#性能优化
200
2026-10-01
deepseek-ai/DeepGEMM-Ascend
S
DeepGEMM-Ascend: clean and efficient matrix multiplication kernel library for Huawei Ascend NPUs
2026-09-11
deepseek-ai/DeepSelect
S
DeepSelect: TopK kernels for DeepSeek Sparse Attention (DSA) and Samplers
2026-10-20
Show HN: KiviDB, a faster Redis-compatible memory database and vector cache
A
Hi all, So I have been working on this project KiviDB from February 2026. It started as a project for me to understand rust, having a product with no GC pauses, memory blow up after snapshots and getting OOM on running pods, and that was actually an issue I faced with Dragonfly. I wanted to have something picking the best of Rust, but also picking the best parts of other alternatives like Garnet and Dragonfly. I used AI a lot, but I was the one making the design decisions based on the tests that claude picked from my history of troubleshooting sessions as part of my job. KiviDB supports both RESP2 and RESP3 protocol. I had some founders credits on my AWS account which helped me with the infra, and running instances I could perform benchmark on. Its keyspace is sharded across threads, it is multi-threaded in nature, and performs best on Linux due to io_uring, I used the tokio_uring crate for this. It also has tokio fallback for other OS's. I hadn't open sourced it until recently as I felt it was not ready yet, the best way to learn it was to build a cloud managed service for it across all three widely used cloud providers i.e AWS, GCP and azure. Since I used to mainly work with cloud, I had some insights on what would be the best ways for the deployments, how to conduct failovers, manage SLAs, etc. And that's what I took help from claude to fast pace it. The website does look like AI slop but it works for now, given my limited resources, that's all I had. You may check it out at : https://kividb.io, I open sourced the core engine under AGPLv3, didn't want to have just source available, because that is available with other tools too. It supports apprx 98-99% of the redis 8.10 commands, few have issues and you can find them at https://github.com/kividbio/kividb/issues, contributors are welcome. Now the numbers, for the final benchmark, I used 2 c7gn.metal servers, in the same placement group in us-east1, and I used memtier-benchmark tool for this. The setting was : 10M keys, 100B each, 32 cores in use with both pipeline and un-pipelined runs. pipeline 16 GET: KiviDB 21.2M ops/sec vs Dragonfly 2.0 9.1M, Redis 8.10 2.4M, Redis Cluster 32 shards 4.7M When not pipelined, Dragonfly performed slightly better at 3.35M ops/sec vs kividb at 3.07M ops/sec. Memory per key was also slightly worse than dragonfly and almost same as Redis, 177B (kividb) vs 134B (dragonfly) vs 178B (redis). With vector ingestion times it was slower but when querying, it was faster than most alternatives at lower ef, however slower than dragonfly at high ef. Most of the technical details are mentioned here so you can check it out: https://kividb.io/benchmarks. easiest way to try it: docker run -d -p 6380:6380 kividb/kividb I am looking forward to building a community around this, provide better rust based tools, I am open to constructive criticism and value informed opinion. Thanks
2026-10-16
leekt/skim
S
Application-specific Ethereum state client: sync (and archive) only the state your app needs, straight from P2P. BAL-driven, verified, local eth_call in ms.
2026-10-14
datalab-to/indra
S
Fast CPU OCR: line detection, CTC recognition, reading order, markdown
2026-09-15
Show HN: Nari Qwen3-TTS and Qwen3-ASR – High accuracy, low latency and cost
A
Hey HN, Toby from Nari Labs here. We've been working on making OSS speech models super-fast. Last year, we built Dia, the first OSS text-to-speech model capable of doing natural dialogue. Since then, so many more great speech models have been released to the public. But the market is still dominated by closed source models. We think that's an inference problem. Existing systems such as vLLM / SGLang are not well suited for multimodal inference. To prove this, we built an inference engine specialized for Qwen3-TTS and open-sourced it ( https://github.com/nari-labs/nari-qwen3-tts). Running at sub-50 ms latency at 10 RPS, this showed open models can be run much faster and cheaper. Since then, we've been working hard to bring cheap, fast, and high quality serving to all. And we've even beat closed models at their game! Measured on the highly cited Coval (YC S24) voice AI benchmarks, our Qwen3-TTS endpoint not just is #2 in latency, but #1 in accuracy (WER) compared to 11Labs, Cartesia etc. while being the cheapest endpoint. Our Qwen3-ASR endpoint has the lowest latency and #2 accuracy, just 0.1% away from #1. It is the second cheapest model on the list. It took a lot of clever inference engineering to make these models quick, perform well while keeping costs low. Interestingly, Alibaba's official endpoints seem to perform worse in terms of accuracy and latency compared to ours. But nonetheless, much love to the Qwen team for OSS-ing these amazing speech models. We want to continue to push prices down to make speech technology a commodity - so that every app can have great TTS and STT without worrying about unit costs. We're also working on other parts of audio such as diarization - as well as video and world model inference. More to come!
