Kinda missed the point, I didnt mean spec it out to the max, which yeah is about 24k. (Granted that gives it 32g RAM 1.2 TB Solid state HD, 2 x 3.2Ghx and kick arse everything else)

I ment spec it out as much as you can with 10k.

From what I can tell specing a computer out on the Mac site and the site NW linked to, the Mac site gets slightly better specs, although it doesnt (to my knowledge) come with the kick arse cooling systems and housings those computers come with.

Basically Macs a slightly (like 10-15%) more expencive than most PC's, But they are better built, more reliable, and 7 fucktonnes cooler.

 

Lol. Injun = mini-mod/voice of reason. Stop slathering us infidels with the cuntjuice that is your Mac hate, m'kay? Like shit ever really stays on-topic in a DLP thread...lol.

In all seriousness, this thread ended with Nuk's link. Anything better would be too expensive or a custom job - which would end up saving money. Can't have that nonsense.

 

All I can say is go with a Solid State Hard Drive and some SLI'd top of the line GPUs with about an 800+ watt PSU. Of course get a great processor, 64 bit XP Pro, and 4 sticks of 1 gig ram. After that, just add a fuck load of cool shit and a bad ass cooling system and you have happiness. That's what I'd get if I had about 2-3K on a Computer. Seeing as you want a hypothetical 10k cpu, I don't even know where the fuck to start. The solid state, probably.

 

No. I can't begin to list why that's not a good idea.

 

merg, why not? I'd love a fucking computer like that.

 

The vid cards could be upgraded to the nVidia 1.5GB cards, three of them running in SLI.

RAM could be 8GB sticks instead of 1 GB sticks

256gb SSD for OS / Games
3x 1TB for storage

The new Intel processors i7's on the nephylm architecture.

Fiber card for FIOS intergration
Coolin/case etc

wireless card.

Hell, for 10K you're looking at low-end Super-computers. Where the speed is measured in teraflops.

 

If you're at all curious about supercomputing, there's a bona fide subject matter expert who's a regular poster here.

 

Oh? Who's that?

 

Perspicacity, yeah?

 

Think it's Irene actually.

 

Yeah. (filler)

 

Depending on just how hard-core your programming chops are, you could go with an IBM QS-22 blade, which can be gotten for under 10 grand. I could probably put together a GPU-driven system that's got more floppage for the cash, but it'd suck down huge power and have other issues. (Programming in CUDA, e.g. Ugh). Besides, you can't beat IBM's quality engineering--their Cell blades are as well designed as anything I've seen in supercomputing. These blades are what the world's fastest supercomputer uses.

If I tally up performance from the PowerXCell 8i chips alone (there's two on the board; there should be a pair of Opteron cores as well, but I didn't see them mentioned on the quote) and ignore the performance of the PPU (a PowerPC chip on each Cell), I get 3.2 GHz * 2 chips * 8 cores/chip * 4 fused-multiply-adds/clock (i.e., 4-way SIMD in single-precision) * 2 floating point ops/FMA to get 0.410 Tflop/s theoretical peak performance. This puts you solidly in "home supercomputing" territory at a power cost that's comparable to or less than that of an AMD Barcelona desktop.

Real-world performance is going to be somewhat lower. Linpack will probably get 40 to 45% of theoretical peak because they're running in double precision and only get half the throughput of single precision with the SIMD. Some specialized single-precision apps might push higher than Linpack--I think I could build a lattice Boltzmann hydro solver that could pull around 300 Tflop/s; I know of one artificial vision kernel that would pull about 60% of theoretical peak. The app from the code team I lead (a plasma particle-in-cell simulation code) would pull around 40% of Linpack for real-world problems.

The thing you'd want to do with such a system is get the fastest DDR ram you could get--EIB (bus) speeds scale linearly with memory speed and feeding the cores is going to be your biggest worry. Our app is memory bandwidth starved; we're fully subscribing every DMA channel and simply can't get the data to the SPE cores any faster. (For what it's worth, we pull 374.25 Tflop/s on all of Roadrunner--not bad for an app with, on average, 4.0 ops/word and random access to memory in the inner loop; there's a reason we were Gordon Bell Prize finalists last year).

 

Out of interest, how practical would such a computer be for home use? I mean, could you play games on it, despite its lack of GPU? Could you even install Windows?

 

With the Cell, the GPU is irrelevant. You'd probably run a flavor of Linux on it for an OS. (Roadrunner uses an incarnation of Fedora).

Supercomputers are built with apps in mind, not general purpose usage. (For general purpose stuff, just buy a cheap Mac or Linux box and connect to it). This machine would scream on apps specially designed to take advantage of the Cell's processing power. (Rendering, scientific apps, folding at home, etc). Making such an app is no mean feat, which is the origin of my first sentence in my prior comment. It took us about a year to redesign and tune our application to do so.

Of course, you could always just buy a bunch of PS3s and connect them with a fast network. This would give you the highest flops per dollar I can think of, especially now that Sony has dropped the price on their consoles. As a scientific supercomputer, it'd be very limited, though--you only have half a gig of RAM per Cell BE; you have no double precision SIMD capability; there's no IEEE-compliant rounding of any kind; any double precision usage would introduce huge stalls into your execution pipeline, etc.

 

Do you hurt us cause you love us?

 

Perspicacity, I thought you had a PhD in physics. I didn't think that supercomputing would be your thing. But I can see how having such a degree would come with supercomputing knowledge. Friend of mine is doing his PhD research with CERN on the LHC project (ATLAS) and he has to know about this kind of stuff in order to write the programs to do the physics stuff.

That being said, A good deal of what you said went over my head. I'm just a lowly Arts major (Politics, Philosophy, and Law) with a hobby in Physics theory. Nothing beyond that. I know how to use a computer, but not that kind of computer. Hell, I can't even be bothered to properly learn any unix platform yet.

 

I do, but my field is plasma physics. Plasmas tend to behave so nonlinearly that interesting problems need computer simulations to make sense of what's going on. While I like to think my analytic chops (think pencil and paper) are reasonably sharp, it's pretty much impossible to get a job without being able to compute.

Granted, I may have taken that to an extreme in the past couple of years...