Pauling runs on Google Cloud Dataflow -- auto-scaling, fault-tolerant, zero infrastructure to manage. You pay for compute, not for hardware collecting dust. Running your own HPC cluster means SLURM scripts, failed jobs at 3 AM, a full-time sysadmin just to keep nodes alive, and $500K+ upfront before the first simulation runs.
| Category | Pauling | HPC Cluster |
|---|---|---|
| Setup time | Minutes (sign up and run) | Months (procurement, racking, networking, OS, SLURM config) |
| Upfront hardware cost | $0 | $500K--$2M+ for a modest GPU cluster |
| Ongoing maintenance | Managed by Pauling | 0.5--1 FTE sysadmin minimum |
| Scaling | Auto-scales to hundreds of workers via Cloud Dataflow | Fixed capacity; scaling means new purchase orders |
| Fault tolerance | Automatic retry and checkpointing (MD save/resume) | Manual requeue; failed jobs need babysitting |
| Job scheduling | Instant -- no queue wait behind other groups | SLURM queue contention; priority fights across labs |
| GPU access | On-demand GPU nodes for UniDock, GROMACS | GPUs idle or oversubscribed depending on the week |
| Software stack management | Always current; tools pre-integrated | You maintain every package, dependency, and MPI build |
| Cost model | Pay per job -- compute only | CapEx + electricity + cooling + maintenance + depreciation |
| Multi-site access | Browser-based, accessible anywhere | VPN + SSH + forwarding configs per user |
| File format handling | Automatic conversion (PDB, SDF, MOL2, SMILES, CIF, PDBQT) | You write the conversion scripts |
| Reproducibility | Every job tracked with full parameter logging | Depends on how disciplined your users are with SLURM scripts |