2026-09-04
sohu-mptc/FlashRec
A
FlashRec is a CUDA-graph engine for generative recommendation: wide beam search (3–5 SID steps, n=50–512+) over a trie-constrained catalog, in-process FP8 serving, and ranked beams on /v1/chat/completions.
2026-10-10
sqliteai/sqlite-multiwriter
S
Concurrent SQLite writes across threads and processes without modifying SQLite. No forks, no schema changes, up to 5.7x faster writes.
2026-09-02
Show HN: Running 104GB Qwen3.8-Flash-Next on 48GB Mac with at ~12 tok/s
A
I built slotstream, a way to run Qwen3.8-Flash-Next 4-bit on a low-memory mac starting from 16GB, a 125B parameter model that would need 100GB+ memory/RAM, thanks to expert-offloading/ssd-streaming. Easy to install/update, and mac-native using MLX and Swift. It ships with auto-mode, which makes a good tradeoff between memory usage and speed. I'll be implementing and porting the MTP module for speculative decoding next
2026-10-01
deepseek-ai/DeepEP-Ascend
A
A high-performance communication library for machine learning training and inference on Huawei Ascend NPUs.
2026-09-23
Show HN: InstinctFlash – High-Performance Serving Runtime for Robotics Models
A
Hey HN, Guanming here, cofounder of General Instinct. We just released InstinctFlash, a high-performance serving framework for robotics models. It’s licensed under AGPL-3.0. On Jetson Thor, we see speedups about 1.2x to 7.9x from runtime optimizations alone and up to 33.78x for LingBot-VA when we combine those runtime optimizations with a distilled few-step diffusion scheduler, going from the original 25 visual / 50 action steps to 2 / 4 steps. Across 50 Robotwin2.0 tasks, we evaluated 1,153 episodes per configuration, LingBot-VA with InstinctFlash at 2 visual / 4 action steps achieved a 90.5% success rate, compared with 92.1% for the baseline at 25 visual / 50 action steps. Here’s an optimized 5B world action model, running in real time on a Jetson Thor: https://youtu.be/nku65iyL5Fw InstinctFlash currently supports 8 VLA / world-action model families, including pi0.5 and NVIDIA Cosmos Policy, across RTX 4090 / 5090 and Jetson Thor. Just give it your fine-tuned checkpoint and InstinctFlash handles the rest, exposing the accelerated model through a Python runtime or an OpenPI-compatible WebSocket server. We started working on this because we kept running into the same problem while deploying robot policies, the models were getting much better, but inference was often way too slow for the control loop we actually wanted. For pi0.5, mixed-precision GEMMs and CUDA graphs speed up computation and reduce launch overhead. For Cosmos, caching avoids redundant computation across diffusion steps. World-action models’ diffusion denoising step depends on the previous one which motivated our work on few-step distillation. Right now, InstinctFlash contains 6 aspects of optimization. - Graph: CUDA graph capture, memory planning and separating prefill from repeated execution. - Cache: Reusing KV and conditioning state across diffusion steps and prediction calls. - Attention: Specialized attention paths for different model architectures. - Kernels: Fused operations and kernels tailored to specific backends and tensor layouts. - Precision: FP8 and mixed-precision execution. - Model: Few-step distillation for diffusion and action generation. Teams at Samsung, Siemens, and other robotics startups have used InstinctFlash for model acceleration on VLAs, WAMs, and diffusion-based world models. Now we are opening up access to you. Feel free to try it here: https://github.com/General-Instinct/InstinctFlash More implementation details and benchmarks: https://general-instinct.com/blog/instinctflash-edge-inferen... Would love to hear your feedback!
2026-09-29
yfyeung/PrunedCTC
S
Memory-efficient CTC loss with exact loss and first-order gradient equivalence under vocabulary reduction, and activation memory that no longer scales linearly with vocabulary size.
