5 Questions About Reservoir Simulation Software for Stone Ridge Technology

Dr. Karthik Mukundakrishnan (SRT Director of R&D) recently presented a webinar hosted by NVIDIA, highlighting the strengths and features of ECHELON, our ultra-fast, ultra-scaleable GPU-based reservoir simulation software.

1. How large a model might you be able to run on a single GPU?

This varies by model complexity and the GPU being used, but up to a 10 million active cells can be run on a single NVIDIA A100 and up to 6 million active cells can fit on most low cost workstation GPUs.

2. How is pre & post processing handled in ECHELON software? How well does it work with 3rd party tools?

ECHELON reservoir simulation software works with legacy simulator and industry-standard input and output and can be used with Petrel, RE-Studio, S3, Tecplot, Tempest View and more.

3. A lot of simulations are still conducted on workstations rather than on clusters - does the GPU still have an edge over the CPU in those setups?

Yes absolutely! Their small hardware footprint makes GPUs a perfect solution for simulations being performed on workstations. By using a single low cost GPU workstation, ECHELON reservoir simulation software easily outperforms other modern simulators using just the CPU.

4. How well can you model unconventional wells?

Marathon Oil has been utilizing ECHELON software for the simulation of unconventionals for about 6 years, routinely running models in the tens of millions of cells on a modest number of GPU-based cluster nodes.

5. Can ECHELON software take advantage of multiple GPUs in a server? Can it run across multiple nodes? If so, what are the requirements for the interconnect?

Yes! Multi-GPU and multi-node support was designed into ECHELON software from its very inception in order to handle the largest of simulation models. ECHELON software can utilize all the GPUs in a server and can run across multiple nodes using MPI communication. Performance is maximized when using Infiniband from Mellanox which allows data to be transferred between nodes without passing through CPU memory.

If you missed our webinar with NVIDIA, you can access it here.