Acceleware has just announced a seismic data processing ‘accelerator,’ a graphics-processing unit (GPU)-based hardware/software add-on that ‘outsources’ number crunching to an array of GPUs. The company has developed a Kirchhoff pre-stack time migration seismic processing library that is tuned for the NVIDIA Tesla GPU hardware.
Benchmark
We asked Acceleware for some benchmarks. This is what they came up with. Adding a 2 GPU* accelerator to a single server with two dual core processors (4 core total) speeds up a complete Kirchhoff time migration by a factor of 8. This speedup includes the preprocessing which is done primarily on the CPU.
8 GPU
A slightly more obscure benchmark showed a 45 times Kitchoff speed up running on ‘an 8 GPU solution with 2 servers each with two dual core processors (8 core total)’ as compared with a single core machine.
* Likely an NVIDIA Tesla ClusterInABox.
© Oil IT Journal - all rights reserved.