... that I deemed "utterly unnecessary" off-the-bat, but I was wrong.

 

Jesus, think of all that porn I could save on that thing...

 

Feh, think of how much porn you could watch...

If you have the money for a the computer, might as well pair it up with the right monitor(s?)

 

Excuse me while I change my underpants...

 

Totally pointless and unnecessary...

and yet so totally freaking awesome.

 

Finally. Now we just need someone to develop software for realtime, hi-res 3-D rendering... of porn.

Download the models, type in any two names... and watch them fuck.

Bring it on, future. Bring it on.

 

While a nice machine, (Cray makes, hands down, the coolest looking supercomputers), the quote about being highest performing computer using standard office power is hyperbole. Just a few hours ago, I used an 85 Tflop/s supercomputer that was just a COTS (commercial off-the-shelf) cluster of networked Opterons. And it's quite a ways down on the Top 500 list; several other systems higher on the list are COTS machines.

The economics of supercomputing is such that it doesn't pay to do much custom hardware development anymore. Even the Roadrunner supercomputer, the only machine to break the Pflop/s barrier, is essentially a bunch of Opterons connected to 12240 IBM PowerXCell 8i chips (a modified version of the chip in the Sony Playstation 3).

 

He probably should've used the phrase "consumer computer", Pers. You would be the one with access to such high-end machinery, given your job.

Bastard I wants mah 'Minority Report' touch/gesture computer.

 

I will publicly declare my love to the first person who invents one of those. Imagine the porn you could play on it!

 

what the hell do you need a computer that powerful? What are you going to filll it up with? lol

 

I will download God.

 

Fuck God, I'll take porn

 

Why the hell would you do that? It take up memory that should be used for porn!

 

I would go beyond simple tasks like that. I would download everything. I run out of space? Buy another 4TB hard drive, cause if you can afford one of those in the first place you can easily afford another.

Eventually I will have the biggest collection of everything in the world. Books, movies, porn. It would all be mine!!! (I am not cackling like a maniac here, I swear!)

 

Cray makes sweet looking hardware.

I wonder, though, if 8 nodes of 64 GB per node and 4 TB of internal storage becomes obsolete by next year, or something:

http://www.technologyreview.com/computing/14742/?a=f

This article talks about InPhase Technologies' "holographic memory," a way of storing memory in 3D blocks of space instead of 2D linear disk systems...

Then we could cream freaking petabytes instead of terabytes into the same amount of space the Cray has...

*drool*

[Edit: Oops, looks like the article is a little old. Where's my 3D memory storage? Gah.]

 

Glorious. No other word for it.

 

Disks are going away soon anyway because mechanical devices are both more power-hungry and failure-prone. You can already buy machines such as my MacBook Air that are 100% solid state storage. The only moving part inside is the fan.

High end supercomputers have tens to hundreds of thousands of cores, depending on how you count them. In ten years, expect billion-way parallelism. The file systems on the machines are enormous, ranging in the hundreds of terabytes to petabytes today, with what amounts to single-purpose supercomputers attached to them to break data into metadata and route it so as to avoid bottlenecks. Unfortunately, these systems still rely on mechanical technology, which is failure prone and power-hungry on the very large scales we're talking about. Less so in a handful of years, when everything goes solid-state and, ultimately, 3D.

Not only is storage going to be 3D in the relatively near future, but processor vendors are looking to layer silicon in an effort to locate memory closer to processors, make the physical size of chips as small as possible, and minimize energy consumption. Today, some of the biggest supercomputers (e.g., ORNL's Jaguar) approach ten megawatts of power; even the fastest supercomputer, LANL's Roadrunner machine, draws more than 2 megawatts. It's impractical to envision supercomputers that use beyond 50 or so megawatts. (At 20 cents/kwh, it'd cost $10k/hr for power to run the machine; this works out to about $90M/year for power, nearly the cost of a machine like Roadrunner at $110M).

Simply put, the speed of light is too slow and memory bandwidth and latency will be the bottlenecks to performance from here on. On most apps, the processor spends most of its time stalling as it awaits data; cache sizes and protocols aren't standardized enough to make it worthwhile to design to the metal in any but the most dedicated apps--video games/rendering engines taking advantage of GPUs, e.g., or special-purpose science apps. Without taking advantage of 3D placement of components, application efficiencies will continue to ratchet downward as machine size increases. Unchecked, we'll reach a point of impracticality somewhere around 10 to 100 Pflop/s for even the most streamlined of apps.

 

Wow Pers that sounds awesome, I am quite looking forward to computers now in the next decade or so. I am uncertain why you said 10 years though for the 3D thing, is that actually feasible?


Also

:awesome