Thursday, 30 January 2014

GeoTeric processing

The most frequently used processes and workflows in GeoTeric utilise a workstation’s Graphics Processing Unit (GPU) when carrying out computationally intensive processing.

GPU Accelerated Volume Processing refers to using a GPU to run a GeoTeric process over a 3D volume, create the results and write the data to disk, just like regular processing using the CPU in GeoTeric would do.

To do this we utilise a core feature of NVIDIA GPUs called CUDA (Compute Unified Device Architecture). GeoTeric will use all available “CUDA enabled” GPUs in the workstation in order to process as long as those GPUs have been enabled in Tools>Options>Accelerated Computing. These GPU capabilities are fully integrated into the GeoTeric Batch Processor and workflows and are activated automatically when a user runs one of GeoTeric’s “CUDA enabled” processes/workflows.

All graphics cards that are currently certified on GeoTeric support CUDA processing- you only need to ensure graphics card driver version is up to date or you are using our certified driver and that “Accelerated Computing” is enabled in GeoTeric. It’s enabled by default but to check the setting go to Tools > Options > Accelerated Computing.

Further performance boost can be given if you install a Tesla card as a Secondary card. Tesla cards are for processing only and have no video out (i.e. cannot be used for Rendering so need to be used in conjunction with another card).

 How much faster is the golden question- this is very much linked to your graphics card/s and system setup so it can be difficult to give general figures- however if you would like more information on this please contact support.

Not all of the algorithms in the GeoTeric processes are suitable for CUDA enabled processing so other parts of GeoTeric use CPU multi-threading. Below is a table outlining what parts are multi-threaded or CUDA enabled. In order to set the number of threads to use- go to Tools > Options > CPU’s and enter the value here. It is worth noting that the threads available to GeoTeric are managed by the Operating System itself. GeoTeric can only advise on how many processes it can use. 

CUDA Enabled Processes
Attributes
DipAzi
Chaos
DipAzi Combine
Fault Enhance
Instantaneous Phase
RGB Segmentation
RGB Export
SO Discontinuity
SO FMH
SO Noise Filter
Smoothing
TDiffusion
Tensor
Volume Combo

CUDA enabled Workflows / Tools
Frequency Decomposition
Adaptive Geobodies
Multi-threaded Processes in the Process Tree
SO FMH
SO Dip Derivative
SO Noise Filter
GO Semblance
SO Dip Derivative
Volume Combo
SO Discontinuity
SO Discontinuity
Voxel Math
Body Morphology
Parser
CookieCutter
DipAzi Combine
Surface
Attributes
SurfaceIn
Bedform indicator
SkinIn
StdDev
Fault Enhance
Channel Edge
Fault Detect
Continuity
Fault Density
Texture
Fault In
Differential
Convert
Smoothing
Time volume
Flexure
SO Semblance
Partially multi-threaded processes
(the following processes consist of several steps where at least 1 or more of the steps have been multi-threaded)
TDiffusion
Eigen Analysis
DipAzi
Tensor
Flexure
SO Semblance
Chaos
Thickness
Conformance
Fault Trends
Deformation
Border

No comments:

Post a Comment