2026-08-27
0xBakeer/qwen38-flash-next-spark
A
Run Qwen3.8-Flash-Next (180B) on a single DGX Spark by keeping its 51B n-gram embedding table on NVMe
2026-10-29
zizco/pcaplat
A
One-way latency, tail statistics and microburst analysis from two packet captures
2026-10-21
indexable-inc/ra-jit
A
Rust compiler daemon on rust-analyzer: fast incremental cargo builds and live edits to running programs. Try: nix run github:indexable-inc/ra-jit
2026-11-08
Th0rgal/satmux
A
Accelerate high-latency GEO inflight Wi-Fi via warm BBR TCP multiplexing + 1-RTT Ground-TLS + zlib (zero dependencies)
2026-10-25
alleyxd/quantpicker
A
Measures which GGUF quant of a model to run on your GPU: real speed and quality loss (KL divergence), not estimates.
2026-10-23
xevrion/hatchzip
A
Extract huge archives in place: the archive shrinks as your files hatch, so you only need room for the output. ZIP, 7z, tar, split archives.
2026-09-23
mattxslv/childstream
A
Stream games from a second Windows desktop while you keep using your PC - via Windows child sessions + Sunshine
2026-09-07
xzf-thu/BreezeTTS2_Mac_Streaming
A
Real-time Breeze-TTS-2 streaming on Apple Silicon — RTF from ~4 down to ~1. mlx-audio 0.5.1's depth decoder recomputes each frame's existing acoustic codes instead of reusing its KV cache. Adds intra-frame KV reuse, one CPU read-back per frame, an optional compiled whole-frame path, and RTF/underflow stats.
2026-10-22
ahtx/kindi
A
A custom LLM inference engine for the NVIDIA Quadro K420: Ternary-Bonsai-2-2B at 20+ tokens/s with an OpenAI-compatible server
2026-11-09
ArneshBanerjee/gpt2.s
A
GPT-2 small (124M) in hand-written AArch64 assembly. No libc calls. About 100 tokens/s on an M1.
2026-10-15
R-Hatim/scry
A
Whole-disk file search for Windows: any file by name in a millisecond, search inside your files, a search bar on Alt+Shift+Space and a tray icon.
2026-09-21
mizorewww/laya-mlx
B
Native MLX runtime for Laya typed decision models — 7–14 ms short decisions on M3 Max. No text generation, PyTorch, or cloud API.
2026-09-19
CYBERVERSE-Research/skyline-speeder
A
Sender-side TCP acceleration: eBPF struct_ops congestion control with a Rust control plane. For links with non-congestive random loss. Kernel 6.12 LTS+.
2026-09-16
anmolkapil/plexo
B
Speed up downloads by combining multiple network connections in parallel
2026-09-11
kevinzakka/mjbatch
B
A Python library for running thousands of MuJoCo simulations in parallel on CPU
2026-09-08
VII-Cae/hyalite--liquid-glass
B
Real refraction liquid glass for the web: SDF lens maps + SVG displacement + backdrop-filter. One file, no WebGL.
2026-10-09
noahdunnagan/fsearch
B
Whole-disk file search for macOS: fuzzy names, typo tolerance, indexed content grep. ~1 ms over 8M files.
2026-09-30
Spottt/elm-compiler
A
Fast, unofficial Elm 0.19.1 compiler written in Rust — drop-in replacement for elm (npm: @spottt/elm-compiler)
2026-09-18
Noisemux/zeptun
A
High-performance tun2socks engine in Zig with TCP, UDP, ICMP, SOCKS5, io_uring, and C API.
2026-08-24
harlixay7/ReclaimArc
A
Transactional in-place archive extractor for Windows that progressively reclaims disk space from archives as it extracts. Extract 100 GB files with only 5-10 GB of free space.
2026-11-09
nico-bohlinger/xmimic
A
Whole-body motion tracking for the Unitree G1 in under a minute. BlinkMimic's world-model trainer rebuilt in JAX, 3.5x faster on four H100s.
2026-09-28
linuxkid473/ppcosxkvm
A
PowerPC Mac OS X Tiger on Apple Silicon with an emulated ATI Radeon 9700 PRO: Quartz Extreme, Core Image and OpenGL via Metal
2026-09-28
orlp/polyxor
A
A 128-bit universal hash function designed for fast file/data stream integrity checks with cryptographic strength, written in Rust.
2026-09-26
agentscope-ai-java/json-guard
B
JSON.parse with hard limits on nesting depth, key count, string length and input size. Refuses parser bombs before they are allocated.
