Support vibe-qc¶
vibe-qc is free and open-source software (MPL 2.0), developed and maintained by one person on personal hardware. Development happens on a MacBook. Test calculations run on a personal gaming PC. The server that hosts vibe-qc.com, the GitLab repository, and the CI/CD pipeline is self-hosted and self-funded.
There is no institutional backing, no grant, no university compute allocation. If vibe-qc is useful to you — or if you think the Cyclic Cluster Model reaching CCSD(T) accuracy on a publicly available open-source code is worth existing — consider supporting the project.
What your sponsorship funds¶
There are two recurring costs that keep development running at all, and a one-off hardware roadmap that defines what kinds of problems vibe-qc can be developed and tested against. Both matter.
Recurring costs (always-on)¶
Claude Max subscription (~$200/month) — the AI coding assistant that is the direct engine of development. Without it, vibe-qc does not get built at the current pace.
Self-hosted server costs — vibe-qc.com, the GitLab repository, and the CI/CD pipeline all run on personal hardware.
📚 Data licences (small annual items)¶
Some of the work needs paid data subscriptions that an academic group would get bundled in their institutional licence — but vibe-qc has none. These are the smallest, most actionable sponsorship items: any single supporter can fund a year outright.
NIST Crystal Data — SRD 3 — single-user annual subscription, $200/year. Curated crystallographic database; the kind of authoritative reference-structure source needed to build the solid-state quantum-chemical reference database vibe-qc’s v2.x Cyclic Cluster Model track is aimed at. With no institutional access, this subscription is the difference between referencing the literature for every benchmark structure by hand and querying a curated database programmatically. A single sponsor can fully fund a year — one of the easiest concrete contributions to make, and a year of access carries the v2.x reference-database build cleanly through the work it enables.
🧮 Compute time on your cluster¶
Direct financial sponsorship is the headline ask, but for academic groups with cluster access there is a second, equally valuable contribution: let me use your cluster.
vibe-qc has no institutional compute allocation. Reference benchmarks against ORCA, PySCF, and CRYSTAL on production- scale ionic crystals — the kind of comparison that makes the release papers credible and that flushes out scaling bugs the dev hardware never reaches — currently run on a personal gaming PC and a single 16-core/128 GB self-hosted box. That ceiling is the actual rate-limiter on a meaningful chunk of the v0.x → v1.0 roadmap, ahead of the hardware fund’s own timeline.
The ask is intentionally simple: a guest user account on your group’s cluster. No formal allocation, no service-grant paperwork, no priority queue — just an account that lets me submit jobs. Even modest, low-priority access to a big-memory node, an Apple Silicon node, or a GPU node unblocks comparison work the dev hardware cannot reach.
Compute providers are credited in release notes and in the acknowledgements of any paper whose results the access enabled — the same convention CRYSTAL, Turbomole, and ORCA papers use for CINECA, JSC, HLRS, etc.
If your group can spare a guest account, please reach out: mpei@vibe-qc.com.
🖥️ Hardware roadmap¶
vibe-qc is currently developed on a personal laptop. To accelerate development and let the project tackle production-scale quantum chemistry problems rather than toy systems, we are raising funds for a dedicated development workstation. This section describes what the hardware enables, what the two funding tiers are, and what your contribution buys in concrete terms.
Why hardware matters for quantum chemistry¶
Quantum chemistry has unusual hardware requirements compared to typical scientific software. Three properties dominate:
Memory capacity. Wavefunction tensors, two-electron integrals, and density matrices grow steeply with system size. A 30–40 atom molecule with a triple-zeta basis set can require 60–100 GB of RAM for correlated methods like CCSD(T). On smaller hardware, development is limited to molecules with a handful of heavy atoms — fine for unit tests, inadequate for benchmarking against published reference data or the real molecules users actually care about.
Memory bandwidth. Most of the wall-clock time in correlated methods (MP2, CCSD, CCSD(T), CASPT2) is spent on tensor contractions. These operations are bandwidth-bound, not compute-bound. Doubling memory bandwidth roughly doubles throughput on the rate-limiting steps.
Unified memory + Metal GPU support. Apple Silicon shares one high-bandwidth memory pool between CPU and GPU, with no PCIe bottleneck. vibe-qc is being designed with first-class support for Apple’s MLX framework and Metal GPU acceleration — currently a gap in the wider ecosystem (most GPU-accelerated QC codes target CUDA only). A capable Apple GPU lets us validate and tune these code paths properly.
