Tesla Kepler Family Product Overview - Nvidia

Tesla ® Kepler GPU AcceleratorsIntroducing the world’s fastest accelerators.NVIDIA Tesla K-series GPU Accelerators are based on the NVIDIA Kepler compute architecture and powered by CUDA ® , the world’smost pervasive parallel computing model. They include innovative technologies like Dynamic Parallelism and Hyper-Q to boostperformance as well as power efficiency and deliver record application speeds for seismic processing, biochemistry simulations,weather and climate modeling, image, video and signal processing, computational finance, computational physics, CAE, CFD, and dataanalytics.The innovative Kepler computearchitecture design includes:SMX (streaming multiprocessor) designthat delivers up to 3x more performanceper watt compared to the SM inFermi 1 . It also delivers one petaflop ofcomputing in just ten server racks.Dynamic Parallelism capability thatenables GPU threads to automaticallyspawn new threads. By adapting to thedata without going back to the CPU, itgreatly simplifies parallel programming.Plus it enables GPU acceleration of abroader set of popular algorithms, likeadaptive mesh refinement (AMR), fastmultipole method (FMM), and multigridmethods.Hyper-Q feature that enables multipleCPU cores to simultaneously utilize theCUDA cores on a single Kepler GPU. Thisdramatically increases GPU utilization,slashes CPU idle times, and advancesprogrammability—ideal for clusterapplications that use MPI.The Tesla K-series family of products includes:Tesla K10 GPU Accelerator – Optimized for singleprecision applications, the Tesla K10 includes twoultra-efficient GK104 Kepler GPUs to deliver highthroughput. It delivers up to 2x the performancefor single precision applications compared to theprevious generation Tesla M2090 GPU in the samepower envelope. With an aggregate performance of4.58 teraflop peak single precision and 320 gigabytesper second memory bandwidth for both GPUs puttogether, the Tesla K10 is optimized for computationsin seismic, signal image processing, and videoanalytics.Tesla K20 and K20X GPU Accelerators – Designedto be the performance leader in double precisionapplications and the broader supercomputing market,the Tesla K20 and K20X GPU Accelerators deliver 10xthe performance of a single CPU 2 . Tesla K20 and K20Xboth feature a single GK110 Kepler GPU that includesthe Dynamic Parallelism and Hyper-Q features.With more than one teraflop peak double precisionperformance, these GPU accelerators are ideal forthe most aggressive high-performance computingworkloads including climate and weather modeling,CFD, CAE, computational physics, biochemistrysimulations, and computational finance.1Based on DGEMM performance: Tesla M2090 = 410 gigaflops, Tesla K20 (expected) > 1000 gigaflops2Based on WS-LSMS performance comparison between single E5-2687W @ 3.10GHz vs single Tesla K20X. Tesla K20X > 650 gigaflopsNVIDIA Tesla K-Series | Datasheet | Oct12