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

[patrickjp93]

Everyone missed the giant caveat in this report. They only used OpenCL. Run equivalent CUdA code and Nvidia wins by a wide margin almost universally. The theoretical flops of AMD's cards are also way higher than the actual reachable, whereas Nvidia's best researchers and collaborators have made applications for scientific computing that get within 7% of the peak theoretical and sustain it.

[Jack.EXE]

Everyone missed the giant caveat in this report. They only used OpenCL. Run equivalent CUdA code and Nvidia wins by a wide margin almost universally. The theoretical flops of AMD's cards are also way higher than the actual reachable, whereas Nvidia's best researchers and collaborators have made applications for scientific computing that get within 7% of the peak theoretical and sustain it.

love how nvidia fans say this, when they bash mantle which does the same for AMD in games.

[Altecice]

All you people cracking jokes at how hot the cards are need to seriously move on... So what it came with a bad reference cooler that made them run warm... nothing that aftermarkets didint fix. If we kept talking about old stuff then we would still all be moaing about the 480gtx's heat output but I fail to see anyone talk about that?

 

stop flogging a dead horse.

[patrickjp93]

love how nvidia fans say this, when they bash mantle which does the same for AMD in games.

Except in my case I'm right. CUDA is also a much easier language to learn and performs amazingly well compared to OpenCL.

[patrickjp93]

All you people cracking jokes at how hot the cards are need to seriously move on... So what it came with a bad reference cooler that made them run warm... nothing that aftermarkets didint fix. If we kept talking about old stuff then we would still all be moaing about the 480gtx's heat output but I fail to see anyone talk about that?

stop flogging a dead horse.

4-year+ old card vs. What AMD released last year...

[Lynx]

all this bashing and etc... <_< it's efficiency in compute perfomance not gamings, the world of GPU didn't revolve around gaming alone

[Altecice]

4-year+ old card vs. What AMD released last year...

 

The principle is still the same. Talking about stuff that's been fixed.

[patrickjp93]

The principle is still the same. Talking about stuff that's been fixed.

Except it hasn't. The heat output of the 290/x is obnoxious.

[bogus]

all this bashing and etc... <_< it's efficiency in compute perfomance not gamings, the world of GPU didn't revolve around gaming alone

Just let it be, you are wasting your time - I mean, people bashed the eizo because they released a 1:1 ratio monitor LOL

[MetroP]

Everyone missed the giant caveat in this report. They only used OpenCL. Run equivalent CUdA code and Nvidia wins by a wide margin almost universally. The theoretical flops of AMD's cards are also way higher than the actual reachable, whereas Nvidia's best researchers and collaborators have made applications for scientific computing that get within 7% of the peak theoretical and sustain it.

The Nvidia powered super computer was running CUDA, it was still 25% less efficient than the AMD powered super computer.

[LukaP]

Except in my case I'm right. CUDA is also a much easier language to learn and performs amazingly well compared to OpenCL.

Except in your case you are just repeating words you heard me say in THE EXACT ORDER, and before you were a lover of oCL... Go sleep, its late for a 14 y/o to stay up so long

 

Except it hasn't. The heat output of the 290/x is obnoxious.

except its not. its within 10% of the GK110, and probably lower than GK210

 

The Nvidia powered super computer was running CUDA, it was still 25% less efficient than the AMD powered super computer.

Dont bother, he is a troll that knows a few words and then uses them in sentences that appear to make sense so he can start arguments. I havent seen him be right once. and when he gets bashed in the head by like 5 people at the time, he just takes what they have been saying, and starts using those words as well, having no idea what they mean

[Opcode]

Everyone missed the giant caveat in this report. They only used OpenCL. Run equivalent CUdA code and Nvidia wins by a wide margin almost universally. The theoretical flops of AMD's cards are also way higher than the actual reachable, whereas Nvidia's best researchers and collaborators have made applications for scientific computing that get within 7% of the peak theoretical and sustain it.

They didn't say anything about using OpenCL nor CUDA. But I can guarantee each of their own are using either OpenCL or CUDA.

 

 

Except in my case I'm right. CUDA is also a much easier language to learn and performs amazingly well compared to OpenCL.

CUDA isn't that much easier than OpenCL. A few steps can be simplified due to it being propitiatory software built for a specific hardware. CUDA doesn't have the same flexibility as OpenCL.

[patrickjp93]

The Nvidia powered super computer was running CUDA, it was still 25% less efficient than the AMD powered super computer.

Except, it wasn't. It was running OpenCL, but thanks for playing.

 

Except in your case you are just repeating words you heard me say in THE EXACT ORDER, and before you were a lover of oCL... Go sleep, its late for a 14 y/o to stay up so long

 

except its not. its within 10% of the GK110, and probably lower than GK210

 

Dont bother, he is a troll that knows a few words and then uses them in sentences that appear to make sense so he can start arguments. I havent seen him be right once. and when he gets bashed in the head by like 5 people at the time, he just takes what they have been saying, and starts using those words as well, having no idea what they mean

 

I never said I loved OpenCL. It's just more platform agnostic and more useful as a language/framework to know.

 

I have been right in all but 1 of our arguments thus far. By the way, Nvidia did go back and switch Volta to Pascal in the IBM supercomputer announcement, so in your face!

[patrickjp93]

They didn't say anything about using OpenCL nor CUDA. But I can guarantee each of their own are using either OpenCL or CUDA.

 

 

 

CUDA isn't that much easier than OpenCL. A few steps can be simplified due to it being propitiatory software built for a specific hardware. CUDA doesn't have the same flexibility as OpenCL.

It doesn't have the platform agnosticism that OpenCL does, but it is MUCH easier to program in and produce highly efficient code which OCL cannot touch.

