Headwave Announces Full Support for NVIDIA® Tesla™ Compute TechnologyHeadwave announces full support for Nvidia Tesla(™) acceleration in all products.
Headwave announces full support for Tesla
When NVIDIA announced the Tesla compute technology in 2007, NVIDIA announced three products, the C870 compute board, the D870 deskside supercomputer and the S870 GPU server. Even then, Headwave applications were fully prepared to immediately leverage the Tesla technology for performance improvements. Today, all Headwave applications and modules are fully "GPU" accelerated, and leverages CUDA-enabled Quadro/Tesla for all computations.
Raising the Bar for InteractivityGeophysics, whether it is data processing, interpretation or reservoir characterisation requires substantial number-crunching.
In data processing, turn around time of months is the norm. In interpretation and rock physics, generating & accessing various attributes and interpreting huge volumes often have significant turn-around times. With the aid of Tesla, these workflows can be carried out faster, much faster. Equally important is it that teraflops of compute performance available to , it is possible to imagine working differently and smarter.
For the Technically MindedHeadwave takes a pragmatic approach to develolpment and industry deployment. On one end, technologies such as NVIDIA Tesla offer game-changing possibilities, yet initially not every client has recent hardware distributed throughout their asset teams.
For these reasons, Headwave typically offers a multi-level implementation - unique for the industry. What this means is that a given workflow, we provide a Tesla implementation and a CPU implementation, and in some cases also a OpenGL/GLSL based implementation. To explain further:
 |
The algorithms are implemented using Nvidia CUDA to take full advantage of NVIDIA Tesla high performance technology, if it is available at the workstation or cluster. Tesla-enabled workflows typically outperform CPU based workflows by a factor of 10 to 100 (or more)
|
|
if Tesla compute capabilities are not available, Headwave will try to fall back to an OpenGL/GLSL based implementation of the algorithm. OpenGL/GLSL algorithms rely on the graphics drivers and normally offer a significant speed up over the CPU based algorithms. Due to the complexity of many algorithms, not all algorithms can feasibly be implemented using GLSL.
|
|
We always offer an optimized, multi-core CPU based implementation as a last resort fallback. From an end user perspective, the above handled transparently by our architecture but the end user will be notified about the fall back and the potential effect on output.
|
Most of the core algorithms, at the architecture level, are implemented in all three variants in order to support a broad user base and choice of hardware.
|
|
|