2026-09-16
slorber/gpu-lexer-textarea
A
Demo of using gpu-lexer with the new OpaqueRange web API (Chrome 152+) to create a universal syntax highlighter within an editable textarea
2026-09-07
pcm469768-droid/llm-api-speedtest
A
Benchmark LLM APIs, OpenAI-compatible gateways, and API proxies by token speed, TTFB, latency, and success rate. Compare multiple providers with reproducible JSON reports and scheduled GitHub Actions.
2026-08-30
jlrouzies-fr/DLSS5-Feeder
B
DLSS 5 neural rendering in D3D11/D12/Vulkan games that ship without any DLSS — feeds a synthetic DLAA contract (ReShade depth + motion vectors) to the DLSS 5 add-on via a private D3D12 device.
2026-11-03
CubeLink-oss/voltdir
A
Read-only Windows disk analysis for AI agents: one call, stable JSON, low memory.
2026-10-28
deepanshu406/scx_pgtier
A
Two-tier, selectively unfair sched_ext (eBPF) CPU scheduler for PostgreSQL servers that run short transactions next to long queries. Open-source implementation of the UFS idea from arXiv:2605.02377.
2026-10-16
remorses/mdastic
A
Markdown and MDX parser in TypeScript. Exact remark mdast output, 17 to 28x faster.
2026-10-05
MohammadHaishemKhawaja/hashyy
A
A full, unpruned 177B MoE model at 11.5 tok/s on one 12 GB GPU. Expert streaming for llama.cpp, with the measurement harness and every dead end.
2026-09-27
NullByte3/midden
A
Fast JVM heap dump analyser: retained sizes, owners and leak suspects from .hprof files. One static binary, no JDK needed.
2026-09-27
crmne/fastframe
B
egui on rails: the shared foundation for native Rust apps built with egui
2026-09-25
demirfirat/ipadScreen
A
Use an old jailbroken iPad (iOS 6+, tested on iPad 2) as a real extended display for your Mac, over USB or Wi-Fi. Native iOS 6 app with libjpeg-turbo NEON, ~35 fps.
2026-09-23
NoizAI/YuE2-Turbo
A
Fast, concurrent inference for YuE2. Same model and recipe: 1.68× faster per song, 3.31× more songs per GPU.
2026-09-15
magna-nz/redis-near-cache
A
Client-side caching on top of StackExchange.Redis. Keeps the values you read in memory and drops them the moment Redis says they changed.
2026-09-13
matthart1983/kernwatch
A
Linux kernel observability in your terminal. Built with Rust and Ratatui.
2026-09-10
MapleTechLabs/effect-clickhouse
A
Type-safe ClickHouse queries, result decoding, and reproducible benchmarks for Effect and TypeScript
2026-09-10
Show HN: Compute polynomials twice as fast
B
A few years ago my coauthor and I was wondering if we could reduce the number of multiplications used for hashing algorithms. We had a construction and a 100 page proof, but we were not 100% sure it was correct. Now we have a full Lean proof, so we decided to publish it. I made this website to make it easy for anyone how has polynomials to evaluate to see how it would be done using our method, as well as a number of previous approaches by Knuth and others.
2026-09-08
Deep-Learning-Profiling-Tools/ptx-optimization
B
Evidence-led NVIDIA PTX optimization skill: profiling, exact-cubin patching, validation, and measured tuning
2026-08-27
lxw112190/lw.PPOCR.C
B
轻量级纯 C PP-OCRv6 Tiny 推理运行时，支持 DET/CLS/REC、完整 OCR、SSE2/AVX2，并提供 C# WinForms 与跨平台 HTTP/Web Demo。Lightweight pure-C PP-OCRv6 Tiny runtime with full OCR, SIMD, C# WinForms and cross-platform HTTP/Web demos.
2026-10-28
juanjsebgarcia/vita-turbo-storage
A
Faster SD2Vita and internal storage for the PS Vita (taiHEN plugin, firmware 3.60-3.74)
2026-10-27
heydeer/pngdeer
A
Smaller PNGs, same picture. Lossy PNG compression that picks the quality for you: 8–12% smaller than TinyPNG at the same quality. Command line and in the browser.
2026-10-26
alrojo/fp4Attn
A
CroissAttn: FP4 attention forward and backward, with analysis of stochastic rounding, zero-sum softmax gradients, and sampled centering.
2026-10-17
yapfast/yapfast
A
The chat app that doesn't eat your PC. A lightweight, native Discord client in Rust.