 

Also, no, the test was uniform usage of an OpenCL benchmark suite, but thanks for playing instead of reading the article and assuming I'm always wrong. 

 

feel free to consult @LukaP since my word can apparently never be right. She will also tell you CUDA is superior in every single way from a performance and programmability viewpoint, and she would be right.

[alexyy]

Imma guess they must of over volted the GPUs to get the insane temps and power draw

[Opcode]

It doesn't have the platform agnosticism that OpenCL does, but it is MUCH easier to program in and produce highly efficient code which OCL cannot touch.

 

feel free to consult @LukaP since my word can apparently never be right. She will also tell you CUDA is superior in every single way from a performance and programmability viewpoint, and she would be right.

You wanna put that to the test? Unless you can provide the forum with some source code and results for the same task ran through each framework then you should stop flapping at the jaw. You spew enough nonsense on the forum as it is. If you want I can compile the OpenCL equivalent to your own CUDA example. Let me know as these results would be beneficial to the forum in card comparison as well. This is your chance to put up or shut up.

 

Also, no, the test was uniform usage of an OpenCL benchmark suite, but thanks for playing instead of reading the article and assuming I'm always wrong. 

If you actually read the original article in no place does it state the use of a framework for any of the three supercomputers. Tho anyone with half a brain would know each one utilizes either OpenCL or CUDA.

[RexinOridle]

It's efficient. It warms up your home and compute at the same time. You don't find things like that just anywhere.

 

Ok, enough with the joking. I think in the perspective of raw power of stream processors(Open-CL), it probably is the most efficient GPU in the market right now.

[patrickjp93]

You wanna put that to the test? Unless you can provide the forum with some source code and results for the same task ran through each framework then you should stop flapping at the jaw. You spew enough nonsense on the forum as it is. If you want I can compile the OpenCL equivalent to your own CUDA example. Let me know as these results would be beneficial to the forum in card comparison as well. This is your chance to put up or shut up.

 

If you actually read the original article in no place does it state the use of a framework for any of the three supercomputers. Tho anyone with half a brain would know each one utilizes either OpenCL or CUDA.

http://arxiv.org/pdf/1005.2581.pdf

Chew on that, a wide range of scientific computing and quantum simulation workloads. CUDA is universally better.

[LukaP]

Except, it wasn't. It was running OpenCL, but thanks for playing.

 

 

I never said I loved OpenCL. It's just more platform agnostic and more useful as a language/framework to know.

 

I have been right in all but 1 of our arguments thus far. By the way, Nvidia did go back and switch Volta to Pascal in the IBM supercomputer announcement, so in your face!

You were proven wrong in every argument ive participated in (cant say for the others, i havent checked them) by atleast 3 people at the time.

 

Please enlighten me, because according to this

 

 

Meanwhile on the GPU side, this supercomputer announcement marks the reintroduction of Volta by NVIDIA since going quiet on it after the announcement of Pascal earlier this year. Volta was then and still remains a blank slate, so not unlike the POWER9 CPU we don’t know what new functionality is due with Volta, only that it is a distinct product that is separate from Pascal and that it will be building off of Pascal. Pascal of course introduces support for 3D stacked memory and NVLink, both of which will be critical for these supercomputers.

 

 

It doesn't have the platform agnosticism that OpenCL does, but it is MUCH easier to program in and produce highly efficient code which OCL cannot touch.

 

Also, no, the test was uniform usage of an OpenCL benchmark suite, but thanks for playing instead of reading the article and assuming I'm always wrong. 

 

feel free to consult @LukaP since my word can apparently never be right. She will also tell you CUDA is superior in every single way from a performance and programmability viewpoint, and she would be right.

Its not superior in every single way. its actually very very inferior to oCL since its closed and proprietary. but in every other way, if you only program for NV cards, CUDA is superior to oCL by a large margin

[Opcode]

http://arxiv.org/pdf/1005.2581.pdf

Chew on that, a wide range of scientific computing and quantum simulation workloads. CUDA is universally better.

I don't want to base anything from an article dated from many years ago (not even going read it once I saw the date). Like I said let's provide a real-world example to see if you're words are as extensible as your mouth.

[patrickjp93]

You were proven wrong in every argument ive participated in (cant say for the others, i havent checked them) by atleast 3 people at the time.

 

Please enlighten me, because according to this

 

 

Its not superior in every single way. its actually very very inferior to oCL since its closed and proprietary. but in every other way, if you only program for NV cards, CUDA is superior to oCL by a large margin

I said from performance/programmability, but thank you for once again shoving words in my mouth.

 

Also, check out Nvidia's Facebook. anandtech will be updating shortly.

[patrickjp93]

I don't want to base anything from an article dated from many years ago (not even going read it once I saw the date). Like I said let's provide a real-world example to see if you're words are as extensible as your mouth.

It's from 2010!

[LukaP]

I don't want to base anything from an article dated from many years ago (not even going read it once I saw the date). Like I said let's provide a real-world example to see if you're words are as extensible as your mouth.

he will never do it. that would mean he would have to have some actual knowledge on the matter, which he doesnt. it would also mean he admits hes wrong, which he doesnt do. it would also mean him supporting his argument, which he doesnt ever do

[Cookybiscuit]

This is totally fake, everyone knows only Nvidia can make good GPU's.

[Opcode]

It's from last year you twit!

I sense someone's credibility is starting to run on empty. If you don't want to at least compile an example for utilizing CUDA then you're barking bigger words than you're capable of. Like I said I will compile the OpenCL equivalent (should be easier now with shared virtual memory) just throw me your source code. Of course I won't be able to benchmark using my machine as I run an ancient GPU. Tho finding someone to run it on modern hardware shouldn't be an issue here (it's LTT there's plenty of hardware here).