SMX design

305835SMX design//images10.newegg.com/BizIntell/item/14/132/14-132-008/14-132-008_h1.jpgSMX (streaming multiprocessor) design delivers up to 3x more performance per watt compared to the SM in Fermi*. It also delivers one petaflop of computing in just ten server racks. (*Based on DGEMM performance: Tesla M2090 = 410 gigaflops, Tesla K20 (expected) > 1000 gigaflops)305835305835Dynamic Parallelism capability//images10.newegg.com/BizIntell/item/14/132/14-132-008/14-132-008_h2.jpgDynamic Parallelism capability enables GPU threads to automatically spawn new threads. By adapting to the data without going back to the CPU, it greatly simplifies parallel programming. Plus it enables GPU acceleration of a broader set of popular algorithms, like adaptive mesh refinement (AMR), fast multipole method (FMM), and multigrid methods.305835305835Hyper-Q//images10.newegg.com/BizIntell/item/14/132/14-132-008/14-132-008_h3.jpgHyper-Q feature enables multiple CPU cores to simultaneously utilize the CUDA cores on a single Kepler GPU. This dramatically increases GPU utilization, slashes CPU idle times, and advances programmability—ideal for cluster applications that use MPI.305835305835ECC memory error protection//images10.newegg.com/BizIntell/item/14/132/14-132-008/14-132-008_h4.jpgMeets a critical requirement for computing accuracy and reliability in data centers and supercomputing centers. Both external and internal memories are ECC protected in Tesla K20 and K20X.305835305835L1 and L2 caches//images10.newegg.com/BizIntell/item/14/132/14-132-008/14-132-008_h5.jpgAccelerates algorithms such as physics solvers, ray-tracing, and sparse matrix multiplication where data addresses are not known beforehand.305835305835Asynchronous transfer with dual DMA engines//images10.newegg.com/BizIntell/item/14/132/14-132-008/14-132-008_h6.jpgTurbocharges system performance by transferring data over the PCIe bus while the computing cores are crunching other data.305835

Questions & Answers