GPU:MassivelyParallelCoprocessornVIDIAAGPUisCoprocessortotheCPUorHostHasitsownDRAMRuns1oo0softhreadsinparallelSinglePrecision:4.58TFlop/sDoublePrecision:1.31TFlop/s
© NVIDIA Corporation 2013 GPU: Massively Parallel Coprocessor A GPU is Coprocessor to the CPU or Host Has its own DRAM Runs 1000s of threads in parallel Single Precision: 4.58TFlop/s Double Precision: 1.31TFlop/s
Heterogeneous Parallel ComputingnVIDIACPULogicoComputeoLatency-OptimizedFastSerial Processing
© NVIDIA Corporation 2013 Heterogeneous Parallel Computing Latency-Optimized Fast Serial Processing Logic() Compute()
Heterogeneous Parallel ComputingDVIDIAGPUCPULogicoComputeoLatency-OptimizedThroughput-OptimizedFastSerialProcessingFastParallelProcessingnn2053
© NVIDIA Corporation 2013 Heterogeneous Parallel Computing Latency-Optimized Fast Serial Processing Throughput-Optimized Fast Parallel Processing Logic() Compute()
Heterogeneous Parallel ComputingnVIDIAGPUCPULogicoComputeoLatency-OptimizedThroughput-OptimizedFastSerialProcessingFastParallelProcessing
© NVIDIA Corporation 2013 Heterogeneous Parallel Computing Latency-Optimized Fast Serial Processing Throughput-Optimized Fast Parallel Processing Logic() Compute()
GPU in Computer SystemnVIDIAPCleDDR3CPUDRAMConnectedtoCPU chipsetbyPCle16GB/sOneWay,32GB/sinbothway
© NVIDIA Corporation 2013 GPU in Computer System Connected to CPU chipset by PCIe 16GB/s One Way, 32GB/s in both way PCIe CPU DRAM DDR3