A reverse-engineering study traces a global-memory load on an Nvidia RTX 4090 from a CUDA vector-add kernel through registers, caches, address translation, interconnects and GDDR6X DRAM. Each 32-thread warp executes an LDG.E instruction that requests four bytes per lane; for contiguous floats, the hardware coalesces those requests into four 32-byte sectors, or one 128-byte cache line. The RTX 4090’s virtually indexed and tagged L1 cache is four-way set-associative and returns hits in about 15.4 ns, or 40 cycles at a locked 2.6 GHz core clock. An L1 miss undergoes virtual-to-physical translation through a per-SM, 16-entry TLB shared across warps. Measurements found TLB misses cost about 4.4 ns, suggesting a larger, fast on-chip translation cache beyond it. Physical requests cross to one of 36 L2 slices, each a 2 MiB, 16-way set-associative cache; L2 hits take about 127 ns. On an L2 miss, one of 12 memory controllers accesses its associated 2 GiB GDDR6X chip. A contiguous 128-byte request requires a DRAM row activation and four column reads. The measured DRAM round trip, including cache and interconnect travel, is about 255 ns or 660 cycles. Other resident warps issue work while a warp waits on the load dependency, helping hide that latency.

Kino is an open-source Rack 3 web server for Ruby 4.0+ that uses a Rust front end built on Tokio and Hyper to manage networking and dispatch requests to Ruby workers. In Ractor mode, it runs Ractor-shareable Rack applications across CPU cores in a single process; in threaded mode, it supports any Rack application, including Rails, but Ruby’s Global VM Lock serializes CPU-bound Ruby execution. The server includes graceful draining, worker crash respawning, bounded queues that return 503 responses under pressure, request timeouts, TLS through rustls, access and application logging, and monitoring endpoints. Benchmarks on an AWS c7a.2xlarge server with eight AMD EPYC cores, Ruby 4.0.5, and YJIT showed Kino Ractor mode serving 77,999 CPU-bound fib requests per second, versus 58,006 for an eight-worker Puma cluster. The Ractor benchmark application used 148 MB proportional set size, compared with 1,068 MB for Puma’s cluster. Kino’s threaded Rails hello-world configuration used 92 MB versus Puma’s 389 MB, but handled 2,637 requests per second compared with Puma’s 12,138 because Puma used eight processes across all cores. Ractors remain officially experimental in Ruby 4.0, and Rails is not currently Ractor-shareable. Kino’s kino --check command identifies objects and captured state that prevent an app from using Ractor mode.

Octane is a React programming model that eliminates the rules of hooks, allowing hooks to be used conditionally and without dependency arrays. It compiles React's hooks, Suspense, and actions ahead of time, focusing on performance without utilizing a virtual DOM.

Data-oriented design focuses on optimizing data structures and memory access patterns to improve performance in software development. This methodology emphasizes the organization of data to enhance cache efficiency and processing speed, particularly in game development.

ast-grep rewrote Tree-sitter's C core in Rust, resulting in a 30% increase in parser speed and a 22% improvement in overall performance for ast-grep. The new core enhances parsing efficiency and accelerates tree reading.

Gigatoken enables language model tokenization at speeds of GB/s, making it approximately 1000 times faster than HuggingFace's tokenizers. It serves as a drop-in replacement, supports a wide range of CPU hardware, and is compatible with multiple commonly used tokenizers.

Elm is a programming language designed for building web applications, featuring friendly error messages, high performance, small asset sizes, and no runtime exceptions. Recent updates have improved build times, enhancing developer productivity.

Prylint is a Rust reimplementation of pylint that delivers byte-for-byte identical output while being 15 to 2300 times faster, with a median speed increase of approximately 85 times. It maintains the same error messages, line and column references, text order, exit codes, and footer as the original pylint.

Zef is a dynamic language with an extremely simple AST-walking interpreter that has been optimized for speed. The optimizations aim to make Zef competitive with established interpreters like Lua, QuickJS, and CPython.

Bun has achieved a 100x speed increase by rewriting Git in the Zig programming language and integrating it into its runtime. This enhancement significantly improves performance for developers using Bun.