Reservoir simulation is one of the most computationally demanding jobs in subsurface engineering. Engineers push hundreds of thousands — often millions — of grid cells through nonlinear solves at every timestep, then wait hours or days for results that inform multi-million-dollar decisions.

When we first introduced Permeon in May, it was a GPU-powered black-oil simulator. Since then it has grown into something broader: a reservoir simulator with the compute built in, covering the full subsurface and delivered as a cloud app, a REST API and an on-premise license. This post is updated to reflect where Permeon is today.

The problem: compute, not just code

Legacy simulators were designed when CPUs were the only option, and their solvers scale poorly as you add cores. But the bigger constraint for most teams isn't the solver — it's the hardware around it.

Running simulation at scale means buying GPU and CPU servers, housing and powering them, keeping drivers, MPI stacks and licenses patched, and queuing for them when everyone needs them at once. Between projects the hardware sits idle; during a deadline it's never enough.

The result is familiar: fewer scenario runs, simplified models, and uncertainty ranges wider than anyone would like, because running enough realizations for a robust P10/P50/P90 is simply impractical.

Permeon removes both constraints: the slow solver and the hardware you'd otherwise have to own.

A GPU-native engine that also scales with MPI

Permeon's solver was written from first principles for parallel hardware:

GPU-native linear algebra. The heart of every timestep is a large sparse linear system. Permeon solves it on the GPU with a CPR (constrained-pressure-residual) two-stage preconditioner — an AMG pressure stage and an ILU smoother — built on cuSPARSE and custom CUDA kernels.
MPI domain decomposition. The same physics runs on a pluggable CPU, multi-CPU (MPI) and GPU backend, so a field model can be split across many cores and nodes when that's the fastest route.
Robust nonlinear solves. Fully implicit and IMPES formulations, Newton iterations with line-search damping, and adaptive timestepping that takes large steps where the solution is smooth and small ones where it isn't.

The result is simulation up to 100× faster than traditional CPU-based simulators: overnight runs that finish in minutes, and ensembles that no longer need a cluster booked a week ahead.

The full subsurface on one engine

Permeon started with black-oil and now covers the energy-transition subsurface — all reading the same Eclipse-compatible input:

Black-oil. Three-phase oil, water and gas with dissolved gas and vaporized oil, Stone I/II relative permeability, WAG hysteresis, dual-porosity naturally fractured reservoirs, and analytic and numerical aquifers.
Advanced wells. Peaceman completions, multi-segment wells, VFP/THP control, group controls and history-mode producers.
Geothermal and thermal. A full energy equation — from low-enthalpy doublets through high-enthalpy water–steam flash to supercritical conditions — plus steam EOR (SAGD and steamflood) for heavy oil.
CO₂ storage. CO₂–brine thermodynamics for carbon storage and CO₂-plume geothermal, with a fully coupled pressure–temperature solve.
Coupled geomechanics. Poroelastic compaction, surface subsidence, a 3-D stress field, thermal stress from cold reinjection, and stress-dependent fracture conductivity.

Every physics capability is validated against common commercial and research codes and analytic solutions, and ships with a reproducible validation report. See the features and applications for the details.

Compute included: no hardware to own

This is the biggest change since launch. You don't need to buy, house or maintain expensive GPUs and CPUs to run simulations — Permeon does that for you.

Upload your Eclipse-compatible .DATA deck and its include files — no conversion.
Pick a compute tier and submit, in the browser or with one API call.
Follow progress and cost while it runs, then download EGRID/UNRST results and summary vectors.

Runs are billed per second, from $10 an hour on an 8-core tier, and every new account starts with $10 of free credit — no card needed. Larger multi-node CPU and GPU tiers are rolling out, with first access for Professional plans. Monthly plans include a compute allowance, and there are never overage bills: when the allowance and your credits run out, new runs simply wait. Full details are on the pricing page.

When data has to stay in-house, the same engine runs on your own servers or private cloud with an on-premise license.

Studies, not just single runs

Fast runs change how you work, so the Permeon app is built around studies:

Sensitivity and screening — one-at-a-time sensitivities and Plackett–Burman screening of many variables at once.
Uncertainty and proxies — Latin-hypercube uncertainty studies and Box–Behnken designs to build fast proxy models.
Geological variations — vary layers, faults and relative permeability across a study.
AI-drafted studies — describe what you want to explore and let Permeon draft the study set-up for you to review.
History matching — match production data and compare the realizations.
Reports — export, summarize and print a study to share with your team.

Everything in the app is also in the REST API, documented with curl and Python examples, so studies can run from notebooks and pipelines too. Short narrated videos for each task are on the tutorials page.

Try it on your own model

Permeon is available today. Create an account and run your first deck on the free credit, or request a demo and we'll walk through your own Eclipse model with you.

The era of waiting overnight for simulation results — and of owning the hardware to get them — is over.

Introducing Permeon: Reservoir Simulation and Compute for the AI Age