AMD FirePro awarded top spot in Green500 list

[CAT-THE-FIFTH]

It appears a supercomputer powerered by FirePro S9150 compute cards is now top of the Green500 list:

http://www.sys-con.com/node/3237476
http://www.marketwatch.com/story/am...green500-list-2014-11-20?reflink=MW_news_stmp

 

[humbug]

It appears a supercomputer powerered by FirePro S9150 compute cards is now top of the Green500 list:

http://www.sys-con.com/node/3237476
http://www.marketwatch.com/story/am...green500-list-2014-11-20?reflink=MW_news_stmp

NEW ORLEANS, LA -- (Marketwired) -- 11/20/14 -- AMD (NYSE: AMD) powers the L-CSC cluster at the GSI research facility achieving the number one position for supercomputers on the latest Green500™ List, a ranking of the 500 most energy-efficient supercomputers in the world,

Number #1, nice

 

[Orangey]

AMD confirmed for better efficiency than Maxwell.

 

[humbug]

AMD confirmed for better efficiency than Maxwell.

Maxwell's Boost technology has nothing to adjust in Compute heavy tasks so the power efficiency remains the same as Kepler.

It seems that perhaps AMD have done it at an engendering level rather than firmware trickery.

 

[drunkenmaster]

There is a reason AMD is using generally a bit more power. Architecture wise Hawaii is an absolute compute monster. It's a 1/2DP throughput beast where as Titan is only a 1/3DP throughout design. Hawaii spanks the Nvidia equivalent in DP throughput on price, power and absolute performance.

A S9150 offers very similar DP performance to a Titan-Z or K80(2xGK210).

As in S9150 or professional Hawaii gives about 2.5Tflops DP , Titan-Z produces about 2.7TF and the new dual gpu K210 based card is only rated at 2.9TF DP.

Hell the K80 Tesla which is 2x K210 only does 8.8TF single precision with it's 2.9TF DP. Desktop 295x2 provides almost 11.5TF single precision but is DP limited(driver/not architecture). if they released a professional version it would be say 11TF SP and 5.5TF DP performance, almost double what Nvidia's new Tesla card can provide.

For a "compute" heavy card Nvidia is both bandwidth and architecture limited this generation.

 

[Kaapstad]

Can it run Mantle without running out of memory ?

 

[humbug]

Can it run Mantle without running out of memory ?

The 8GB one can

 

[layte]

There is a reason AMD is using generally a bit more power. Architecture wise Hawaii is an absolute compute monster. It's a 1/2DP throughput beast where as Titan is only a 1/3DP throughout design. Hawaii spanks the Nvidia equivalent in DP throughput on price, power and absolute performance.

A S9150 offers very similar DP performance to a Titan-Z or K80(2xGK210).

As in S9150 or professional Hawaii gives about 2.5Tflops DP , Titan-Z produces about 2.7TF and the new dual gpu K210 based card is only rated at 2.9TF DP.

Hell the K80 Tesla which is 2x K210 only does 8.8TF single precision with it's 2.9TF DP. Desktop 295x2 provides almost 11.5TF single precision but is DP limited(driver/not architecture). if they released a professional version it would be say 11TF SP and 5.5TF DP performance, almost double what Nvidia's new Tesla card can provide.

For a "compute" heavy card Nvidia is both bandwidth and architecture limited this generation.

Yet for all its supposed failings. Nvidia is raking in the super computer design wins, has exclusive contracts with the likes of IBM and have just announced what will be the fastest super computer on the planet.

Pretty good going. It's almost as if the headline grabbing theoretical maximum flop numbers only paint a small part of the picture. Seems being able to make use of that performance in real world HPC situations counts for something.