Is Tianhe-2 Overrated?

[Guynextdoor2]

yawn you didn't get my point

Tianhe-2 is the fastest supercomputer in the world (when running LINPACK) ...I am not disputing that. M y point was it doesn't use all that power because Chinese researchers are having a hard time programming code to unlock all that power. and the cost just to use it per day would set you back $60,000 to $100,000

U.S,Western, and Japanese have better software

and looking at all the parameters that make a supercomputer super Tianhe-2 ranks 6th

Brief Introduction | Graph 500


the Japanese Kei Supercomputer reigns supreme

@Nihonjin1051

Guess the prestige of havig 'fastest supercomputer' was more important than actually using it.

 

[JSCh]

China has five supercomputer center, one Tianhe-2, two Tianhe-1a, one Sunway Bluelight, one Sugon 6000.

With a increasing connected world, there is increasing demand of computing power. One of the biggest demand is from cloud computing.

There is a smaller Tianhe-2 LvLiang Solution, currently rank as no. 34 fastest supercomputer, working as a cloud computing center in Lvliang.

China currently has around a hundred supercomputer in the top 500 supercomputer of the world. Tianhe-2 currently might not be utilized fully all the time for commercial and administrative reason like the other hundred are, but there are applications where money is not that much of a concern.

Because of that, Tianhe-2, or any other fastest supercomputer that China could get, is definitely very important for a big country like China to have.

 

[C130]

If you bother to look at that list,notice how many nodes and cores the top above Tianhe-2 were using. Right now, Tianhe-2 has the best computational power but it is not needed to use all because there is nothing needed for it to use all that power. It is that simple. In the future, we might need those power but right now we don't.

An analogy is like you used a Mustang vs a Honda Civic. If you are on a highway trying to escape a cop, you would certainly choose a Mustang over a Honda Civic but you are not in trouble, so you don't need a Mustang. It is always good to have it, just in case.

let's say there were two Tianhe-2 in the world

one in the hands of the Chinese and the other in the U.S

the only thing can different the two is the program/codes written for them

the code written for the Chinese Tianhe-2 isn't good while the one in the U.S has good code.

you know the theoretically peak for Tianhe-2 is 54.9 Pflops. that can only be achieved with 100% perfect software

that's all I was alluding to

 

[jhungary]

The Tianhe2 is a Brute Force supercomputer, basically it does not have a good network infra (At least up to the point I read about it 2 years ago, may have upgraded now, I don't know) What gave Tianhe2 the sheer computing power is the sheer number of node, CPU and RAM that attached to it. That is the reason why Tianhe2 draw that much power, as they have a lot (70,000 CPU) to power.

On the other hand, US supercomputer is based on economic design and with small node and try to maximize the power to process ratio. Where US usually have a slower supercomputer, but they have a lot of them.

There are good and bad coming from each system, the good for the Chinese is they can show the brute power, basically build a computer to show off to the world number one position (Ironically, by using American CPU) and the way to connect them (all 70000 cpu and controller) itself is a sheer feat. And that should be commended.

However, in reality, building the fastest supercomputer in the world twice as fast as the next one does not spell out in reality. Because to be honest, not that many process require that much computing power to begin with and most of the job can be finish with a lower power consumption. But every time Tianhe2 start a process, any process, it will cost as if they are using all the power the supercomputer gonna give.

Under the statistical assumption, the standard deviation favour the median task, unless China really need have that many high demand computing process to justify the use of Tianhe-2 full power, they are losing money and would be better off using more potent and economical machine.

 

[Jlaw]

Those guys who compiled Top-500 must be commies!

Yeah exactly. You know all these tests that supercomputers go through for rankings must be all set up by China. These computer experts are not credible, but SCMP "reporter" is most credible!

Tianhe-2 : 24 MW for 33.86 Pflops
IBM: 3.9 MW for 8.59 Pflops
If you scale IBM machine to 33.86 Pflops, it could consume to 16 MW. IBM machine is only 50% energy efficient. Tianhe-2 sacrifices 50% energy efficiency for 4 times boost on speed. It is a great achievement since it can save much time to reduce development cycles greatly.

the American is jealous that a commie is beating them in supercomputer. It's like saying a Lamborghini uses more gas than a Ford Focus! Ford must be better than Lamborghini!

 

[Abacin]

For most projects from infrastructures to smartphone manufacturing, the development cycle is at least cut to half for China compared to USA. That's why companies like Apple like to design in USA but manufacture in China

so it's not overrated

good conversation




the point is Tianhe-2 can't even run at it's full potential because of the lack of coding/software




34% capacity


you mention scaling which means jack if the higher performing machine isn't running at it's peak performance.


it can reach 200 miles per hour but is only going 60 miles per hour (another car analogy)

the 10 Pflops Mira with it's lower cost is operating at the same speed.


which brings me back to saying Tianhe-2 is overrated.

But for each client, the development cycle is still reduced greatly. This is why China becomes the center of world manufacturing center because China has the lowest development cycle. Companies such as Apple are forced to give the manufacturing jobs to Chinese companies, otherwise their products can be delayed several months and lose competitions. Similarly chinese military technology is rapidly catching that of USA since we have shorter development cycles.

 

[Aepsilons]

Irrespective of precision, energy efficiency, one thing is clear is the ingenuity of the Chinese science sector. China is a powerful competitor , and this excellence only will drive counter development. Good Job to our friends in the Chinese Science Industry. Keep it up, Keep it up.

This is why China becomes the center of world manufacturing center because China has the lowest development cycle.

