Intel's "Beacon" the World's Most Energy-Efficient Supercomputer

H

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
The latest Green500 list announced today named The National Institute for Computational Sciences' "Beacon" system as the world's most energy-efficient supercomputer. "Beacon" is powered by Intel Xeon Phi coprocessors 5110P and Intel Xeon processors E5-2670 and delivers nearly two-and-a-half billion floating-point operations per second per watt (2.49 GFlops/Watt) and peak performance of 112,200 gigaflops while consuming only 44.89 kW of power. Sean Koehl from Intel Labs shares his view in a blog on why this achievement is a significant milestone for Intel and Intel Xeon Phi coprocessor and what it takes to design the world's most energy-efficient supercomputer.
 
Is it supercomputer season? There seems to have been loads built in the last week or so. There must be some reason...
 
Actually, I believe this specific super computer was built specifically as a prototype to test/demonstrate the Xeon Phi.
 
Also of note... each Xeon Phi 5110P is capable of over 1.0TFLOPS so a good sized system but not huge. 40kW/ rack is not uncommon so guessing this is 1-2 racks maximum. Not exactly a monster sized system. That helps with keeping power requirements down.

BTW - the reason for all of the Supercomputer stuff in November is that the big show is on (Google SC12).
 
This stuff is getting out of hand, nearly all of the entities that can afford these outside of medical research scare me.
 
pjkenned said:
Also of note... each Xeon Phi 5110P is capable of over 1.0TFLOPS so a good sized system but not huge. 40kW/ rack is not uncommon so guessing this is 1-2 racks maximum. Not exactly a monster sized system. That helps with keeping power requirements down.

BTW - the reason for all of the Supercomputer stuff in November is that the big show is on (Google SC12).
Click to expand...

How do you get 333 amps at 120v into a single rack?
 
I was all like, "Intel has bacon?!" until I read it a second time and then I was back to being a sad Skribbel.
 
Jeremy C said:
Skynet needs to start somewhere...
Click to expand...

Skynet was only 90TFlops, but then again, that wasn't bad tech considering it was a ZISC based neural-network multi-processor architecture with an adaptive-learning capability. ;)

Yes, I'm entirely serious; at its core, Skynet was only 90TFlops.
Kind of shows you how far we've advanced in some areas... and have yet to advance in others.
 
You must log in or register to reply here.
Back
Top