Is Tianhe-2 Overrated?

[Lure]

You made that or you bought it?

If you talk about buying my friend Iraq was the first country to buy and use super computers in the seventies!!! look where we were and where we are now because of dictators.

Hey Malik. I didn't understand what you've meant by "making" or "buying" it? Today a huge portion supercomputers uses US chips and US originated softwares at some point. It's not feasible to make your "own" supercomputer for any country today. I mean think about it, you create your own OS, your own CPU, your own chips etc. China has to do it because US put sanctions for high volume CPU imports that can be used for supercomputers. So US actually makes China a favor. They are now more motivated to make their CPU's more widespread.

So in order to answer your question, from hardware perspective it's bought like the rest of the world. I mean with that logic K computer is also largely bought because it also uses US CPU and Tianhe-2A is bought because of the same reason.

If we are talking from a software perspective, yeah there are custom softwares produced by ITU for that machine.

 

[Malik Alashter]

Hey Malik. I didn't understand what you've meant by "making" or "buying" it? Today a huge portion supercomputers uses US chips and US originated softwares at some point. It's not feasible to make your "own" supercomputer for any country today. I mean think about it, you create your own OS, your own CPU, your own chips etc. China has to do it because US put sanctions for high volume CPU imports that can be used for supercomputers. So US actually makes China a favor. They are now more motivated to make their CPU's more widespread.

So in order to answer your question, from hardware perspective it's bought like the rest of the world. I mean with that logic K computer is also largely bought because it also uses US CPU and Tianhe-2A is bought because of the same reason.

If we are talking from a software perspective, yeah there are custom softwares produced by ITU for that machine.

Ok, then why they don't buy the hardware to make their own super computer like the others and write the software.

 

[Beast]

it's been the fastest supercomputer in the world since it's debut in June 2013. running at a blistering speed of 33.86 Pflops, but this monster also consumes 17.6 MW of power (24MW including cooling) that's enough power to power 240,000 homes in the U.S


fast as it is, it's inefficient. coding for Tianhe-2 is a challenge , and the cost just to use it is insanely high.

World's fastest computer, Tianhe-2, might get very little use | South China Morning Post

Only sourgrape will claim Tianhe-2A is overrated. It helps to build China almost all strategic system and weapon. It is a projected needed at all cost. Project sponsored by PLA.

To think Tianhe-2A is only used for commercial uses is just plain stupid and sourgrape attempt by loser western to discredit China.

 

[C130]

Only sourgrape will claim Tianhe-2A is overrated. It helps to build China almost all strategic system and weapon. It is a projected needed at all cost. Project sponsored by PLA.

To think Tianhe-2A is only used for commercial uses is just plain stupid and sourgrape attempt by loser western to discredit China.

Tianhe-2A is just a inefficient design especially if it's used for military purposes

http://www.graph500.org/results_nov_2015


even the weaker K supercomputer beats Tianhe-2A

so my point of it being overrated is spot on

 

[Jlaw]

Tianhe-2A is just a inefficient design especially if it's used for military purposes

http://www.graph500.org/results_nov_2015


even the weaker K supercomputer beats Tianhe-2A

so my point of it being overrated is spot on

So your whole premise is that Tianhe 2 is inefficient because it waste more energy than some other supercomputers? I guess in your world, the Nissan Sentra must be better than a Lamborghini Aventador

 

[FairAndUnbiased]

Tianhe-2A is just a inefficient design especially if it's used for military purposes

http://www.graph500.org/results_nov_2015


even the weaker K supercomputer beats Tianhe-2A

so my point of it being overrated is spot on

inefficient is one thing but raw computational power is another.

with the new Shenwei chips, the next one will be both efficient and have the raw computational power.

 

[vtnsx]

So your whole premise is that Tianhe 2 is inefficient because it waste more energy than some other supercomputers? I guess in your world, the Nissan Sentra must be better than a Lamborghini Aventador

But China doesn't build Lamborghini and/or Nissan Sentra.

inefficient is one thing but raw computational power is another.

with the new Shenwei chips, the next one will be both efficient and have the raw computational power.

lol, I don't know why you nerds are drooling over computer technology. It's been proven that no computers/supercomputers today can match the awesome Quantum computer.

