Massive $208M Petascale Computer Gets Green Light

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
The University of Illinois at Urbana-Champaign and its National Center for Supercomputing Applications said it has entered into an agreement with IBM to build a $208M, 200,000 processor system know as Blue Waters

Blue Waters is expected to deliver sustained performance of more than one petaflop on many real-world scientific and engineering applications. They will be coupled to more than a petabyte of memory and more than 10 petabytes of disk storage. All of that memory and storage will be globally addressable, meaning that processors will be able to share data from a single pool exceptionally quickly, researchers said.
 
I bet it will draw a few watts of power, will take a serious tower to hold it all and
most likely have a lot of wires. Wanna bet?? :rolleyes:
 
Wow, that was really dumb. You could build a petraflop computer using GTX 260s with only 1200 of those. And with the G200 architecture you get the same double floating point precision. When are the supercomputers builders going to wake up and start using GPUs instead of CPUs to build platforms that are highly multithreaded?
 
Wow, that was really dumb. You could build a petraflop computer using GTX 260s with only 1200 of those. And with the G200 architecture you get the same double floating point precision. When are the supercomputers builders going to wake up and start using GPUs instead of CPUs to build platforms that are highly multithreaded?

Wow that is a great idea!!! I will tell my execs about your awsome idea. :p

Sorry, but I have a feeling these guys know what they are doing.
 
Wow, that was really dumb. You could build a petraflop computer using GTX 260s with only 1200 of those. And with the G200 architecture you get the same double floating point precision. When are the supercomputers builders going to wake up and start using GPUs instead of CPUs to build platforms that are highly multithreaded?

A GTX is the last thing I want to base scientific study on that could involve millions of lives. :p

Also, you can't get 10 petabytes of high performance storage nor a petabyte of memory out of a gtx cluster :)
 
A GTX is the last thing I want to base scientific study on that could involve millions of lives. :p

Also, you can't get 10 petabytes of high performance storage nor a petabyte of memory out of a gtx cluster :)

lol. Well storage you could do the same, but on a CUDA based platform you'd get the full Petrabyte of memory. Remember, CUDA isn't like SLI and it's AFR implementation, it gets the full memory amount from each card, it wouldn't be 1200 cards with only 896MB each. (That would really suck!)

Now where you find a mobo with 1200 PCI-E 2.0 slots, or where you find a 250,000 Watt PSU.... that's another story :p
 
Wow that is a great idea!!! I will tell my execs about your awsome idea. :p

Sorry, but I have a feeling these guys know what they are doing.

A GTX is the last thing I want to base scientific study on that could involve millions of lives. :p

Also, you can't get 10 petabytes of high performance storage nor a petabyte of memory out of a gtx cluster :)


Actually you're both wrong and using GTX280's would be an awesome implementation, cheaper and faster. Basically it could be done for a fraction of the cost ..200million plus ...that's absurd amount of money.

I would bet that they have a budget that they have to spend a certain amount of money every year to receive additional funding the next year, if they don't meet that quota then they lose the amount the next year and get less to spend. A lot of goverment funded/operated buisnesses/organizations are this way, same would be for universitys and research institutes receiving grants and other funding.

This guy proves it can be done for about $900,000 U.S not including the storage....a lot less than $208million. :rolleyes:


http://helmer3.sfe.se/
 
But will it run crysis?

Probably not.

AOM said:
Actually you're both wrong and using GTX280's would be an awesome implementation, cheaper and faster. Basically it could be done for a fraction of the cost ..200million plus ...that's absurd amount of money.

GTX260s would be used, simply because they are much cheaper per flop.
 
The University of Illinois at Urbana-Champaign and its National Center for Supercomputing Applications said it has entered into an agreement with IBM to build a $208M, 200,000 processor system know as Blue Waters

Jeez you would think that they would at least hang a text to speach box on it and call it HAL 1000 (leaving 7999 revision numbers until 9000), I bet that would have got some press coverage.

Obscure movie references aside, who the hell wants to rewrite/modify every freaking app to CUDA compliance. That would cost more than 208 big ones most likley and lock you into this machine or even worse end up with 2 code trees to maintain. And by app I dont mean notepad I am sure some of the models they will be running have years and years of development in them and you dont fark with a couple of million lines of code unless you absolutely have too. This is one of, if not the finest computing centers we have on the planet, as mentioned above, I am certain they know precisely what they are doing and have very good reasons for it even with all the red tape feldercarb. Or not.
 
Actually you're both wrong and using GTX280's would be an awesome implementation, cheaper and faster. Basically it could be done for a fraction of the cost ..200million plus ...that's absurd amount of money.

I would bet that they have a budget that they have to spend a certain amount of money every year to receive additional funding the next year, if they don't meet that quota then they lose the amount the next year and get less to spend. A lot of goverment funded/operated buisnesses/organizations are this way, same would be for universitys and research institutes receiving grants and other funding.

