China successfully developed "Milky Way One" Supercomputer

[Chumbucket843]

You know that theorically RV770 has more computing power than G200?

you know that in the real world a 600 gigaflop gtx280 beats a 1.2teraflop 4870. comparing flops across architectures is almost as bad as comparing clock speed. ATi has a "markitecture" just like intel did.

[annihilat0r]

"in the real world" doesn't mean "in Crysis" for the Chinese supercomputer users, though

[Jamesrt2004]

Wonder why they used 4870x2 if GT200 is really better for GPGPU.

Fake? LOL. It's a live video made by CCTV!

mm ati have better gpgpu perf actually check dp benchmarks not nooby stuff like f@h

[Sam_oslo]

what are you on about?

the programs these guys use would be looking for more teraflops, not ease of programming, the 4870X2 has over 2Tflops, even the GTX 295 has nowhere near that number.

You know that theorically RV770 has more computing power than G200?

You are both right, but the API for converting that power to computing power (GPGPU) is still "crappy". CUDA would do a better job, for sure. Those 2Tflops is good for games, but can't get converted to computing power easily, is my point.

But 4870x2 got 2GB of really fast memory, so i was wondering if they are using the memory as some kind of caching, and maybe that's why he said ""Equipped with 6,144 Intel CPUs and 5,120 AMD GPUs, Tianhe was able to store all 27 million books in the National Library of China four times over", meaning that they are keeping this much info in cache ?

[AbelJemka]

you know that in the real world a 600 gigaflop gtx280 beats a 1.2teraflop 4870. comparing flops across architectures is almost as bad as comparing clock speed. ATi has a "markitecture" just like intel did.

I know that a GTX280 has 933 gigaflop of computing power (yours numbers us "markitecture"? )
I also know that without good software the best hardware is pointless.
Nvidia has a good software (CUDA).
Apart from this I know nothing

You are both right, but the API for converting that power to computing power (GPGPU) is still "crappy". CUDA would do a better job, for sure. Those 2Tflops is good for games, but can't get converted to computing power easily, is my point.

+1 you posted before me.

[rogueagent6]

you know that in the real world a 600 gigaflop gtx280 beats a 1.2teraflop 4870. comparing flops across architectures is almost as bad as comparing clock speed. ATi has a "markitecture" just like intel did.

I'm sure China took ample time evaluating the offerings from both vendors. If Nvidia's solution is so much better than ATI's, then why didn't China go with Nvidia?

[hurleybird]

"in the real world" doesn't mean "in Crysis" for the Chinese supercomputer users, though

Just to add to that, GTX 280 blows at double precision, while the 4870 doesn't. Obviously both are worse than 5870 and especially Fermi whenever it comes out.

[eric66]

you know that in the real world a 600 gigaflop gtx280 beats a 1.2teraflop 4870. comparing flops across architectures is almost as bad as comparing clock speed. ATi has a "markitecture" just like intel did.

lol real world do you mean games or something there are no nvidia sponsored titles in real world

[Splave]

yeah but can it play crysis

to quote the pic
"wow look at all those harddrives guyz"

[annihilat0r]

Five thousand 4870x2's... I think Crysis was created with this computer in mind.

[Chumbucket843]

mm ati have better gpgpu perf actually check dp benchmarks not nooby stuff like f@h

i would say its dependent on the application. video encoding works on ATi hardware much better but good software isnt really there yet. ATi certainly has a unique advantage in double precision in that it can pair 2 single precision units into one DP unit. they dont waste any die space doing that. nvidia cant do that. they have to dedicate die space for DP.

the current f@h ATi client has been tainted by r600. the entire ATi implementation is a mess. the two main theings holding it back is the fact there is no cache in r600 and rv770 and up have cache but its not used. the speed would easily double if they utilized it. the other bottleneck is the cpu. it has to be used for random number generation. this is why the 5870 has the same ppd ad the 4870. but dont worry gpu3 will fix all of those problems. it uses opencl too. i would expect improvements in the nvidia client too. i would not blame all of the problems on stanford molecular dynamics is really complicated and nvidia is taking a smart approach and helping programmers with libraries and optimizations. ATi has not really done this well. the reason why companies in the past with massively parallel architectures failed is usually because of software.

[Jamesrt2004]

i would say its dependent on the application. video encoding works on ATi hardware much better but good software isnt really there yet. ATi certainly has a unique advantage in double precision in that it can pair 2 single precision units into one DP unit. they dont waste any die space doing that. nvidia cant do that. they have to dedicate die space for DP.

the current f@h ATi client has been tainted by r600. the entire ATi implementation is a mess. the two main theings holding it back is the fact there is no cache in r600 and rv770 and up have cache but its not used. the speed would easily double if they utilized it. the other bottleneck is the cpu. it has to be used for random number generation. this is why the 5870 has the same ppd ad the 4870. but dont worry gpu3 will fix all of those problems. it uses opencl too. i would expect improvements in the nvidia client too. i would not blame all of the problems on stanford molecular dynamics is really complicated and nvidia is taking a smart approach and helping programmers with libraries and optimizations. ATi has not really done this well. the reason why companies in the past with massively parallel architectures failed is usually because of software.

