AMD's Hawaii Is Officially The Most Efficient GPGPU In The World To Date, Tops Green500 List

[Faa]

Obviously, but GPU's do take up a large margin. Consider that, in terms of actual Computer Components, the GPU takes up the most power out of any component, by a huge amount. They draw 2 or 3 times the amount as a CPU (or more, depending on what CPU). AND modern Super Computers have hundreds or thousands of GPU's.

 

Not to mention that the most demanding aspects - example: cooling/environmental control - are going to be fairly similar for any modern supercomputer, as they are all going to have similar requirements.

 

But the fact remains that, even if they were able to save power in other areas, it does not diminish the accomplishments AMD has made towards GPGPU compute and supercomputer efficiency.

Hmm if your target is 60°, bumping that up to 80° saves a ton of energy. youtube.com/watch?v=PBx7rgqeGG8 @ 3:20

It's not going to be similar for all of them, different country different ambient, far from it imo.

Seem like they are only measuring the power the supercomputer actually pulls, so cooling excluded: http://www.green500.org/docs/pubs/RunRules_Ver1.0.pdf

[HippY]

@patrickjp93

yet people use AMD no matter how bad they are in your opinion...

 

you know what? lets look at differently at this datasheet:

 

1. You and me (and many others) are not going to build a datacentre at our backyard

2. The people who created the No 1 SC according the list were running oCL because reasons. I am 100% sure that there is a reason why.

3. These company made a list for oCL Best of list and they did well. Why? If you want to use CUDA, you are going to use Nvidia. End of the story. You do not care which one is the best of all GPUs, as you buy an Nvidia GPU. 

4. Now we realize that this list is nothing more, than a show off list for the shareholders to pet them on the shoulder.

[szomeret]

Well, the R9 290X is very warm. This is a function of its rather high, but not insane, TDP. But also a function of its relatively poor thermal design. The reference Nvidia cards have better coolers.

Secondly, let's do a brief comparison of the Tesla K40 and the FirePro S9150. Both are 235W. K40 has peak of 1.66 DP-TFLOPS and 5 SP-TFLOPS. The S9150 has peak of 2.53 SP-TFLOPS and 5.07 SPTFLOPS. For doing pure mathematics, the AMD GPU are beast. Their weakness is their rasterizaters(ROPS) and related rendering features. This reduces their performance in video games from their pure mathematics capabilities. The new Tesla K80 is better from a power efficiency perspective for compute performance, though. It just hasn't appeared in any actual applications yet (no less a super computer).

Finally, looking at the specifications of this super-computer: they are ridiculously GPU heavy. Even the worst energy burning GPU from a node back (GTX 480) are more energy efficient for doing highly parallel math than just about any CPU on modern nodes, such as Ivy Bridge or Haswell. Most other super computer have a slightly higher ratio of CPU, which reduces their energy efficiency score for vector math tests like these. Highly parallel processors are very good at these supercomputer energy efficient benchmarks.

[patrickjp93]

I don't believe I spouted anything about false equivalencies or even mentioned another card what so ever so your argument is invalid

Hashing and GPGPU is a false equivalency. You said AMD cards specifically. The only reason AMD was good for coin mining was a pair of instructions which just ran faster for them. It had nothing to do with AMD's raw compute power or efficiency.

[Lord Baal]

Hashing and GPGPU is a false equivalency. You said AMD cards specifically. The only reason AMD was good for coin mining was a pair of instructions which just ran faster for them. It had nothing to do with AMD's raw compute power or efficiency.

I don't believe I mentioned raw compute power or power efficiency I simply stated another element the cards were good for so the fact that AMD may be better for some specific uses isn't surprising to me in the slightest when it's well known the current maxwell chips are fairly gaming focused.

You can read more into what I said then was actually said all you like but I don't have the patience to deal with your 14 year old fanboy crap.

[The_Strict_Nein]

I don't believe I mentioned raw compute power or power efficiency I simply stated another element the cards were good for so the fact that AMD may be better for some specific uses isn't surprising to me in the slightest when it's well known the current maxwell chips are fairly gaming focused.

You can read more into what I said then was actually said all you like but I don't have the patience to deal with your 14 year old know it all fanboy crap.