Funding tiers¶
Tier 1 — Development workstation, ~$3,499¶
16-inch MacBook Pro, M5 Pro chip, 64 GB unified memory, 2 TB SSD, standard display.
Spec |
Value |
|---|---|
CPU |
18 cores (6 performance + 12 efficiency) |
GPU |
20 cores |
Memory |
64 GB unified |
Memory bandwidth |
307 GB/s |
Storage |
2 TB SSD |
What this tier unlocks:
DFT benchmarks on systems up to roughly 50 heavy atoms with triple-zeta basis sets.
MP2 and CCSD on small-to-medium organic molecules.
Initial MLX and Metal GPU development on the 20-core GPU.
Fast iteration on core SCF, integral evaluation, and tensor contraction code paths.
This is the threshold where development stops being toy-system- bound and becomes useful for testing against real published benchmarks.
Tier 2 — Full development capability, ~$5,599¶
16-inch MacBook Pro, M5 Max chip, 128 GB unified memory, 2 TB SSD, standard display.
Spec |
Value |
|---|---|
CPU |
18 cores (6 performance + 12 efficiency) |
GPU |
40 cores |
Memory |
128 GB unified |
Memory bandwidth |
614 GB/s |
Storage |
2 TB SSD |
What this tier additionally unlocks beyond Tier 1:
CCSD(T) and other post-HF methods on production-scale molecules (60–100 heavy atoms).
Roughly 2× speedup on bandwidth-bound tensor contractions — the rate-limiting step for correlated methods.
Serious GPU development with 40 Metal cores and 614 GB/s bandwidth shared between CPU and GPU.
Headroom for the next four-to-five years of feature work without hitting hardware ceilings.
The Max tier is what allows vibe-qc to be benchmarked against ORCA, PySCF, and CRYSTAL on the kinds of problems users will actually run — not just the small molecules that fit on the Pro tier.
Why a laptop, not a desktop or server?¶
A reasonable question. Three reasons:
Conferences and demos. vibe-qc is presented and demonstrated at scientific meetings. A portable workstation lets the same machine that runs production benchmarks also run live demos.
Apple Silicon is the development target. Since one of the goals is first-class MLX and Metal support, the development machine has to be Apple Silicon. The 16-inch MacBook Pro is the most capable Apple Silicon device available with 128 GB of unified memory.
Power efficiency. The same chip runs cool and quiet on battery and is competitive with desktop workstations on sustained load. There is no separate “travel laptop plus dev desktop” overhead.
Beyond the laptop — the long-term cluster¶
Once a Tier 1 or Tier 2 development workstation is funded, the next hardware milestone is a self-hosted mini-cluster for routine CI benchmarks against reference codes on real ionic crystals (MgO supercells, perovskites, slab models), validated commit-by-commit instead of by hand. This eventually grows into a multi-node MPI testbed as vibe-qc moves toward the Cyclic Cluster Model at correlated levels of theory. Estimated cost ~$15-25k for a credible starting node count — but it makes more sense to fund after the development bottleneck is solved.
How to support¶
Recurring monthly support. Zero fees. Preferred for ongoing contributors.
One-time donations. No GitHub account required.
Every contribution goes toward the hardware fund until the relevant tier is reached. Progress updates are published, and we document which features and benchmarks shipped as a result of each tier being funded. Sponsors at any level are acknowledged in the release notes for the versions their contribution helped fund. Sponsors who choose to be listed publicly appear on the dedicated sponsors page.
For organizations interested in larger sponsorships, commercial support arrangements, or feature-directed funding, please reach out directly: mpei@vibe-qc.com.
Other ways to help¶
Donate compute time on your group’s cluster — see the 🧮 Compute time section above. For academic groups with allocations, this is often more impactful than direct funding.
Star the repository on GitLab.
Report bugs and install failures via GitLab Issues — the v0.4 bug arc showed that non-dev machines surface problems the dev machine never will.
Contribute a tutorial — the format is documented in the tutorial guide.
Cite vibe-qc in publications — see the citation guide for the software citation, the pob-TZVP basis paper to cite when you use those basis sets, and the libint / libxc / spglib references. The repository ships a
CITATION.cffthat GitLab and citation managers parse automatically.Spread the word — if you work in computational chemistry or solid-state physics and vibe-qc covers your use case, telling colleagues is genuinely useful.