yeah it's stupid, they both went the wrong way, at i could do with good software, and nvidia the hardware (atleast till fermi's out ..)

gpu3 sounds interesting, and actually usefull im sure a lot of people nvidia and ati folders a like are waiting for it

[blob]

i would say its dependent on the application. video encoding works on ATi hardware much better but good software isnt really there yet. ATi certainly has a unique advantage in double precision in that it can pair 2 single precision units into one DP unit. they dont waste any die space doing that. nvidia cant do that. they have to dedicate die space for DP.

the current f@h ATi client has been tainted by r600. the entire ATi implementation is a mess. the two main theings holding it back is the fact there is no cache in r600 and rv770 and up have cache but its not used. the speed would easily double if they utilized it. the other bottleneck is the cpu. it has to be used for random number generation. this is why the 5870 has the same ppd ad the 4870. but dont worry gpu3 will fix all of those problems. it uses opencl too. i would expect improvements in the nvidia client too. i would not blame all of the problems on stanford molecular dynamics is really complicated and nvidia is taking a smart approach and helping programmers with libraries and optimizations. ATi has not really done this well. the reason why companies in the past with massively parallel architectures failed is usually because of software.

Maybe you should checkout Milkyway@home and Collatz Conjecture

Quote by Andreas Przystawik, programmer who optimized the Milkyway@Home client for GPU Computing:
I have no idea what Folding is doing in its code. But you may want to look at two other distributed computing projects using ATI cards: Milkyway@home and Collatz Conjecture.

Both scale virtually perfectly with the higher shader count. That a HD5870 achieves this with the Collatz Conjecture project is also showing that ATI has done a tremendeous job with the memory controller efficiency as it is quite bandwidth hungry.

And I guess I don't have to tell you how pathetic a GTX285 looks compared to a HD5870 at the Milkyway project which uses double precision calculations (it barely matches the performance of an old and rusty HD3850). Nvidia badly needs a factor 8 improvement here to even approach a Cypress, let alone to pass it. And by the way, an nvidia engineer helped the project with its CUDA application, I guess one can assume it's near the maximum performance the hardware is capable of.
I have high hopes for Fermi, but frankly I still doubt a bit it will pass Cypress on pure number crunching.

Oh, by the way, both projects use the Brook+ layer as a base for the ATI applications. So that has nothing to do with the possible scaling.

[WhiteFireDragon]

Professor Zhou Xingming from National University of Defense Technology, said, "Simply put, if everyone of the 1.3 billion people that live in China took a digital photograph, this supercomputer could hold all of them."

i'm questioning if this guy is really qualified to be a professor. my netbook can do this as long as i have enough external storage drives, you don't need a super computer. he doesn't even know the difference between processing power and storage space

[Sam_oslo]

LOL, i hope this is not becoming a ATI and Nvidia fight (again), LOL.

Anyways this is a really strange supercomputer with that many CPU and GPUs, but not a word about the amount of RAM.

[haylui]

TianHe (天河), I don't understand why the official translation to be Milky Way.
Milky Way should be called 银河 rather than 天河....a huge difference...lolz

[Spawne32]

and everyone within 100 miles has lead poisoning

[annihilat0r]

and everyone within 100 miles has lead poisoning

this totally cracked me up

[Serra]

Aside from apparently storing a lot of data on the processors (?) does anyone know what they plan on using it for?

[570091D]

Aside from apparently storing a lot of data on the processors (?) does anyone know what they plan on using it for?

filtering pr0n

[eric66]

filtering pr0n

thats what japanese do

[ownage]

You know that theorically RV770 has more computing power than G200?

Yes and thats what allot of nubs don't know. People don't blind yourself with words like GPGPU, Cuda, OpenCL or anything like that. When the software is right, RV770 can be faster, and thats already proven by expensive decryption software that cant be compared with games or Cuda crap atm.

[apt403]

i'm questioning if this guy is really qualified to be a professor. my netbook can do this as long as i have enough external storage drives, you don't need a super computer. he doesn't even know the difference between processing power and storage space

I'm not sure he can even do basic math...

Working off the assumption that he's slightly mentally deficient and talking about total cache of the system.

Best case senario, the CPUs have 8MB L3, 1MB of L2, and 256KB of L1 (Standard quad-nehalem specs), for a total of 9472KB of cache per CPU. Assuming there are 6144 of these in the system, that's 58,195,968KB of total CPU cache, or about 55.5GB. With 5210 GPUs at 2GB of cache a piece, that's 10420GB of GPU cache, and a total cache of 10475.5GB/~10.2TB for the entire system.

With 10.2TB of space, if you wanted to store 1.3 billion pictures, they'd need to be on the order of 8KB each. I'm not aware of the image quality of cameras in China, but they've either got some damn fine compression, or have a max res of about 100x100 to take pictures that only take up 8KB of space...

[Utnorris]

Aside from apparently storing a lot of data on the processors (?) does anyone know what they plan on using it for?

Hacking the U.S. Super Computer.

[Hu1kamania]

Hacking the U.S. Super Computer.

lol yeah, but I think the US one will win!