This guy proves it can be done for about $900,000 U.S not including the storage....a lot less than $208million. :rolleyes:


http://helmer3.sfe.se/

Ok, I have seen a lot of crap on forums, but this takes the cake.

Talk is cheap.

Put a proposal together, that meets all the requirements of the UIU for a lot less money.

I will pass it over to my collegues over at the our Supercomputing devision and our Academic Sales Team. I am sure they will pay you handsomely for their new found profit.

/sarcasm
 
Actually you're both wrong and using GTX280's would be an awesome implementation, cheaper and faster. Basically it could be done for a fraction of the cost ..200million plus ...that's absurd amount of money.

I would bet that they have a budget that they have to spend a certain amount of money every year to receive additional funding the next year, if they don't meet that quota then they lose the amount the next year and get less to spend. A lot of goverment funded/operated buisnesses/organizations are this way, same would be for universitys and research institutes receiving grants and other funding.

This guy proves it can be done for about $900,000 U.S not including the storage....a lot less than $208million. :rolleyes:


http://helmer3.sfe.se/

Hey look, another keyboard scientist.

There is a reason why you spend big bucks on hardware like this. Keep the comedy coming.
 
Ok, I have seen a lot of crap on forums, but this takes the cake.

Talk is cheap.

Put a proposal together, that meets all the requirements of the UIU for a lot less money.

I will pass it over to my collegues over at the our Supercomputing devision and our Academic Sales Team. I am sure they will pay you handsomely for their new found profit.

/sarcasm

You mean you don't think it's a good idea to stick desktop motherboards with known problems and desktop graphics cards and then stack them?! I mean, it will only do some nuclear calculations or some biomedical or chemical reactions... no big deal if we get some numbers wrong or anything... lol :D
 
Ok, I have seen a lot of crap on forums, but this takes the cake.

Talk is cheap.

Put a proposal together, that meets all the requirements of the UIU for a lot less money.

I will pass it over to my collegues over at the our Supercomputing devision and our Academic Sales Team. I am sure they will pay you handsomely for their new found profit.

/sarcasm

Hey look, another keyboard scientist.

There is a reason why you spend big bucks on hardware like this. Keep the comedy coming.

You mean you don't think it's a good idea to stick desktop motherboards with known problems and desktop graphics cards and then stack them?! I mean, it will only do some nuclear calculations or some biomedical or chemical reactions... no big deal if we get some numbers wrong or anything... lol :D


I never claimed to be a scientist or that the link I posted was my idea. I'm glad you think that the insane amount of money being used is necessary to perform the task at hand on said hardware.

Money is basically thrown away everyday in world b/c the simple fact of people don't honestly give a damn. I'm not saying the idea this guy had is perfect nor what they would need exactly, but I'm pretty sure it could be done cheaper and I was just showing an example.

I also don't know anything about CUDA or the code necessary to make said programs operate/function as they need them too. I do believe that the same tasks could be performed quicker, cheaper and just as reliably if the time was taken to do so, but why do that when you can just blow insane amounts of cash on something less efficient b/c you're given the funds too.

I could go on and on about waste in this world, but don't have the time or energy to argue on the internet or discuss it today.

Here's the original article that I referenced too, in order to find the info from the original sources website for what it's worth.


http://www.hardocp.com/news.html?news=MzQ1NTQsLCxoZW50aHVzaWFzdCwsLDE=
 
You mean you don't think it's a good idea to stick desktop motherboards with known problems and desktop graphics cards and then stack them?! I mean, it will only do some nuclear calculations or some biomedical or chemical reactions... no big deal if we get some numbers wrong or anything... lol :D

G200 core hass double floating point accuracy. It's not a Pentium 2 (or was that P3? I don't remember)
 
All the people in this thread claiming that GPUs are somehow not up to "important" scientific tasks and calculations are directly dissing the Folding@Home project, and most of the distributed computing projects in the world.

Do you people really think that modern hardware makes calculation errors and then is unable to catch said errors?

Just because they are spending 200+ Million does not mean that they need that much money to build such a powerful computer. There are certainly less expensive ways to achieve the processing power - I think that the real challenge is the memory and storage interfaces.
 
You mean you don't think it's a good idea to stick desktop motherboards with known problems and desktop graphics cards and then stack them?! I mean, it will only do some nuclear calculations or some biomedical or chemical reactions... no big deal if we get some numbers wrong or anything... lol :D

Newsbreak: IBM has been found to have sold a fake supercomputer to the UIU.

Investigation into the Top Secret supercomputing centre reveals that behid this "super curtain" it was nothing more than a smart chinese girl typing answers on the console after solving the PetaComplex problem using a giant f***ing Abacus.

xin_400703202205761087325.jpg


The UIU declined to comment, but sources say they are happy with the results, as they are much cooler than other options.
 
Say bye, bye to the meat-sack and hello to our new machine overlords. What no Skynet jokes yet?
 
Back
Top