Computational Engine

Every result below is computed in your browser by the QualiaDB engine compiled to WebAssembly — the same deterministic solver code that serves as the CPU reference oracle for the native engine and its GPU accelerators. No server, no external API call required.

v0.0.41 Architecture: CPU Reference Solvers & Etalon GPU Certification (ETA-14, ETA-15, ETA-17)

For a quick overview: QualiaDB separates semantic authority and exact reference math (`qualia-core-db`) from reusable hardware acceleration (`etalon-*`). The interactive browser cards below run the deterministic CPU reference solvers directly in WebAssembly. On native desktop and WebGPU hardware lanes, etalon-wgsl-forge (bridged via qualia-core-db::wgsl_forge::etalon_bridge) compiles typed KernelSpec + Schedule intermediate representations into target shaders (WGSL, SPIR-V, HLSL, and MSL), validates every module through the Naga AST verifier, and certifies GPU dispatches against these CPU reference oracles:

See the WGSL Forge Architecture and v0.0.41 Release Overview for full hardware qualification details.

What runs here (CPU-exact, in-browser)

Dense linear algebra, statistics & hypothesis tests, symbolic algebra, special functions & number theory, arbitrary-precision exact arithmetic, SI units & dimensional analysis, Fourier / Laplace / Z transforms, graph reasoning, and cryptographic hashing (SHA-2 / SHA-3 / BLAKE3 / HKDF) — wrapping crate::solvers::* and the RustCrypto primitives, the wasm-clean layers. In the browser these execute the exact CPU reference floor (`ExactnessProfile::Exact` / `DeterministicApproximate`).

What runs on the 0.0.41 Etalon GPU & Native Engine

The 9 specialized domain libraries (finance, medical, chemistry, engineering, full ML), the advanced CAS (symbolic integration / limits / series / ODE), the pure-Rust etalon accelerator stack (etalon-wgsl-forge, etalon-sim, etalon-signal, etalon-tensor, etalon-nquin, etalon-agent), and key-generating cryptography run on the native / MCP engine (see the 0.0.41 Release Preview). Several already have live demos on the Sciences pages.

LIVE VIBESCRIPT COMPUTE & ROOFLINE PLANNER · vibe:0.3/{gpu,p64,infer,forge}

VibeScript Compute, Roofline & P64 Quantization Planner

Interactive WebAssembly planner executing eval_program_with_receipt_wasm to size KV-caches, P64 64-byte aligned quantized tensors, workgroup occupancy, and hardware Roofline bounds against the 42 MB Prolog Sentinel arena.

Probing WebGPU / WebGL2… astHash: initializing…
8.0 B
0.5B7B/8B32B70B
8192 tok
51240961638432768
KV-Cache Footprint (`infer.kv.kv_cache_bytes`)
—
P64 Quantized Row (`p64.runtime.quant_row_bytes`)
—
Workgroup Occupancy (`gpu.compute.workgroup_occupancy`)
—
Arithmetic Intensity (`estimate_arithmetic_intensity`)
—
Touch / hover bars or Roofline curve to inspect quantization tier metrics 42 MB Prolog Sentinel Budget Enforced
Editable VibeScript Compute Planner (vibe-wasm)
Evaluated Planner Output & CBOR-LD Tag 4200 Receipt
Evaluating in vibe-wasm…
LIVE IN-BROWSER BENCHMARK SUITE · CPU ORACLE + VIBESCRIPT KERNELS

Interactive Throughput & Speedup Benchmark Suite

Kernel / Solver Latency Throughput Speedup
Initializing benchmark suite…

QualiaDB v0.0.41 • Computational Engine • bundle docs/playground/qualia_core_db_bg.wasm (--features wasm-full) • exports: crates/qualia-core-db/src/wasm_bridge/engine/