2026-10-10
Zagorodnyi/filter-ast
A
Parses a declarative filter object with $op(value) strings into a typed, database-agnostic filter AST. Zero dependencies, ~50 ns per call.
2026-10-04
Meinianda-L/lowpaper-live-wallpaper
A
Ultra-lightweight live / animated video wallpaper for Linux (X11 & Wayland), designed for low-end and old hardware. Minimal Bash script around mpv.
2026-10-01
linuxkid473/maclator
A
Run arm64 macOS apps on Intel Macs: a reverse-Rosetta emulator (interpreter + x86-64 JIT + AOT cache) with a Metal bridge to the host GPU
2026-09-28
roman-starcevic/RUG
A
RUG — Rug Understands Grammar — is a C++98 parser generator and runtime library that compiles an ABNF grammar into a packed LALR(1) parsing table.
2026-09-07
Jarvis322/macos-sysdata
B
See what is really inside macOS System Data and delete it, item by item, from the menu bar.
2026-11-02
Hona/strobe
A
Measure and explain performance: web pages, Electron apps, Bun/Node, any process. Agent-native CLI + MCP.
2026-10-16
PaddlePaddle/FlashMask
A
Efficient Mask-Aware Distributed Attention for Long-Context Training
2026-10-14
rohit267/net-merge
A
Use all your network adapters at once on Windows: per-connection load balancing over Wi-Fi, Ethernet and phone/USB tethers (TUN + Wintun)
2026-10-05
Meinianda-L/lowpaper
A
Ultra-lightweight live / animated video wallpaper for Linux (X11 & Wayland), designed for low-end and old hardware. Minimal Bash script around mpv.
2026-09-29
B0yko/taskdistill
A
Distil an expensive LLM API call on a narrow task into a small local model: capture traffic, curate, LoRA fine-tune with MLX on Apple Silicon, evaluate against the teacher, and serve an OpenAI-compatible cascade that escalates low-confidence requests.
2026-09-14
tailscale/lokv
A
lokv implements an append-only log over a sorted, create-only key/value store
2026-09-05
truespar/paddock
B
Native Rust inference server for open models on NVIDIA GPUs. OpenAI- and Anthropic-compatible APIs, GGUF + safetensors, FP8/NVFP4/MXFP4/Q8/Q4, built-in Studio
2026-09-01
faisalkindi/DLSS5oneclick
B
One-click setup of the leaked DLSS 5 neural-rendering build for any DX11/DX12 game on RTX 20–50, with or without DLSS. ReShade + RenoDX add-on (or OptiScaler engine); DLSS5-Feeder + LumeniteFX for games without DLSS; dlss5-bridge for DX11. Rust, single exe.
2026-10-29
Humpalumps/Ultra-Water
A
Real-time FFT ocean renderer for three.js (WebGL2): shore breakers, foam, caustics, underwater rendering and buoyancy at 200+ fps
2026-10-27
LukaszRozmej/BBTree
A
Fast ordered maps for .NET: a C# port of bbtree, a B+-tree with SIMD inline-key search
2026-10-23
linyusenzz/UltraSlim
A
Android 离线照片转码工具：JPEG / HEIC / AVIF / JXL 互转，保留 Ultra HDR / ISO 21496-1 增益图，支持批量与后台转换
2026-10-08
selanavot/matrix-multiplication-all-fields
A
Lean formalization work extending the matrix multiplication 9/4 argument to arbitrary fields
2026-10-04
Yamz-Labs/kyojin
A
Kyojin: the Yamz inference engine for AMD Strix Halo (ROCm, gfx1151), built on ExLlamaV3. Runs 300B-class MoE models on one 128 GB mini PC.
2026-10-04
christopherowen/dgx-spark-memory-saver
A
Recover usable memory on DGX Spark with a 64 KiB kernel: an opt-in NVIDIA UVM page-table packing patch, build tools, and measured validation.
2026-09-26
bertaye/agentic-cuda-optimizer
B
A langgraph based workflow with a C++ CUDA harness to optimize CUDA kernels
2026-09-26
Show HN: Agentic CUDA Kernel Optimizer
B
Hello; I was working on optimizing some CUDA kernels and I thought may be it is a good oppurtunity learn langgraph as well. I created a simple C++ CUDA Test Harness and handed that to AI agents. They can run kernels, get benchmarks, and even can profile via nsight
2026-09-18
bnsd55/jevmlx
A
Jev-style parallel constrained decisions for any MLX model on Apple Silicon. Typed, schema-valid JSON in one forward pass.
