GPU in seismic processing, ‘don’t try this at home’

Headwave presentation reveals potential of GPU-based seismic compression.

Speaking at the Las Vegas SEG meet late last year, Headwave’s Steve Briggs provided an inside track on the company’s work on using graphics processing units (GPU) in seismic data analysis. Headwave’s focus is prestack volume visualization and analysis. Prestack visual quality control mandates quick look access to data sets of terabyte size. Headwave’s technology uses wavelet data compression to enable this to be performed on a notebook PC. This opens up access to processing that was hitherto only available to the processing center such as prestack horizon picking. The technology can run on GPU or CPU. With NVIDIA’s CUDA, programming the GPU is much easier but Briggs suggests, ‘don’t try this at home.’ While you can do a lot of ‘cool stuff’ with GPUs, ‘they are not going to handle full datasets.’ GPUs are good for testing but still present bottlenecks and roadblocks, and this is in the face of bigger datasets, now around a terabyte for 10 US offshore blocks prestack. Headwave uses GPUs to compress data for visualization and to accelerate attribute calculation. Current Gig Ethernet infrastructure is ‘painful’ as you need to move lots of bytes. Even compressed data sets can take weeks to get from tape. Storage is a problem, disk cache is useless for these volumes – you need ‘RPM s,’ lots of spindles and multiple 10GigEthernet, Infiniband pipes. Compression needs large JBOD arrays and maybe SSD. But investment in infrastructure bandwidth really does pay in terms of visual and computational bandwidth. More from or

Click here to comment on this article

Click here to view this article in context on a desktop

© Oil IT Journal - all rights reserved.