Topic: Episode #262

[Beep Boop Boop]

You can expect to pay a bit over $20,000 for a top system.  Still pretty good for a desktop super-computer.

What the hell do you get for $20,000?  Mind I am asking because I'm curious as to what the difference between our super-computer is from a top system. 

[Citizen Skeptic]

the real super computers that I've seen have been large dedicated boxes with at least 1000 nodes. ncar has a few supercomputers. a friend of mine used to operate one at caltech that has 1000 or maybe more nodes. they are very cool when they run because of the node activity indicators on the front. my friend said he eventually could tell who was using the computer by the light show.

[Mormegil]

You can expect to pay a bit over $20,000 for a top system.  Still pretty good for a desktop super-computer.

What the hell do you get for $20,000?  Mind I am asking because I'm curious as to what the difference between our super-computer is from a top system.

NVidia had prices between the mid $3000 to just over $20,000 listed.

This might be a typical pre-built system:  http://www.supermicro.com/products/system/4U/7046/SYS-7046GT-TRF.cfm

One of the reseller's top product (probably around $20,000), has their top GPU server doing 8.24 Teraflops (in single precision) with 8 NVidia cards ( http://www.amax.com/CS_GPUserverDetail.asp?cs_id=Xn-F4101G )

They've also got a workstation that looks more like a PC, with about half the performance (guessing 4 NVidia cards - http://www.amax.com/CS_GPUpscDetail.asp?cs_id=FSC2n&Type=Clusters&value=GPU%20Workstations )

[Citizen Skeptic]

This was in 2004.

NASA's New Supercomputer Powered by Intel® Itanium® 2 Processors

At an event marking a significant milestone for Intel and the scientific research community, the U.S. National Aeronautics and Space Administration (NASA) unveiled its newest supercomputer, the Columbia, which is powered by 10,240 Intel® Itanium® 2 processors. Several Intel executives joined Intel President and Chief Operating Officer Paul Otellini at the ribbon-cutting ceremony at NASA Ames Research Center in Mountain View, California, home of the new supercomputer.

NASA's 10,240-processor Columbia supercomputer is built from 20 Altix systems, each powered by 512 Intel Itanium 2 processors. (Image courtesy of Silicon Graphics, Inc.)
The powerful system, built by Silicon Graphics Inc. (SGI), promises to revolutionize the rate of scientific discovery at NASA. Using only 16 of its 20 installed systems, Columbia achieved sustained performance of 42.7 trillion calculations per second, or 42.7 teraflops.

"If you could do one calculation per second by hand, it would take you a million years to do what this machine does in a single second," said G. Scott Hubbard, Ames' director.

The new performance record eclipses the performance of every supercomputer operating today and knocks NEC's Earth Simulator off its perch as the world's No. 1 machine—a spot it has held since 2002.

It's a beautiful thing.

[Citizen Skeptic]

And I think this is still the fastest:

Jaguar

XT5 Partition
The XT5 partition contains 18,688 compute nodes in addition to dedicated login/service nodes. Each compute node contains dual hex-core AMD Opteron 2435 (Istanbul) processors running at 2.6GHz, 16GB of DDR2-800 memory, and a SeaStar 2+ router. The resulting partition contains 224,256 processing cores, 300TB of memory, and a peak performance of 2.3 petaflop/s (2.3 quadrillion floating point operations per second).

XT4 Partition
The XT4 partition contains 7,832 compute nodes in addition to dedicated login/service nodes. Each compute node contains a quad-core AMD Opteron 1354 (Budapest) processor running at 2.1 GHz, 8 GB of DDR2-800 memory (some nodes use DDR2-667 memory), and a SeaStar2 router. The resulting partition contains 31,328 processing cores, more than 62 TB of memory, over 600 TB of disk space, and a peak performance of 263 teraflop/s (263 trillion floating point operations per second).
The SeaStar2+ router (XT5 partition) has a peak bandwidth of 57.6GB/s, while the SeaStar2 router (XT4 partition) has a peak bandwidth of 45.6GB/s. The routers are connected in a 3D torus topology, which provides an interconnect with very high bandwidth, low latency, and extreme scalability.

The operating system for Jaguar is the Cray Linux Environment. This consists of a full-featured version of Linux on the service nodes and the Compute Node Linux micro-kernel on the compute nodes. The micro-kernel is designed to minimize partition overhead allowing scalable, low-latency global communications.

Now that's a computer! 

[stands2reason]

What I'm more impressed by is the programming and compiler/language design necessary to actually utilize that kind of hardware.

[MisterMarc]

Cool, but doesn't distributed computing like BOINC put all the supercomputers to shame for sheer processing power? I seem to recall reading that.