Choosing a PIC code in 2026: WarpX, EPOCH, Smilei, PIConGPU and friends
A practical comparison of open-source particle-in-cell codes for laser–plasma, astrophysical and low-temperature plasma research: strengths, hardware and learning curve.
There is no single “best” PIC code. The right choice depends on your physics, your hardware and your team. Here is how we think about the main open-source options when advising clients.
The short version
| Code | Best for | Hardware | Learning curve |
|---|---|---|---|
| WarpX | Large 3D laser–plasma and accelerator runs, mesh refinement, boosted frame | GPU-first (NVIDIA, AMD, Intel), CPU | Moderate–steep |
| EPOCH | Robust laser–plasma and general EM PIC, QED physics | CPU (MPI) | Gentle |
| Smilei | Laser–plasma, astrophysics, load balancing, many physics modules | CPU, growing GPU support | Moderate |
| PIConGPU | Very large GPU runs, synthetic radiation diagnostics | GPU | Steep |
| OSIRIS | Laser–plasma, astrophysics, very mature | CPU/GPU (licence required) | Moderate |
| VPIC | Huge kinetic runs, magnetic reconnection, space plasma | CPU, GPU (VPIC 2.0) | Steep |
Questions to ask first
- Which physics do you need? Ionisation, collisions, QED, radiation reaction and moving windows are not equally mature in every code.
- What hardware will you run on? If your allocation is on GPUs, a CPU-only code wastes most of it. Most new national supercomputers are GPU-heavy.
- How big are your runs? Small 1D/2D studies run happily on a workstation with EPOCH or Smilei. 3D production campaigns benefit enormously from GPU codes.
- Who will maintain the workflow? A code with an active community, good documentation and Python tooling saves months over a PhD.
Low-temperature and discharge plasmas
Electrostatic, collision-dominated problems (discharges, sheaths, thrusters) usually need PIC-MCC with careful collision models and often implicit or energy-conserving schemes. Several of the codes above support collisions, but dedicated PIC-MCC codes are often more efficient. Verification against benchmark cases from the literature is essential here.
Our advice
Start with the physics, then pick the code. Before investing a PhD-year, run a small benchmark of your actual problem in two candidate codes. You will learn more in a week than from any comparison table, including this one.
Not sure which code fits your project? Use the free PIC Simulation Planner for a first estimate, or ask us. Code selection is often the first step of our engagements.
Enjoyed this? Get Radial Notes monthly.