What is Digg?71 Comments
- inactive, on 10/11/2007, -7/+37They didn't need to do that. If they wanted Open Office to run fast all they had to do was turn off Java
- xerus, on 10/11/2007, -1/+27Man this hits close to home. Just the other night I was driving back from the office and right in the road 2 Petaflops jumped right out in front of me. If I didn't swerve at the last second I probably would've hit them.
- BirkBum, on 10/11/2007, -3/+21I bet it has a 1.21 jiggawatt power supply!!!
- stouffer67, on 10/11/2007, -6/+24ALL RIGHT 2 PETAFLOPS....... what the ***** is a petaflop?
- spuffe, on 10/11/2007, -2/+11About time that Sun Microsystems came back in the game :-)
- xShad0w, on 10/11/2007, -0/+7I'm waiting for the exoflop, go oceans 13
- junk2006, on 10/11/2007, -0/+4Imagine playing pong on that thing; there must be no loading time
- ohnoitschris, on 10/11/2007, -1/+5To the non-geek, bits; bytes; and flops will always be funny words.
- noseeme, on 10/11/2007, -1/+5Yet another digger that doesn't seem to know how Java and the JRE work...
- xShad0w, on 10/11/2007, -1/+5Correction: Maybe this will run vista well
- Ramble, on 10/11/2007, -0/+3Everyday I'm amazed by the stupidity of some people.
- inactive, on 10/11/2007, -2/+5Hmmm, maybe this will run Vista?
- geminitojanus, on 10/11/2007, -0/+3They were pretty big in the market along with Cray in the late 90s. Sun used to be one of those giant do-it-all Mega-corp computer companies before the dot bust.
- acdcfanbill, on 10/11/2007, -1/+3a thousand trillion floating point operations a second. ...time 2
- hodrige, on 10/11/2007, -0/+2Photos: http://blogs.sun.com/simons/entry/petascale_unveiled_photos_from_dresden
- Error601, on 10/11/2007, -2/+4Back into? Have they ever been in the supercomputer market? Their biggest stuff have also been more in the big-ass database server market rather than big-ass number crunchers.
- mrASSMAN, on 10/11/2007, -0/+2Well those are two different situations. The PCs aren't "faster" than what you want because we're just pushing the limits of what can be mass-produced cheaply to consumers.. it's actually still leading Moore's law, so nothing is slowing down.
As for the internet connection speed.. that's just due to lack of government initiative, and the monopolized telecom industry in general. It could really all change with a government that paid more attention to these matters. Here's hoping for 2009. - superyounan1, on 10/11/2007, -1/+3wow, you could probably run the java virtual machine on that and not worry about performance
- mrASSMAN, on 10/11/2007, -0/+2I'm surprised by how many here don't know what a FLOPS is. I thought we were all geeks..
- Nicksname1, on 10/11/2007, -1/+3ZING!
- zackkitzmiller, on 10/11/2007, -0/+2You'd like to think that.
- mrASSMAN, on 10/11/2007, -0/+2Not simultaneously.. it's the number being done per second.
- Nicksname1, on 10/11/2007, -0/+2Oh you gotta run the PSU at 1.21 jiggawatts...No More, No Less!
- anonym41414, on 10/11/2007, -0/+2I think you need to come to a deep and intimate personal understanding with the phrase "I/O-bound operation."
- rhabd0mancer, on 10/11/2007, -0/+1Sun bought out Thinking Machines back in the '90s.
- catalysis, on 10/11/2007, -1/+2It must be obvious day at camp stupid.
- pascalosti, on 10/11/2007, -0/+1I want a petaflop
- geminitojanus, on 10/11/2007, -0/+1PS3 clusters haven't been made, and the CELL processor hasn't even made it into very many machines total yet, though I bet someone somewhere is planning on buying a thousand node CELL cluster.
Secondly, Sony is selling the PS3s at a loss, and whatever application you'd want to build this supercomputer to work on wouldn't need access to a next-gen DVD drive (and would need a much better network infrastructure like Myranet or Infiniband or GigE at least). CELL Cluster nodes would cost around two grand a piece, the total solution would cost a lot more. (For example, here's a "workgroup"-sized model of a Cell cluster: http://www.linuxdevices.com/news/NS3591350722.html ). - NinjaBoy, on 10/11/2007, -1/+2http://en.wikipedia.org/wiki/Petaflop
- longbow486, on 10/11/2007, -0/+1bingo
http://en.wikipedia.org/wiki/Flops - tuzziel, on 10/11/2007, -1/+2"Petaflops" watch it flops. (Am I the only one that is not inpressed?)
- mrASSMAN, on 10/11/2007, -0/+1well technically it's "gigaflops" and "teraflops" and so on.. but yeah that's the jist of it. but stouffer might have actually been kidding anyway..
- geminitojanus, on 10/11/2007, -0/+1And SI for quadrillion is "peta-".
- Error601, on 10/11/2007, -1/+2Scale of 2 petaflops - Most estimates (googling around some research papers)are Toy Story could be rendered in real time with 1 gigaflop of processing. This is 2 million times faster, so it could render the entire movie 50 times in one second.
- inactive, on 10/11/2007, -0/+1Good intro alpha. But seeing all these noobs asking such basic questions ( with a serious need of /sarcasm tag if joke intended ) Going into advanced math algorithms might not be the best choice.. "For" loop would be better.
Also, you skim a serious topic: FLOPS as a merit of performance; much like "The megahertz myth" higher numbers are not the best (from the good old days of Apple using IBM PPC chips).