 

[C130]

So your whole premise is that Tianhe 2 is inefficient because it waste more energy than some other supercomputers? I guess in your world, the Nissan Sentra must be better than a Lamborghini Aventador

if the Lamborghini Aventador can only hit second gear then yes. I would say that was a massive waste of money and resources.


you spent $400 million on Tianhe-2 that's gimped!! when with those same resources you could buy 4 Mira or Piz Daint instead

 

[xunzi]

if the Lamborghini Aventador can only hit second gear then yes. I would say that was a massive waste of money and resources.


you spent $400 million on Tianhe-2 that's gimped!! when with those same resources you could buy 4 Mira or Piz Daint instead

If you understand supercomputer which I always want you to do more research on, you must understand that there is no such thing as having enough processing speed. We ALWAYS need higher speed. The more the better as doing multiple intensive application is where Tianhe-2 will shine. On a daily basis, every small supercomputer can do the job but to do the heavy lifting such as 3d aircraft model, nuclear simulation, etc... you need tianhe2.

 

[C130]

If you understand supercomputer which I always want you to do more research on, you must understand that there is no such thing as having enough processing speed. We ALWAYS need higher speed. The more the better as doing multiple intensive application is where Tianhe-2 will shine. On a daily basis, every small supercomputer can do the job but to do the heavy lifting such as 3d aircraft model, nuclear simulation, etc... you need tianhe2.

processor speed is one thing but it's the others subsystems that make a supercomputer great

this is why Kei Supercomputer is the best in the world



The benchmark used in Graph500 stresses the communication subsystem of the system, instead of counting double precision floating-point



1.Kei- 18,000 GTEPS
2.Titan- 16.600GTEPS
3.Mira- 14,300GTEPS
5-Tianhe-2- 2,000GTEPS


the intel processors are good but Chinese subsystems like the interconnect and what not is garbage compared to the Japanese and American ones

 

[xunzi]

processor speed is one thing but it's the others subsystems that make a supercomputer great

this is why Kei Supercomputer is the best in the world



The benchmark used in Graph500 stresses the communication subsystem of the system, instead of counting double precision floating-point



1.Kei- 18,000 GTEPS
2.Titan- 16.600GTEPS
3.Mira- 14,300GTEPS
5-Tianhe-2- 2,000GTEPS


the intel processors are good but Chinese subsystems like the interconnect and what not is garbage compared to the Japanese and American ones

You are obviously pretty clueless. Everybody can make up arbitrary benchmark and the list would look different. Graph500 is not an official recognizable benchmark like Linpack's Top500 because it measures a different scenario. In other words, Graph500 is a complementary benchmark to measure a specific needs of supercomputer speed. If you care to do research as I advice you to do, you notice how many cores and node your Titan and Kei contain compare to Tianhe2.

When designers design their supercomputer, they don't focus on Graph500 criteria, instead they are focusing on the simple fact that every layman can understand, how fast can a supercomputer solve a linear problem. That's why the Linpack is the acceptable and official benchmark to measure peak performance. An anology is the 100m is the official track & field race but not the 50m, 60m, etc.. It's the same as putting all the athletes in a race, and whoever finish from point A to point B in 100m, is the fastest man alive. If Usain Bolt fail to meet your expectation and lost the 50m race as he is known to be a slow starter, would you say he is overrated? That's stupid right? This is the easiest and most simple method for layman to compare and understand. So like I said, tianhe2 is developed to test peak performance that meet Linpack benchmark. Every other benchmark, HPC, Graph 500, etc.. are simple trying to complement the Linpack benchmark and promote their own version of testing supercomputer performance. In this case, it is testing multi variable in a parallel scenario.

 

[C130]

You are obviously pretty clueless. Everybody can make up arbitrary benchmark and the list would look different. Graph500 is not an official recognizable benchmark like Linpack's Top500 because it measures a different scenario. In other words, Graph500 is a complementary benchmark to measure a specific needs of supercomputer speed. If you care to do research as I advice you to do, you notice how many cores and node your Titan and Kei contain compare to Tianhe2.

When designers design their supercomputer, they don't focus on Graph500 criteria, instead they are focusing on the simple fact that every layman can understand, how fast can a supercomputer solve a linear problem. That's why the Linpack is the acceptable and official benchmark to measure peak performance. An anology is the 100m is the official track & field race but not the 50m, 60m, etc.. It's the same as putting all the athletes in a race, and whoever finish from point A to point B in 100m, is the fastest man alive. If Usain Bolt fail to meet your expectation and lost the 50m race as he is known to be a slow starter, would you say he is overrated? That's stupid right? This is the easiest and most simple method for layman to compare and understand. So like I said, tianhe2 is developed to test peak performance that meet Linpack benchmark. Every other benchmark, HPC, Graph 500, etc.. are simple trying to complement the Linpack benchmark and promote their own version of testing supercomputer performance. In this case, it is testing multi variable in a parallel scenario.

only way to find out if you or I am right is have Tianhe-2 and Kei solve a problem and see who does it first