 

I have to agree with patrick here, loathe as I sometimes am to do so

 

You said that people shouldn't be surprised that AMD are better for compute performance, as "there's a reason they were being used for coin mining". Coin mining is based on a specific set of instructions, that as patrick said, ran with 2 cycles on Nvidia and 1 on AMD, effectively doubling hashing performance for free.

 

The comparison between direct compute power and hashing is invalid, because hashing instructions are completely different to compute instructions, typically.

 

As such, your argument "I'm not sure why anyone's remotely surprised" is invalid when talking about hashing.

 

I'm fairly sure (but not 100%) that for many instructions not based on OpenCL, the raw GFlops of Nvidia cards easily beat out AMD

[Lord Baal]

I have to agree with patrick here, loathe as I sometimes am to do so

 

You said that people shouldn't be surprised that AMD are better for compute performance, as "there's a reason they were being used for coin mining". Coin mining is based on a specific set of instructions, that as patrick said, ran with 2 cycles on Nvidia and 1 on AMD, effectively doubling hashing performance for free.

 

The comparison between direct compute power and hashing is invalid, because hashing instructions are completely different to compute instructions, typically.

 

As such, your argument "I'm not sure why anyone's remotely surprised" is invalid when talking about hashing.

 

I'm fairly sure (but not 100%) that for many instructions not based on OpenCL, the raw GFlops of Nvidia cards easily beat out AMD

Thing is this did rely on OpenCL and pointing out another scenario in which the cards under specific conditions can perform better then the competition regardless of GFlops was the only reason I mentioned coin mining what so ever. 

[dalekphalm]

Thing is this did rely on OpenCL and pointing out another scenario in which the cards under specific conditions can perform better then the competition regardless of GFlops was the only reason I mentioned coin mining what so ever. 

Yeah as much as crytocurrency hashing is a very specific task, it's still a part of GPU Compute. You're still using OpenCL.

 

So while it may not reflect GPU Compute or OpenCL performance as a whole, it IS important to note. There may be other OpenCL/GPGPU tasks that AMD cards are extremely efficient at as well. You cannot just blanket dismiss @Lord Baal's comment simply because not all GPGPU tasks will be that good.

[cesrai]

"Einstein's relativity work is a magnificent mathematical garb which fascinates, dazzles and makes people blind to the underlying errors. The theory is like a beggar clothed in purple whom ignorant people take for a king... its exponents are brilliant men but they are metaphysicists rather than scientists."

New York Times (11 July 1935), p. 23, c.8

"Today's scientists have substituted mathematics for experiments, and they wander off through equation after equation, and eventually build a structure which has no relation to reality."

-Nikola Tesla

Hi I got really confused are you calling Einstein an idiot and saying Reltivaty is wrong?

What did that expermint show when the bars moves? They moved vertacliy but 'horzantly' they weren't moving why?

Can you please explain or give me a Linn to an article or anything that would help me understand.

[mr moose]

Hi I got really confused are you calling Einstein an idiot and saying Reltivaty is wrong?

What did that expermint show when the bars moves? They moved vertacliy but 'horzantly' they weren't moving why?

Can you please explain or give me a Linn to an article or anything that would help me understand.

 

I think he is just trying to say that if you keep setting up more and more different tests that all focus on a slightly different aspect of something, all you will do is end up with a whole lot of data that can't be compared to anything else and is therefore useless in understanding reality.   You can pull apart the green 500 methodology and dispute it but it won't change reality, and like Einsteins theory, those that understand it the least will argue the most over it's relevance.

[deathjester]

Hi I got really confused are you calling Einstein an idiot and saying Reltivaty is wrong?

What did that expermint show when the bars moves? They moved vertacliy but 'horzantly' they weren't moving why?

Can you please explain or give me a Linn to an article or anything that would help me understand.

 

http://www-history.mcs.st-andrews.ac.uk/Extras/Einstein_ether.html

 

Einstein failed his college entrance exam and married his own cousin. I do not deify Einstein and treat him as an infallible genius. I like empirical evidence. Also I never called Einstein an idiot and neither did Nikola Tesla. He called him a brilliant man and said that he simply thought he was wrong.