Even if PC1 can do 20 FLOPS and PC2 only does 10 FLOPS, if PC1 takes twice as long to load from memory as PC2, the performance is essentially equal. What about integer math? What about branching? (I am getting at IPC or CPI for those who know what they mean)
It all depends what the computer is used for (ergo benchmarks). Computer architecture is quite tricky and no one architecture is "best for all," yet 2PFLOPS is absolutely impressive. - PaulOwen, on 10/11/2007, -1/+2RIKEN's MDGRAPE-3 achieved 2 Petaflops this time last year. I know it's not a popular fact with US manufacturer flag-wavers, but a fact nonetheless:
http://en.wikipedia.org/wiki/MDGRAPE - geminitojanus, on 10/11/2007, -0/+1Quite frankly, there are two sets of problems; problems that are I/O Bound, and problems that are CPU Bound. This is an I/O Bound supercomputer. Rendering Toy Story (or anything really) is a I/O Bound operation, meaning that while it would be capable of doing the task, it wouldn't do so efficiently.
A much better fit would be a CPU-bound cluster for the I/O bound application like rendering. This means extremely fat memory interconnects and single- or dual- CPU nodes instead of Octo-core nodes with cross-bar memory. The I/O Bound cluster (like this one) is a much better fit for applications like molecular dynamics, genetics, less so fluid dynamics (anything where you can keep intermediate variables within the machine or better yet within the CPU cache; this machine is really good at doing "embarrassingly parallel" work). - Error601, on 10/11/2007, -0/+1The I/O is distributed just like computation. There would be no point having all that crunching power and no where to put the results.
- inactive, on 10/11/2007, -0/+1Today's graphics do just that.. graphics: they have great GFLOPS numbers (350 on NVIDIA's G80 ) due to the large amount of ALU space working in parallel.
Data that passes gfx processors is rarely reused and it represented in many pixels; this leads to massively parallel opportunities like having a lot of simple (read FAST) units for computation (float and integer, hell both!)
Since data is not reused often, fat ass pipelines can be created with no worries with branches. The last thing I can remember is that gfx (and scientific code) has few data dependencies which again leads to fast chips with fat pipes.
Contrary to gfx, general code is all the opposite to this: lots of code jumping, code and data reuse, and not to fail to mention I/O ( It doesn't belong here really except in respect of how the CPU responds to a call ).
Moore's law HAS slowed, took a right hand turn years ago. The "power wall" stops _single_ CPUs from getting any faster/watt and the "memory wall" has also crippled the computation potential due to such slow access.
well is this long enough? should I go on? - Error601, on 10/11/2007, -0/+1It might be a good market for them. The PS3 is a totaly unworkable form factor for a data center, but they could slap those processors into a box designed for number crunching.
- SomaSynth, on 10/11/2007, -0/+11 gigaflop to render Toy Story in real-time? I think you might need to rework your calculations a bit.
- inactive, on 10/11/2007, -0/+1C'mon digg community... I thought that you where smarter than this. Not knowing scientific units, basic meaning of a flop, lack ANY sort of serious discussion whatsoever.
- Error601, on 10/11/2007, -0/+1The failure recovery element isn't new for Sun. The kernel has been capable of failing out basically any part since I think the 2.8 release. The last I really messed with it, we were still on 2.6, and it could fail ot some parts but not everything. The promise at the time was we would be able to pull and replace any CPU or I/O board in the case without shutting down the OS.
- satx, on 10/11/2007, -0/+1Jigga who?
- cantoral, on 10/11/2007, -0/+1The people that brought us SPARC may again be ready to leap forward. Let us see what Andy Bechtolsheim has in store for us.
- alpha754293, on 10/11/2007, -1/+1FLOP is an acronym for floating point operation. It is a measure of computational work done by the giant/over glorified calculator commonly referred to as the computer. FLOPS is the rate of computational work done, i.e. floating point operations per second.
For example, in matrix algebra, Gauss Elimination requires a cubic polynomial (i.e. O(n^3) or more specifically, approx. n^3) number of floating point operations for an n-order matrix (i.e. square matrix of n x n dimensions). (Likewise is true for LU decomposition). While the specific implementation of a FLOP vary between CPU architectures, the requirements are (loosely) universal; hence it is the most portable method of measuring system performance.
Also hence why the Top500 list exists, that measure performance based on LINPACK and HPL LINPACK. Not often used for general public largely because for some reason, FLOP always requires an explanation, but GHz (often) doesn't. *shrug*
Wiki:
FLOP, Gauss elimination, matrix algebra, LU decomposition, sparse matrix, LINPACK, HPL LINPACK, FORTRAN, HPC, Top500
Additional references:
Top500 Processor Architecture list (and previous detailed publications) on specifics.
IBM Journal of Research and Development - PaulOwen, on 10/11/2007, -1/+1Yet another digger that doesn't seem to know that not "understanding" Java doesn't excuse its sluggish performance.
- specialK16, on 10/11/2007, -1/+1Most performance issues are due bad coding, specially when it comes to Java. (Not saying Open Office is badly coded, as a matter of fact, it runs great on my computer with Java on).
- SomaSynth, on 10/11/2007, -0/+0Whatever your sources are, they're either inaccurate or perhaps were referring to a raster approximation, much like that time Nvidia claimed the Geforce 3 could render Final Fantasy in real-time. One Teraflop I might be willing to believe.
- jonathono2000, on 10/11/2007, -1/+1Oh the irony.
-
Show 51 - 71 of 71 discussions
The Digg Toolbar for Firefox lets you Digg, submit content, and keep track of Digg even when you're not on the Digg site. Download the official 
© Digg Inc. 2009
— Content created and posted by Digg users is