CHIP DESIGNER Nvidia has announced what it claims is the fastest GPU accelerator high performance computing (HPC) applications.
Based on Nvidia's Kepler compute architecture, the Tesla K40 GPU accelerator is what Nvidia is calling "the world's highest performance accelerator ever built". It is designed for delivering higher graphics performance to science, engineering, and enterprise applications.
"Providing double the memory and up to 40 percent higher performance than its predecessor, the Tesla K20X GPU accelerator, and 10 times higher performance than today's fastest CPU, the Tesla K40 GPU is the world's first and highest-performance accelerator optimised for big data analytics and large scale scientific workloads," Nvidia said.
The Tesla K40 GPU accelerator boasts 12GB of ultra-fast GDDR5 memory so users can process datasets twice as big as its predecessor to analyse massive volumes of data.
The GPU supports 2,880 CUDA parallel processing cores to accelerate applications by up to 10 times compared to using a CPU alone, Nvidia claims. Dynamic parallelism technology enables GPU threads to spawn new threads so users can quickly crunch through adaptive data structures. The K20 GPU also features Nvidia's GPU Boost technology to control performance via benchmarking.
Nvidia's Tesla K20 will rival AMD's Firepro S10000 GPU that was announced last week and which also boasts 12GB of graphics memory. Designed for big data HPC workloads, the GPU was the "industry's first supercomputing server graphics card", AMD claimed.
The AMD Firepro S10000 12GB GPU features ECC memory plus DirectGMA support, which the company said allows developers working with large models and assemblies to use the parallel processing capabilities of AMD GPUs, which are based on the AMD's Graphics Core Next (GCN) architecture.
Nvidia touts the Tesla K20 as having computing speeds of 4.29 teraflops single-precision and 1.43 teraflops double-precision peak floating point performance.
Nvidia's Tesla K40 GPU is available from today, and in the coming months will appear in products from a variety of server manufacturers, including ASUS, Dell, HP and IBM.
Speaking of IBM, today it announced plans to collaborate with Nvidia on GPU accelerated versions of IBM's enterprise software applications on IBM Power Systems.
As the first time that GPU accelerator technology will be used in enterprise data centres as opposed to supercomputing, the collaboration will allow IBM customers to process, secure and analyse massive volumes of streaming data across scientific, engineering, big data analytics or other high performance computing (HPC) workloads. µ