Printable View
If Fermi powers that one:Quote:
Originally Posted by Hu1kamania
- Sir, we have a problem. the 5000 Fermi videocards we bought for our supercomputer...
- ...... yes? What's the problem?
- Wooden scre-
- DAMN YOU JENSENNNNNNNNNNNNNNnnnnnnnnnnnnnnnnnnn:::::::.:::: :...:..... .... . . (slowly disappearing scream)
What do you have against wooden screws? Thats just nvidias way of going green.Quote:
Originally Posted by annihilat0r
Exactly.Quote:
Originally Posted by Spawne32
So U.S. hardware hacking U.S. hardware. Kind of ironic.Quote:
Originally Posted by Utnorris
Maybe not the US computer, but US encryption perhaps. Digital Fortress anyone?Quote:
Originally Posted by Utnorris
The new rage is Aion, man they are every where. Login with a new character and get slamed with gold seller spam. Bots and sellers non stop.Quote:
Originally Posted by rogueagent6
what the point for such server farms to have massive GPUs running for?
BIG Bro observation?
i was talking to a friend last night about enermax's chinese website, its hilarious how every product they have has to be an "anti-us" product :rofl:
woah crazy.. imagine if they changed the 4870x2 to 5970 gpus... but imagine having to physically swap all 5,120 gpus out :eek:
It will be used for national weather forecast center. The setup uses Intel Xeon processor, 4870x2, and a specially designed OS. According to some reports, when using NV's GPU during early stage testing, 4870x2 only trailed behind for 2% in efficiency in certain test, and others pull ahead. Since 4870x2 is cheaper, we all know what the outcome would be.Quote:
Originally Posted by Serra
Im sure someone there is willing to do it for .05 an hour ;)Quote:
Originally Posted by eva2000
I think bill gates' desktop is of similar performance no? :rofl:
Do you have a link, or more details about this "specially designed OS"?Quote:
Originally Posted by duckbillclinton
Do I see it right?, is it 2 CPU + 2 GPU on each blade in this picture?
http://i301.photobucket.com/albums/n...ard/radeon.jpg
looks like quad cpu + 2x hd4870x2
And 16 DIMMs per plate. So they are using Brook+, that makes it almost impossible to switch to Nvidia hardware. Why didn't they use OpenCL and program the whole thing from scratch? I assumed that should be the case for mainframes like this. Two years in development and they use the standard API? With so many people they only had one guy assembling the cards? :rofl:
I'm not sure on the figures about processing power against power consumption and price. Is a GPU better for crunching in power consumption too?Quote:
Originally Posted by the article
The idea of using the GPU as a "accelerator to make the computer run faster" is not new. But then as you mentioned, it would be more natural to use OpenCL (or other proven and standard API), but they have chosen to use a "specially designed OS".Quote:
Originally Posted by mursaat
This approach is totally "off" (from the "normal" approach for using the GPU as accelerator). If I'm not mistaking, it seams they are not using those GPUs for delivering processing power (GPGPU), at least not as their main function. And it makes me wonder if they are using those 2GB fast GPU-memory as some kind of cache?. I guess the answer (which is really interesting) lies in what is going on with that "specially designed OS".
Yeah, AES be dare xdQuote:
Originally Posted by twwen2
I updated the OP with some info about processors and graphics cards used, and also performance, cooling used, and its computing purposes.
It is confirmed that it does use 4870x2.
why are they saying it is a petascale supercomputer? it only gets 560 teraflops in linpack.
Can you run linpack on GPUs?Quote:
Originally Posted by Chumbucket843
Didn't think so.
as far as i am concerned it doesnt matter that the gpu's cant be used. you either can run the benchmark or you cant.Quote:
Originally Posted by Solus Corvus
They didn't build a supercomputer to benchmark linpack all day. So even if it couldn't run linpack at all, your point is still moot.
no one builds a supercomputer to run linpack. linpack is blas so its going to reach peak flops better than anything else and they arent getting in the top500 with out it. you dont know if it will be faster running the software for its intended purpose so how can you claim it will reach over a petaflop?Quote:
Originally Posted by Solus Corvus
I'm not the one claiming a petaflop, the designers are. And they obviously aren't claiming over a petaflop in linpack. They designed it to be over a petaflop for a specific usage. Does it actually reach over a petaflop in that application? I don't know and I'm willing to bet that you don't either. But if it does then it's definitely a petaflop supercomputer, regardless of whatever it scores in linpack.
You seem like a smart person so I'm kind of curious why you are so hung up on the linpack thing. Linpack isn't the only measure of flops in the entire universe. Sure, linpack is standard and the score won't get them into the top500. But they didn't build it to benchmark linpack or to get into the top500 - they built it to perform a certain type of computation and how good it is at that is the important thing.
ahaha :ROTF:Quote:
Originally Posted by annihilat0r
it will probably be used in a one time winner take all supercomputer crysis tournament, Fatal1ty vs. chinese gold farm.. or whoever is their best.
on topic though i'm sure if this can help with early warning weather systems then that's great.
1.000.000.000.000.000 so in other words 1 peta flop... impressive... :DQuote:
the computer can do more than one quadrillion calculations per second at its peak speed
so while nvidia talks gpgpu, ati makes it happen? :D
im curious though, how are they going to use the gpus?
open cl?