gunnchOS
Research · STAGING_WORKER
Educational and research baseband platform. The GPU and native radio code do not run on this Worker.
Evidence class: SYNTHETIC_SIM. The selector lists CPU synthetic benchmark rows. The GPU gate stays blocked.
# gunnchos GPU NR Baseband Platform CPU-first, Aerial-aligned NR PUSCH-oriented baseband vertical slice with explicit separation of **educational substitute** vs **standards-path** vs **CPU validation** vs **CUDA candidate** vs **GPU measured** vs **3GPP conformance**. **Not a dissertation paper. Not 3GPP conformance. Not carrier-grade.** Supervisor path: [`docs/START_HERE_SUPERVISOR.md`](docs/START_HERE_SUPERVISOR.m
- Apple M2 / CPU-only: build & test the CPU standards path. - Missing NVIDIA GPU → `BLOCKED_GPU` / `BLOCKED_HARDWARE` / `BLOCKED_GPU_RUNNER` JSON (never invent timings). - CompactQcLdpc + (16,8) LDPC live under `educational/` — **not** NR acceptance. - Real BG1/BG2 tables: `third_party/sionna_ldpc_codes/` (Apache-2.0) → `ldpc_bg_tables.hpp`.
make bootstrap test make validate-phy-independent make cuda-correctness cuda-equivalence # BLOCKED_HARDWARE on Mac make gate4-cpu-reference optimization-study-cpu orchestrator gate6-dry-run
Methods are the CPU benchmark JSON files in results/benchmarks and the blocked gate in results/blocked_gpu/BLOCKED_GPU.json. CPU rows with status PASS copy median_us from those files. The blocked-GPU file says gpu_timings_present is false and numeric_gpu_claim is false. No GPU kernel timing is added here.
Native research runtimes stay on a local checkout. This page shows repository records only.
results/benchmarks/cpu_crc24a.jsonresults/benchmarks/cpu_eq_mmse_4x4.jsonresults/benchmarks/cpu_ldpc_nr_bg2.jsonresults/benchmarks/cpu_ldpc_nr_scaffold.jsonresults/benchmarks/cpu_ldpc_short.jsonresults/benchmarks/cpu_memory_layout.jsonresults/benchmarks/cpu_ofdm_64.jsonresults/benchmarks/cpu_ofdm_64_fft.jsonresults/benchmarks/cpu_scheduler_mac.jsonresults/benchmarks/cpu_scheduler_pf.jsonresults/benchmarks/cuda_graphs_streams.jsonresults/benchmarks/gate4_cuda_build.jsonPinned source 3931f51d43b7bee87db6d710a9df5a7c1136fcfa
Educational and research baseband platform. The GPU and native radio code do not run on this Worker.
The research code is not executing here. This is a research surface.
It gives this repository a public web surface inside the gunnchOS ecosystem without pretending the original runtime runs on Cloudflare Workers.
Use the primary action to inspect repository records, then open the source for the local toolchain.
STAGING_WORKER
Runtime mode on this page: PRECOMPUTED. No remote model call is made.