Amazon Launches High Performance Cloud â€“ Hackers in Love
|May 18, 2012||Posted by admin||
Calling it a “nuclear-powered bulldozer”, yesterday, Amazon announced and blogged about its newest cloud infrastructure service, the “Cluster GPU Instance”, which delivers supercomputer calculation power for as little as $2.10 per hour.Â The new instance type employs the same NVIDIA Tesla processor used in three of the five fastest supercomputers.Â It is rated at 515 gigaflops (515 billion double-precision floating point calculations per second) and each Amazon instance employs two of them, giving each instance more than one teraflop of processing power.Â Amazon further allows instances to be clustered “up through and above 128 nodes” for even more power.Â
Theoretically, a 128-node cluster of the new Amazon EC2 instances would qualify as the 50th fastest computer in the world.Â The new instance type enables a wide variety of calculation-intensive workloads for applications that include energy exploration, weather prediction, graphics rendering, and video transcoding.Â And, oh, it is also good for enabling encryption code breaking and identity theft.
I doubt Amazon’s founder Jeff Bezos or its web services evangelist Jeff Barr are feeling like Oppenheimer did when he witnessed the first test of his creation, the atom bomb, and famously quoted Hindu scripture about becoming “the destroyer of worlds”, but maybe they should feel that way.Â This is a big moment for good and evil.
First, the Good
Not long ago, computing power of this magnitude was a very precious resource, only available to the large, well-funded companies, government agencies, and academic institutions who could afford to buy and manage expensive supercomputers from companies like Cray and IBM.
Then, things started to change with the growth of computer gaming, scientific visualization, computer animation, and media streaming, which drove the development and volume production of processing chips called “graphics processing units” (GPUs) by companies like NVIDIA and ATI.
In their namesake application of graphics processing, GPUs perform the complex “floating point” decimal arithmetic needed to render and manipulate highly detailed graphics and photorealistic computer-generated imagery, or “CGI”.Â (Conventional CPU chips, like the x86, ARM, and others, can only perform integer arithmetic “in hardware”, making them ill-suited for efficient graphics processing.)
But, there are many other, non-graphical applications that also require floating point calculations and it was not long before GPUs were being used as mathematical co-processors in high-end scientific workstations and aggregated in servers.Â Although they are much cheaper than first-generation super computers, these systems are still quite expensive, with workstations costing $10K and up and servers going for multiples of that.
Yesterday that all changed.Â Now the tiniest company, even the lone quant, can have the same computational power, for as little as $2.10 per hour, with no up-front investment and access from anywhere in the world.
Amazon Cluster GPU Instances can lower the cost and accelerate the progress of fighting famine and disease, building safer, more fuel efficient vehicles and aircraft, finding and exploiting new sources of energy, and, of course, producing breathtaking visual entertainment.Â We have not yet begun to imagine the new businesses and research projects this kind of cloud computing will make possible.
In their blog entry entitled “A Couple More Nails in the Coffin of the Private Compute Cluster” large scale computing specialists Cycle Computing provide a very detailed picture of how they have used this technology to build a value-added computation service in the Amazon public cloud to support these kinds of applications.