RTM Performance
RTM Performance Numbers
The following table shows our AxRTM performance numbers. Feel free to use the links to review more detailed information below!
* on a single GT200 GPU
** on an NVIDIA Tesla C870 GPU
Detailed Result Brochures (html):
SEG Salt Model A1 Performance Results Brochure
SEG Salt Model A2 Performance Results Brochure
Marmousi 2D Performance Results Brochure
Foothills 94 Performance Results Brochure
Results Summary
SEG Salt Model
 |
- Industry standard synthetic dataset
- Represents complex marine salt-body geology where RTM is often used
- 13.5 x 13.5 x 4.2 km volume with 20m grid spacing
- 2D and 3D synthetic surveys are available
- More information visit
research.seg.org/3dmodel/salthome/intro.html
|
SEG Salt Phase A1 Survey 3D
 |
- 3D narrow-azimuth marine acquisition simulation
- 138 shots at 80m spacing
- 6 streamers of 65 receivers at 40m spacing from 160m to 2760m offset
AxRTM Performance Result:
- 149 seconds per shot on a single GT200
|
Salt A1 Velocity Model
 |
Salt A1 Reverse Time Migration
 |
SEG Salt Phase B Survey 3D
 |
- 3D survey over entire model volume of 13.5km x 13.5km x 3.7km depth
- 35 shot lines at 320m spacing
- Wide-azimuth acquisition
AxRTM Performance Result:
- 20m grid spacing results in 87 million cell volume
- Processing is underway
- Less than 10 minutes per shot on a single GT200 GPU
|
Marmousi Model - 2D
 |
|
 |
- Migration volume of 9.2km x 3km depth with 4m grid spacing
AxRTM Performance Result:
- 239 shots at 44.8 seconds per shot on a single GT200 GPU
|
"Foothills 94" Dataset
Hard-edged velocity model is the "perfect" input to the migration.
|
2D RTM image of Foothills model using hard velocity model produces the best possible image.
|
Smoothed Model
|
Smoothed model minimizes the effects of errors in the velocity model that are always present when building a model from field data.
|
2D RTM image of Foothills model using smoothed velocity model produces an excellent image.
|
|
The "Foothills 94" dataset is a synthetic dataset modeling the folded and faulted "thrust belt" geology typical of the Foothills region of Western Canada. The 2D model spans 25km to a depth of 10km and has a vertical relief of approximately 1.5km.
The synthetic data consists of 277 shot records with 5 seconds of trace data. For migration, we used a model size of 1700x1000 cells and a 25Hz source wavelet.
More information: http://software.seg.org/datasets/2D/Model_1994
|
The large amount of vertical relief at the surface present special data processing challenges which Reverse Time Migration handles seamlessly.
These Reverse Time Migration images feature highly accurate reflector positions, 90-degree and overhanging reflectors, low noise (coherent and random), reflectors imaged right to surface, and constant reflector amplitudes. These features are difficult or impossible to achieve with simpler, lower-cost migration algorithms.
AxRTM Performance Results:
Computation time for 10,000 timesteps was 1 minute 10 seconds per shot running on an NVIDIA Tesla C870 GPU.
|
|