Oak Ridge National Laboratory gets Kepler K20 "Big K"

[tviceman]

Perf/watt of Phi is not really great. Intel is not even really beating Fermi with Phi:

http://www.techpowerup.com/171436/I...e-Details-of-Intel-Xeon-Phi-Co-Processor.html

Your graph does show its beating Fermi.

More than two and a half years later, Intel was able to barely beat out 2009-2010 technology!

You mean the sites that uses SA etc as source? Or AT that just makes a guess?

So let me get this straight - are you round-about saying we should take your word over AT's? That you are a more accurate journalist than Anandtech is? That you know the specs and Anandtech does not?

 

[Jaydip]

That chart is not probably accurate in the first place.Check my earlier post.

 

[antihelten]

More than two and a half years later, Intel was able to barely beat out 2009-2010 technology!

And neither Intel nor Nvidia are capable of beating IBM*, interestingly enough.

*The Blue Gene/Q IBM systems are running at about 2100 MFlops/watt, and Titan should achieve about 1875 MFlops/watt assuming 20PFlops peak, 9 MW and similar efficiency (Rmax to Rpeak) as Jaguar.

 

[tviceman]

And neither Intel nor Nvidia are capable of beating IBM*, interestingly enough.

*The Blue Gene/Q IBM systems are running at about 2100 MFlops/watt, and Titan should achieve about 1875 MFlops/watt assuming 20PFlops peak, 9 MW and similar efficiency (Rmax to Rpeak) as Jaguar.

Interesting! I did not know that. Makes me wonder why IBM isn't crushing everyone in HPC then...

 

[antihelten]

Interesting! I did not know that. Makes me wonder why IBM isn't crushing everyone in HPC then...

Price and timing would be my guess, their hardware is possibly significantly more expensive, and the Blue Gene/Q system is quite new.

 

[exar333]

I would bet Oak Ridge wouldn't miss 1 or 2 of those big Ks...

 

[tviceman]

And neither Intel nor Nvidia are capable of beating IBM*, interestingly enough.

*The Blue Gene/Q IBM systems are running at about 2100 MFlops/watt, and Titan should achieve about 1875 MFlops/watt assuming 20PFlops peak, 9 MW and similar efficiency (Rmax to Rpeak) as Jaguar.

Interesting! I did not know that. Makes me wonder why IBM isn't crushing everyone in HPC then...

Well that didn't last very long! Titan has now been placed #1 on the top500 list.
http://www.anandtech.com/show/6446/nvidia-launches-tesla-k20-k20x-gk110-arrives-at-last

 

[BenSkywalker]

And neither Intel nor Nvidia are capable of beating IBM*, interestingly enough.

Top500 has been updated, Titan wins performance/watt by ~18%. Still extremely impressive for IBM to be going toe to toe with a much newer system.

 

[SirPauly]

NVIDIA GK110: Tesla K20 Launch Preview

http://www.rage3d.com/previews/video/nvidia_gk110_preview/

We inquired about SLI to be told that the board doesn't support it, but there are clearly two SLI connectors on the PCB in the promotional images supplied to us... oversight in the mockup or indication of GK110's future destination - the desktop?

 

[nabat]

Thats impressive.But why do you say CUDA compared to x86 is horrible?the learning curve is mostly same.

x86 is quite easy compared to CUDA. Search for Xeon phi

Xeon Phi 5110P. It will provide 1 TFLOPS with a 225 W that is called "acceptible" and usual. Xeon phi is quite power efficient. They are true CPUs which are very x86, unlike what some people said here. Only its SIMD is different, because it supports 512 bit vector instruction which main CPUs don't provide.
You can use openmp on Linux (no other OS supported now)..

Thanks!
​