[kscherer]

Note to Mac users: M$ is building the world's fastest cumputer. According to the MS specks for XBox2 (see below) the unit processes FP at 1 Teraflop! Does this mean that 70 of these $300 gaming machines will constitute a supercluster capable of overpowering the Earth Simulator?

Am I lost and not knowing my tech terms, or is M$ BS-ing us?

http://www.xbox.com/en-US/xbox360/factsheet.htm

[sunrain]

Well, it's a measurement of calculating speed, so I guess it depends how many operations they're measuring.

This quote from the Wikipedia should illuminate the marketing spin that's likely at play here.

Pocket calculators are at the other end of the performance spectrum. Each calculation request to a typical calculator requires only a single operation, so there is rarely any need for its response time to exceed that needed by the operator. Any response time below 0.1 second is experienced as instantaneous by a human operator, so a simple calculator could be said to operate at about 10 FLOPS.

[kscherer]

Certainly. Micro$oft's direct claim, according to the tech specs, is "Overall System Floating-Point Performance = 1 teraflop". If my understanding is correct, this is the same unit of measurement used to calculate the speeds of the world's fastest computers, of which the VT cluster is a member. Apple's claims for the G5 are somewhere in the neighborhood of 10 gigaflops. Interesting.

[Gargoyle]

That is very weird. I can't believe that each core is running at 333 gigaflops!

I think they are counting every possible floating point operation in the system and adding them up! ie. each core + gpu.

[chucker]

First of all, they use single precision for this claim. The Top 500 list uses double precision. Big difference.

Second, what people keep forgetting is how specialized these Xbox CPUs are. Sure, you could probably hack your way through, make a few drivers and get Linux or even Darwin to boot on them, but they're not optimized for that. They're optimized for their purpose: a game console.

The Xbox 360 is not the world's fastest computer (as in personal computer), because it is neither a personal computer, nor a particularly fast one.

[kscherer]

Nor am I claiming it to be so. It just seems a bit odd that this machine is capable (?) of such speeds.

By the way, I haven't got a clue to the difference between single and double point!

[Brad]

The difference is that the variables are twice as wide, moving from 4 to 8 bits each. That means, among other things, using twice as much bandwidth.

It all depends on the processor, though. In many cases, single precision is faster to calculate than double. Some will perform single and double precision at about the same speed; some will actually perform double precision faster. I believe the older x86 processors (maybe they still do?) actually calculated both as double precision, converting single to double as a variable was moved to the register stack. Of course, only assembly-writers need to remember that because compilers should take care of the conversion ahead of time for you.