Welcome to the Technical Overview of Ansys Lumerical 2025 R1
This release focuses on GPU performance enhancements, optimizing simulation speed, memory efficiency, and resource control, which are crucial for accelerating complex photonics designs.
Key Features and Updates
- FDTD Tab and Sweeps Panel:
- Seamless configuration of CPU, GPU, or multi-GPU jobs.
- Select resources directly and switch hostnames for local or remote job launching.
- Use a dropdown to assign GPUs for sweeps.
- Multi-GPU Support:
- Includes both a custom mode for manual assignment and an auto mode that uses all available resources.
- GPUs can be renamed or selectively enabled, ideal for shared environments or controlled cluster use.
- GPU Memory Estimation:
- Performed before meshing begins, allowing you to know in advance if a model is likely to fail due to memory limits.
- Saves compute time and reduces trial and error cycles.
- Simulation Checker:
- Provides detailed reports and highlights misconfigurations for both CPU and GPU setups.
- Helps prevent runtime crashes.
- Meshing Enhancements:
- Significantly accelerated using multi-threading and a hierarchical decomposition approach.
- For flat metastructures like metalenses, meshing time has been reduced by about 20%.
- Mesh scaling follows an N log N over T trend, improving efficiency as model complexity or thread count increases.
- FDTD Simulations:
- Using PML boundaries now uses 50% less GPU memory with no loss in accuracy.
- Enables larger systems to run on existing hardware with a slight boost in performance.
- Up to 10% speed improvements in small to mid-size models, benchmarked using a dipole source in a PML box.
- Data Distribution:
- Lumerical now distributes meshing data more efficiently across multiple GPUs.
- By asynchronously transferring data and intelligently slicing geometry along the X or Y axis, near-linear scaling is achieved.
- Especially effective on NVIDIA A100 clusters, with benchmarks showing up to 80% NVLink bandwidth utilization.
- 8 GPUs provide 8X acceleration for metalens simulations.
- Solver Support:
- Now supports GPU simulations for perfect electrical conductor materials, dispersive models, dipole sources, and ZBF import data.
- Unlocks accelerated workflows for CMOS sensors, OLEDs, and co-packaged optics.
- Eliminates the need for GPU express mode.
- Expands GPU simulation to a broader set of photonic design scenarios without compromising accuracy.
Contact Information
Thank you for watching. For more information, consulting, or technical mentoring on how to implement these features in your workflow, reach out to the experts at Ozen Engineering, Inc.
Contact us at [email protected].
Welcome to this technical overview of what's new in Ansys Lumerical 2025 R1, focused on GPU performance enhancements. This release includes updates that optimize simulation speed, memory efficiency, and resource control, crucial for accelerating complex photonics designs.
The FDTD tab and sweeps panel now allow seamless configuration of CPU, GPU, or multi-GPU jobs. You can select resources directly, switch hostnames for local or remote job launching, and use a dropdown to assign GPUs for sweeps.
Multi-GPU support now includes both a custom mode for manual assignment and an auto mode that uses all available resources. GPUs can be renamed or selectively enabled, ideal for shared environments or controlled cluster use. GPU memory estimation is now performed before meshing begins.
This means you'll know in advance if a model is likely to fail due to memory limits, saving compute time and reducing trial and error cycles. The simulation checker provides detailed reports and highlights misconfigurations for both CPU and GPU setups, helping prevent runtime crashes.
Meshing has been significantly accelerated using multi-threading and a hierarchical decomposition approach. The following is the Ansys simulation approach. For flat metastructures like metalenses, meshing time has been reduced by about 20%.
Mesh scaling now follows an N log N over T trend, improving efficiency as model complexity or thread count increases. FDTD simulations using PML boundaries now use 50% less GPU memory with no loss in accuracy.
This enables larger systems to run on existing hardware with a slight boost in performance. You'll see up to 10% speed improvements in small to mid-size models, as benchmarked using a dipole source in a PML box. Lumerical now distributes meshing data more efficiently across multiple GPUs.
By asynchronously transferring data and intelligently slicing geometry along the X or Y axis, near-linear scaling is achieved, especially on NVIDIA A100 clusters. Benchmarks show up to 80% NVLink bandwidth utilization, with 8 GPUs providing 8X acceleration for metalens simulations.
The solver now supports GPU simulations for perfect electrical conductor materials, dispersive models, dipole sources, and ZBF import data. This unlocks accelerated workflows for CMOS sensors, OLEDs, and co-packaged optics, while eliminating the need for GPU express mode.
These additions expand GPU simulation to a broader set of photonic design scenarios without compromising accuracy. Thank you for watching. For more information, consulting, or technical mentoring on how to implement these features in your workflow, reach out to the experts at Ozen Engineering.
Contact us at [email protected].
