Supercomputing Simulation Employs 156,000 Amazon Processor Cores

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
Need an extra 155,996 processor cores for a project? Give Amazon a call, they got you covered. :cool:

On Tuesday, the company announced a new record use of its technology: a simulation by Mark Thompson of the University of Southern California to see which of 205,000 organic compounds could be used for photovoltaic cells. The simulation drew upon the power of 156,314 processor cores at Amazon Web Services for 18 hours to test the chemicals with Schrodinger Materials Science software for computational chemistry, Cycle Computing said in a blog post.
 
So... did he end up finding which organic compounds could be used for photovoltaic cells?
 
Article doesn't indicate the flops they peaked at, would be interesting to do the conversion and see what fraction of the BTC network it was equivalent to.
 
Some of the comments are comedy.

"I've got an overclocked i7 with a Titan, these are great for supercomputers!"
 
Some of the comments are comedy.

"I've got an overclocked i7 with a Titan, these are great for supercomputers!"

Well, he is right that CUDA is an extremely powerful technology for use in complex computations and the Titan is the best at these calculations as an individual card. He's wrong that it outstrips 160,000 CPU cores, but for a lot of simulations, GPUs are much better suited. Data centers are not always properly equipped for these calculations, though at some you can pay a premium to use GPUs, even Amazon:

http://aws.amazon.com/gpu/
 
It's a matter of scale = 156,000 cores. Yes, Teslas are in many supercomputers. But an i7 and a Titan ... aren't.
 
Not impressed. Anyone with enough money can pay someone else for processors but it takes a totally amazing, super awesome cat person to actually own 155,996 processors. Dog people are disqualified just because their dog would make all that processor power pointlessly demand to be put outside so it can poop and then not have the decency to bury it.
 
So, if I hypothetically wanted to single-handedly outproduce the [H]orde at F@H, how many processors would I need, and how much would that cost me?
 
One of the more interesting security talks I've seen in the last couple years is using Azure for fuzzing tasks... Fuzzing is trivially parallel and is great for spanning over hundreds of instances. Scale out to reduce run durations giving up all of your instances when you're done can save a huge amounts of money and more importantly time.
 
Back
Top