IMEX ON GPU
CMG-IMEX 2025.30/31 benchmarked on NVIDIA A30 and A100 against CPU-only runs, across UNISIM-I-R and UNISIM-II-R reservoir models — run independently by UNISIM, with no involvement from CMG.
UNISIM-I-R
Single porosity · single permeability · Black Oil
UNISIM-II-R
Dual porosity · dual permeability · Black Oil
GPU A30 · 24GB
% reductionGPU A100 · 80GB
% reductionGPU A30 · 24GB
-
8 cores 2.9× – 3.3×
-
12 cores 2.6× – 2.7×
-
16 cores 2.4× – 2.8×
-
20 cores 2.1× – 2.7×
GPU A100 · 80GB
-
8 cores 3.0× – 3.3×
-
12 cores 2.9× – 3.0×
-
16 cores 2.7× – 2.9×
-
20 cores 2.6× – 2.8×
01 | 53–70% runtime reduction on GPU versus CPU, consistent across both models and all core counts tested. | |||||||||||||||||||
02 | The largest, undivided MIG profile wins on both cards: 7g.80gb on the A100, 4g.24gb on the A30. | |||||||||||||||||||
03 |
Larger models benefit more — UNISIM-I-R’s 3.4M active blocks saw stronger GPU gains than the smaller UNISIM-II-R case. | |||||||||||||||||||
“
The simulator demonstrated excellent performance, achieving up to 70% time reduction — recommended for users who need to accelerate simulation for models with a large number of cells.
UNISIM is the Reservoir Simulation and Management research group based at CEPETRO, the Center for Petroleum Studies at UNICAMP (University of Campinas), Brazil. The group develops open benchmark models — including UNISIM-I-R and UNISIM-II-R — and runs independent performance studies on the simulators used across the industry.
This evaluation was conducted independently by UNISIM, with no involvement, funding, or review from CMG. Learn more at unisim.cepetro.unicamp.br.
