MARKOFF

[Jack Black]

Hmmm, wonder what SCO thinks about this project? Actually I think MS is actively fighting it, going as far as to change the hardware of the box in mid run to make it more difficult.

[Dan from Michigan]

Anyone remember Intellivision? TRS-80s?

[Jack Black]

Sony is much more suportive of Linux than MS. (Duh!) They will sell you everything you need for $199. The Xbox is a stronger platform to start with though. It has a faster CPU and a hard drive.

[RightOnline]

Truth is, there isn't a single thing your SGI Origin can do that I couldn't do with an Intel-based workstation for a fraction of the cost.........and a helluva lot faster. That's a fact.......and also why SGI is scrambling in the dungeon stock-value-wise. They used to be great, but like Intergraph and others before them, got caught napping and failed to capitalize on industry trends away from proprietary UNIX platforms on the desktop. I regularly destroy UNIX workstations with sub-$4000 Intel-based workstations running Linux.........it's not even close any more.

[Bogey78O]

I took a look at it. The Asus has support for S-ATA drives. I was thinking of going that route eventually. So I orderered one.

Does 176$ including shipping sound fair?

[RadioAstronomer]

Great price :-)! mine was 225$

[Bogey78O]

Should I get the 800Mhz bus chip or is 533Mhz good enough? Can a difference be detected?

[HiTech RedNeck]

But it runs Windows, right?

[RadioAstronomer]

The 800Mhz FSB will run the memory at full speed. I think it makes a difference :-)

The 2.8Ghz 800MHz FSB with Hyper-threading is ideal. Not so expensive that it breaks the bank. The 3Ghz processors are expensive. I bought the 3 Gig before I knew that Intel released the 2.8GHz with the faster FSB.

[RadioAstronomer]

Truth is, there isn't a single thing your SGI Origin can do that I couldn't do with an Intel-based workstation

You are comparing a multi CPU Origin 3800/3900 cluster supercomputer to an Intel based workstation. Hmmmm... yea sure!

[Bloody Sam Roberts]

From those crappy Atari's we all bought 15 years ago

Hey! I love my Atari 800xl!

I still have it set up with two 1050 5¼" floppy drives daisy chained, ready to run any one of 400 titles of the greatest software from the '80s.

[Hatteras]

"Anyone remember Intellivision?"

http://www.intellivisionlives.com/

My brother and I used to have knock down, drag outs after a few heated laps on the racing game. We found a way to cut through the woods to other tracks. It's all a blur now, that was a long time ago. Seems to me we also found a way to get the hockey players to fight...

[Frumious Bandersnatch]

On the other hand, give a programmer 100GB on the hard drive, 512MB of RAM, and 2 MHz of processor speed, and what do you get? Applications written in Visual Basic.

Barely...

[Frumious Bandersnatch]

The definition of a super computer (as defined by IBM) is (I believe) 1,000,000 mflops/sec. This means that your average computer could be probably considered in this category nowadays.

[Frumious Bandersnatch]

He's may be correct. OTOH, with the SGI you can get something done in minutes which may take months on his computer.

[=Intervention=]

I agree. This isn't capitalism. It's government spenind our money on a dubious scientific value. In other words, the usual waste and fraud. (Don't get me wrong. You're free to hook up however many PS2s you want -- just don't use my money to do it and call it "science").

[2 Kool 2 Be 4-Gotten]

There are, I think, all kinds of misconceptions and misperceptions in this article and the discussion thread.

Sure these guys built a "supercomputer" out of PS2's as long as you define "supercomputer" as a parallel architecture that executes so many floating point operations per second.

And sure Iraq ordered a bunch of PS2's that they were going to cobble together in some similar way.

But let's get real for a second folks. Aside from the Gee-whiz aspect of all this what has really been achieved? And more to the point, what real-world tasks is some rogue nation likely to be able to accomplish with this raw computing power. And what are the limitations of this science project?

I think that there are some real limitations on this architecture. Amount of memory is mentioned in the article and that has to be big. Also speed of the memory, and overall bandwidth of the bus as well as what type of cache coherency is implemented.

Assuming you could solve all of the above memory problems, (amount, speed and coherency) then you're on to your next big problem and that is software. Raw MFLOPS does you little to no good. You need custom codes that have been tuned for the particular machine in question. Ditto for the compilers that are going to compile these codes.

Let's not get carried away here folks - sure this is a neat science project, but Dr. Evil is not going to be able to unwrap 1000 PS2s from their shrink wrap, buy a cheap Ethernet switch off the internet, and have a computing platform that is going to enable him to burrow through to the center of the earth with a ""LASER"". It'll take a little more hardware and software expertise to accomplish that.

[RightOnline]

No, I'm comparing an Origin system to an Intel-based workstation. I didn't say I was comparing clusters to clusters........but that would be fun as well. :)

No, SGI lost any supremacy on the graphics desktop they ever had quite a while ago. The only advantage.......the ONLY advantage.........a UNIX-based workstation has any more is memory addressability. Certain apps. need to deal with very large datasets, and Windows is pretty much limited to 2GB of memory and Linux can be stretched to just under 4GB of memory on IA-32 architecture systems. However, that will change and change fast with x86/64 processors.

FYI, I could build a system right now with a decent sized Linux cluster........with IA-32 systems, mind you.........and the proper "hard-wired frame buffer" (not yet commercially available) that could mop the floor with a $1 million + Onyx with beaucoup pipes at a fraction of the cost.

[tortoise]

The only advantage.......the ONLY advantage.........a UNIX-based workstation has any more is memory addressability.

Huh? Only on low- to mid-end Unix workstations, and even then only for the narrow range of high-performance applications that only tax the CPU and no other part of the system. For the rest of us who don't do ridiculously CPU-bound computation, the overall bus and memory architecture performance (both in terms of latency and bandwidth) have a far greater impact on application performance than the CPU. While many people talk about GFLOPS as a metric for supercomputers, most people actually requiring general purpose supercomputers use STREAM type metrics that profile memory and I/O performance to evaluate these systems. I would remind you that even the smaller x86 SGI boxes use a crossbar switch rather than a bus for I/O.

Sure, for rendering x86 gives killer bang for the buck. But rendering generates a very atypical system resource load that doesn't apply to most other applications you need a beefy box for. On high-end x86 systems, my CPUs mostly sit idle waiting for data to squeeze through the memory bottleneck. AMDs HyperTransport will be a big step up in this regard and will help narrow the current chasm between high-end commodity x86 and high-end Unix systems.

Real Big Iron non-cluster supercomputers are notable more for their I/O and memory bandwidth/capacity than their CPU capacity. When you can find me an x86 motherboard that will give me anywhere near 25 Gbytes/sec of memory bandwith, you let me know. I'll be the first one in line.

[Richard Kimball]

Hey! I love my Atari 800xl!

XL? Newbie. I had the original 800 with 48k of RAM, 300 baud modem and Atari Basic.

I never did get used to programing without line numbering.