TurkishSieve CPU/GPU prime sieve found errors in Nicely's tables
Found bugs in 30-year-old twin prime data; RTX 5090 hits 1.1T candidates/sec.
The Turkish Sieve Methodology: Deterministic Computation of Twin and Cousin Prime Pairs Using an N/6 Bit Data Structure
1.13 trillion candidates/sec on RTX 5090 with 6x memory savings over classical sieves.
Number theory researchers, computational mathematicians
Prime95 · yafu · primesieve
What we've achieved:Full Spectrum Stats: We didn't just stop at twin and cousin primes. We’ve now published detailed enumeration data for General Primes up to 100 trillion.
High-Scale Verification: These tables provide a bit-perfect reference for researchers, confirming the accuracy of our modular-arithmetic-free N/6 bit methodology.
Correction of Historical Records: Our engine continues to verify and correct legacy datasets, including the well-known discrepancies in Dr. Thomas Nicely’s tables.
Tera-Scale Performance: Measured a peak throughput of 1.13 Trillion candidates/sec on an RTX 5090.
Coming Soon: Version 2.0.0 (The General Prime Update)The current version (V1.0.0) is specialized for twin/cousin detection. We are now in the final optimization stages of Version 2.0.0, which will:Integrated General Prime Engine: Full, high-speed support for general prime detection and enumeration within a single, unified interface.
Standalone Executable: Like V1, the new version will be released as a pre-built standalone .exe, ensuring that researchers can run tera-scale experiments without needing to manage complex CUDA/C++ environments. Hardware-Level Optimizations: Advanced memory management for the upcoming "General Prime" module to maintain our signature efficiency.The updated statistics are live in our repository. I’d love to hear your thoughts on the scalability of N/6 bit structures for general prime enumeration!
Found bugs in 30-year-old twin prime data; RTX 5090 hits 1.1T candidates/sec.
4096-bit hardware Proth prover on $200 FPGA finds new prime with deterministic proof.
Direct2D GPU PDF renderer with CPU fallback, but alpha-stage and Windows-only.
Infinite canvas terminal with minimap—tmux tabs but actually visible.
Linux finally gets offline voice typing; Ctrl-tap + Vulkan GPU support vs cloud-dependent alternatives.
The repo does one practical thing well: quantify the real-world impact of Apple Silicon's unified memory on analytics by running six TPC-H queries plus a GPU-favorable QX and shipping the raw charts and code. It's specific and empirical — you get MLX vs NumPy vs DuckDB numbers and PNGs, not just hand-wavy claims — but it's narrowly scoped to M4 hardware and small-ish scales, so its conclusions are useful for experimentation rather than sweeping generalization.