China Builds 1-Petaflop Homegrown Supercomputer 185
MrSeb writes "Drawing yet another battle line between the incumbent oligarchs of the West and the developing hordes of the East, China has unveiled a new supercomputer that uses entirely-homegrown processors — 8,704 of them, to be exact. The computer is called Sunway BlueLight MPP and it has a peak performance of just over 1 petaflop — or around the 15th fastest supercomputer in the world. Sunway uses the ShenWei SW-3 1600, a 16-core, 64-bit MIPS-compatible (RISC) CPU. The process used to make the chips is not known, but it is likely 65 or 45nm, a few generations behind Intel's latest and greatest. Each of the 139,264 cores runs at 1.1GHz, the entire system has 150TB of memory and 2PB of storage, and of course it's water-cooled. The ShenWei chips are based on the Loongson/Godson architecture, which China — as in, the country itself — probably reverse engineered from a DEC Alpha CPU in 2001 and has been developing ever since. Sunway is significant for two reasons: a) It's very low-power; it consumes just one megawatt, about half of its contemporaries and one seventh of the US's Jaguar — and b) This is China's first significant supercomputer to be built without Intel or AMD processors."
"Homegrown"? (Score:3, Insightful)
The ShenWei chips are based on the Loongson/Godson architecture, which China — as in, the country itself — has been steadily developing since 2001. It is believed that the Loongson family of processors, including the ShenWei SW-3 found in Sunway, were created by reverse engineering a DEC Alpha CPU.
So you're saying that the entirely homegrown processor was started by reverse engineering a DEC Alpha CPU? Sounds very telling of China's position on innovation (copy/paste). I'm very excited someone is putting pressure on the nations of the world to compute like a boss but it does rub me the wrong way when the title of the article is titled with a "West vs. East" prefix. I'm not trying to get all "Rah Rah USA" here but isn't all the fabrication and chip design built on top of so much history from all around the world? Calling anything entirely "homegrown" in supercomputers or chip design seems kind of unbelievable to me. Unless China's got something radically original [slashdot.org], I'm guessing they owe at least a little credit to so much work done in the USA, Europe the rest of Asia. I mean, it is RISC, right?
This "East vs. West" and "homegrown" stuff is kind of misleading and I find this amusing:
Lest you think this is merely serendipitous happenstance, think again: China has repeatedly stated that it wishes to sever its reliance on American/Western high-tech — and now it can add supercomputers to its rapidly growing list of (mostly reverse-engineered) successes.
And when that is deemed "too slow" where do you turn to move forward? Do you draw on your internal innovation to come up with a new design and process to defeat your opponents or do you merely go back to re-engineering your opponent's latest chip?
Very soon, perhaps by 2020, the only edge that the US will have is in the realm of research and innovation ...
Reverse engineering is innovation? Okay so when China outstrips the United States and defeats the evil Western corporations, who then will they turn to for reverse engineering targets? Also, what is driving this chip to innovate? Who are the competitors for Loongson/Godson? Nobody inside their borders, the government is funding that! That's the problem when your government pays for and decides what you're going to use. Once that's in place, you can sit back and soak up that fat federal funding. Where's the competition going to come from?
Hey man, I love FUD if it kicks our politicians into dumping more of that Military Industrial Complex cash into Science and Research but ... feel free to call me skeptical of your last conclusion. The fact is that by 2020 they're still going to be using this same reverse engineered chip design -- unless they're on their way to reverse engineering another.
The first knockoff supercomputer. (Score:2, Insightful)
The ShenWei chips are based on the Loongson/Godson architecture, which China â" as in, the country itself â" probably reverse engineered from a DEC Alpha CPU in 2001 and has been developing ever since.
This should be a greater argument against handing technology to China, since they just simply copy off of everyone else.
It's the truth, no matter how far you modbomb.
Re:The first knockoff supercomputer. (Score:4, Insightful)
>It's the truth, no matter how far you modbomb.
You get modbombed because you don't bring anything to the discussion except "hate china" and your ideas are lame. Supply them with faulty technology like you suggested in a previous message? Dude, they can get the latest processors off the shelf. And it's not like they don't have fabs for making their own. We gave it to them, willingly.
So if you have anything to say bad about anybody, maybe you should look at US businesses, who in their greed for short term gains, decided to hand the Chinese everything they wanted.
I don't fault the Chinese for anything they do now. I do fault US boards and CEOs for fucking everyone here for a quick buck.
So yeah, you get modbombed because you're not contributing.
Have a nice day.
--
BMO
Re:The first knockoff supercomputer. (Score:5, Insightful)
reverse engineered
Licensed from MIPS.
DEC Alpha CPU
Loongson is MIPS-compatible.
in 2001
The company that makes Loongson was founded in 2002.
Wow, almost every single word in that clause is wrong.
Re:"Homegrown"? (Score:5, Insightful)
Calling anything entirely "homegrown" in supercomputers or chip design seems kind of unbelievable to me.
And all cars are German because everybody has been copying Daimler and Benz (car analogy, w00t), it's a matter of degree. It's at least homegrown in the sense that it's domestically produced and they are not currently relying on foreign companies do produce it. And since a 2001 era Dec Alpha would be built on 180nm process and this is supposedly on 45 or 65nm, they've clearly redesigned it quite a lot adjusting timings, gates and all that. You can't just take a design and make it 1/4th the size. That tells me they actually know a lot about this technology themselves.
And when that is deemed "too slow" where do you turn to move forward? Do you draw on your internal innovation to come up with a new design and process to defeat your opponents or do you merely go back to re-engineering your opponent's latest chip?
That's not an either-or question.
Reverse engineering is innovation? Okay so when China outstrips the United States and defeats the evil Western corporations, who then will they turn to for reverse engineering targets?
Just because it's unsustainable in the long run, doesn't mean it makes sense now. Innovation is possible, but they're so far behind copying is faster. As long as you're ignoring IP laws, that seems logical. Hollywood ignored copyright laws, now that the balance is in their favor they enforce it with vigor.
Also, what is driving this chip to innovate? Who are the competitors for Loongson/Godson? Nobody inside their borders, the government is funding that! That's the problem when your government pays for and decides what you're going to use. Once that's in place, you can sit back and soak up that fat federal funding. Where's the competition going to come from?
You might as well say the Apollo program had no domestic competition, the country was founding it. China wants homegrown CPUs and supercomputers so they will run a program to get it, and it'll run for as long as they need it to run.
The fact is that by 2020 they're still going to be using this same reverse engineered chip design -- unless they're on their way to reverse engineering another.
You must not have been paying very good attention to what China is doing, they're absorbing high tech at a huge rate. Their high speed rail is a good example, they imported technology from Germany and Japan, then kept building on it. They now have the largest high speed rail network in the world, with their own train designs. You think that isn't their goal with CPUs? Grab what you can, build on top. It doesn't have to be #1, just good enough they don't rely on anyone else.
Re:Why can't the US just give them a bad Concorde? (Score:4, Insightful)
He said "subtly defective" not "Batshit insane"