Super ComputersNot just for the Big BoysSuper ComputerHow “Fast” is FASTSuper Performance Over TimeSingle vs DoubleSingle vs Double cont.GPUs catching up to (and surpassing?) CPUsNVIDIA 280 vs Intel dual coreBuy a pre-built NVIDIA TeslaOr make your own with an NVIDIA CoreYour personal Super Computer won’t make the top 500, but it is equivalent to the top of the line in 2000 for less than $10,000ReferencesPersonalChris WardCS147 Fall 2008Recent offerings from NVIDA show that small companies or even individuals can now afford and own Super Computers.Just what is a “Super Computer”?“A supercomputer is a computer that is at the frontline of current processing capacity, particularly speed of calculation. “ –wikipediaThe term supercomputer itself is rather fluid, and today's supercomputer tends to become tomorrow's ordinary computer.In computing, FLOPS (or flops or flop/s) is an acronym meaning FLoating point Operations Per SecondToday’s two fastest Super Computers operate at > 1 petaflop/s (represents one quadrillion floating point operations per second. )Most applications use double precision math for the following reasons:1. To minimize the accumulation of round-off error,2. For ill-conditioned problems that require higher precision3. The 8 bit exponent defined by the IEEE floating point standard for 32-bit arithmetic will not accommodate the calculation, or4. There are critical sections in the code which require higher precision.Some of the latest GPUs and CPUs have LARGE differences in flops for single vs double precisionResearchers are recognizing this trend and re-examining their algorithms to take advantage of single precision whenever possibleGPU Graphics Processing UnitCPU Central Processing unitDevice Transistor Count Year ReleasedMakerRV770 956 000 000 2008AMDGT200 1 400 000 000 2008NVIDIADual-Core Itanium 2 1 700 000 000 2006IntelQuad-Core Itanium Tukwila 2 000 000 000 2008IntelTesla S1070 Computer is a four-teraflop 1U system powered by the world’s first one-teraflop processor.Each of the 4 cores contains 240 GPUs (960 total)Single Precision floating point performance (peak) 3.73 to 4.14 Tflops Double Precision floating point performance (peak) 311 to 345 GflopsStarting at about $3,995 to ~ $10,000•1 Board= 240 Processing Cores•~ 1 TFlop performance•4GB GDDR3 RAM•PCI-Express 2.0, • Write your code in C and run in Windows or Linux•$1,699.99F
View Full Document
Unlocking...