Without a doubt, yes. Might i also add streamlined production aspect in the manufacturing sector in China. Lean processing et al.

 

[seven7seven]

Why is C130 not banned for this thread?

This is clearly a troll bait thread.

Those who can do, do. Those who can't, whine and cry. USA is becoming the latter. JealoUSy is the new thing!!

 

[Abacin]

Computer calculations and simulations have become very important in reducing development cycles. That's why all countries including China and USA are pushing for faster supercomputers, even with lower energy efficiency and higher cost. For example in designing stealth fighters, computer simulations can greatly reduce the number of real tests and reduce development cycles. While USA always argues that China stole those technologies from USA, which makes China progress so rapidly, one important reason is that China integrates computer simulations better than USA since F22 and F35 were first designed when supercomputers were very poor.

 

[xunzi]

Beat our speed first, then talk about its overrated status. I don't understand how the US suck at competition. They are scare of fair competition so they try to slow the race by banned Intel. What happen to free market competition big mouth? LOL

 

[seven7seven]

Computer calculations and simulations have become very important in reducing development cycles. That's why all countries including China and USA are pushing for faster supercomputers, even with lower energy efficiency and higher cost. For example in designing stealth fighters, computer simulations can greatly reduce the number of real tests and reduce development cycles. While USA always argues that China stole those technologies from USA, which makes China progress so rapidly, one important reason is that China integrates computer simulations better than USA since F22 and F35 were first designed when supercomputers were very poor.

Totally agree. China is catching up fast in nearly all areas and probably ahead in some. I never bought into this idea China were decades behind US in most things. What has already been learned by the West can be learned quickly by others and China has done just that. China is now beginning to show that they can build on this acquired knowledge and start becoming the innovators themselves. However, China should not become complacent like the West and publish all of their research of strategic importance, believing other nations are too primitive to understand and make use of it.

China once gave the white man the secret of gunpowder and they came back later and used it against China. China should never do this again.

 

[nang2]

Baby steps. First, make it work. Then make it beautiful.

 

[jammersat]

unless those s-400 missiles are staged i don't see how missiles that long can bring down a raptor an f-18 or an f-16

 

[Lure]

I would even go as far to say that the Argonne Mira is a better supercomputer than Tianhe-2

With that logic Japan dominates the supercomputing industry.

Lists

But there a lot of metrics out there to measure the performence of a supercomputer. Energy efficiency is of course one of them. However there is data intensive modeling performence for big data analytics and there is matrix calculation performence for simulation needs. Tianhe-2A is currently the fastest supercomputer in simulation purposes (LINPACK Benchmark source : November 2015 | TOP500 Supercomputer Sites and ranked 6th in data intensive computing performence. Japan's K Computer is the fastest supercomputer for data intensive supercomputing.

November 2015 | Graph 500

fast as it is, it's inefficient. coding for Tianhe-2 is a challenge

This has no meaning. Parallel processing is a hard field. Users are responsible for using the machine efficiently. Not the builders of the machine. Users should optimize their codes with the best way they could in order to run them fast. How many threads will be used, how many processes should be used, reinventing the sequential algorithms and come up with parallel implementations are user responsibility in every center.

Lazy and uneducated programmer will complain. That's a fact. But hey, what the heck is a mediocre programmer doing near those machines? A simple deadlock could create a lot of problems in those machines. The code should be tested intensively before running it on supercomputer. Because the supercomputer time is limited and valuable.

Those are the challenges that you experience in every system. I'm sorry to say that but there is no other challenge specific to Tianhe-2A.

“It seems a lot of these massive machines, usually made with large government investment, lie idle after they are made, or are even abandoned midway, due to fundamental defects in China’s traditional bureaucratic management system,” remarks He.

This doesn't make a supercomputer overrated. It only shows that China has more supercomputer capacity than it actually uses. However as the time goes by and the technology gets to be well known by a larger scientist group the utilization rate will go high. Tianhe-1 had also a very utilization rate when it first got out. Today it's very hard to book a time.

Chinese researchers are having a hard time programming code to unlock all that power. and the cost just to use it per day would set you back $60,000 to $100,000

U.S,Western, and Japanese have better software

This is not about Tianhe-2A. This is about the parallel computing ability of Chinese researchers. Researchers not writing efficient piece of sotware doesn't mean the machine is overrated. It means Chinese researchers lack the experience of using a supercomputer. With time, they will have the experience.

which is why I brought up why didn't China just build three 10 to 15 Pflop supercomputers??

once could havefdone aerospace research, another nuclear energy/nuclear weapons research, and the third on whatever

This is also nonsense. Supercomputers can be divided into pieces and multiple users can use them simultaneously. Merging them into one center actually makes it more efficient, then making three different centers.

the only thing can different the two is the program/codes written for them

the code written for the Chinese Tianhe-2 isn't good while the one in the U.S has good code.

you know the theoretically peak for Tianhe-2 is 54.9 Pflops. that can only be achieved with 100% perfect software

This is not about the machine, but about the experience of the researchers in the field of parallel processing. If they write inefficient code, they will get inefficient results. That doesn't mean the machine is overrated.

 

[Irfan Baloch]

@xunzi
I have removed the personal insult from your post but next time you will get infraction instead. please stay civil and contribute in an educated manner
stay classy and all is well

@Nihonjin1051 hehe, I am a little confused buddy, are you posting at the expense of some people here? some of your posts are rubbing @jhungary the wrong way. I value both of you very much but in this thread I am at not sure if your posts are off the mark or @jhungary woke up wrong side of the bed
stay cool