2026-09-18
fabeloper/sf4enhanced
A
Rollback netcode for Ultra Street Fighter IV (Steam). Downloads only.
2026-09-09
coccofresco/TriAevum
B
Experimental native PC recompilation runtime for Ocarina of Time 3D. Windows x64, NRI/Vulkan, single-screen UI and ROM-only setup.
2026-08-31
orloxgr/AdaptiveDisk
A
Copy files from damaged disks FAST — skip bad sectors, map readable areas, and recover as much data as possible without wasting hours on unreadable regions.
2026-10-23
yftx888/Pulse
B
轻量的 Windows 桌面悬浮监控：电池功耗与续航、CPU/GPU/内存占用与功耗、网速、真实延迟与断网提醒。原生 C++ / Direct2D，迷你模式约 0.1% CPU。
2026-10-21
mr-finster/vf3-recomp
B
Virtua Fighter 3 (Sega Model 3, 1996) for Mac, Windows and Steam Deck: the arcade's PowerPC program recompiled to native code, up to 6x, widescreen, smooth 120 fps. Bring your own vf3.zip.
2026-10-11
AFAlimru/fastly-hosts-adjust
B
Find the most stable nearby Fastly CDN route and pin it in hosts to speed up GitHub downloads and X images. Windows & Linux, menu-driven, English/中文. 自動找出附近最順的 Fastly 路線並寫入 hosts，加速 GitHub 下載與 X 圖片。支援 Windows / Linux，中英文選單操作。
2026-09-29
kofiadeyemiq/iterleak
B
Go analyzer that reports iter.Pull and iter.Pull2 calls whose stop function can go uncalled and leak a goroutine.
2026-09-26
architectds/collabosm
B
Run Qwen3.8-Flash-Next (125B-A6B MoE) on ONE Colab A100-80GB High-RAM: pinned runtime, weights from HF, OpenAI-compatible endpoint, measured 2.8k-3.9k t/s prefill / 90-97 t/s decode, plus a verified pinned-RAM KV tier.
2026-09-23
rbayuokt/expo-adaptive-glass
B
Adaptive native glass surfaces for Expo and React Native. Scales blur, refraction and highlights to device capability and live frame performance.
2026-09-14
Show HN: Makefaster.dev
B
I had a bunch of extra Fable credits, so I spent around $10k in api costs doing autoresearch loops on the top 200 github repos with frontends. The goal was to speed up the frontend / improve the lighthouse score. I distilled the wins down into a leaderboard of most common improvements, then built a wrapper that iterates through them. It ties into your claude/cursor/codex subscriptions, so there's no subscription or payment - it should "just work" as long as you have a sub to one of those. It will ask you to submit your score / thinking trace at the end, but that's optional. Lemme know what you think / if it helps you out!
2026-09-02
shadcn-ui/cn
B
cn is a new engine for Tailwind class merging and conflict resolution. It replaces tailwind-merge and clsx. Same APIs. Full parity. And it is 30× faster.
2026-08-27
seanyourhighness/vllm-sm12x-nvfp4-dflash2
B
All-NVFP4 vLLM v0.27.1 + DFlash2 K7 speculative decoding for Blackwell (SM120/SM121): NVFP4 target/draft/KV, 262K context, optimized language-only serving.
2026-11-09
AmSach/maple
B
MAPLE: Manifold-Activated Pruning of Latent Experts for Sub-2-BPW MoE LLMs
2026-10-29
yt1988/me2cloud
B
Free cloud latency test from your own browser: median, jitter and loss to 360+ regions of AWS, Azure, Google Cloud and 14 more providers. 10 languages.
2026-10-20
nfzerox/VirtualIntelMacOnAppleSilicon
B
Run Intel versions of macOS Mojave, macOS Catalina, and macOS Big Sur on Apple silicon with Metal and OpenGL acceleration. This marks an exciting new milestone in the preservation of the history of Mac.
2026-09-25
ruc-system/quiver-ae
B
A low-latency, high-throughput, and billion-scale GPU-SSD vector search system.
2026-09-16
PerryTS/turnloop
B
Embeddable, cross-platform event-loop driver that the host turns: bounded turn(), completion-shaped I/O on epoll, kqueue, IOCP, WASI and the web.
2026-09-11
zmodelerlover/dlss5-neural-amd
B
ReShade add-on running the DLSS-NR network on AMD GPUs via HIP, RenoDX-shaped. Tested on PCSX2. Published to be forked. Discord: https://discord.gg/wYhvS3JSHM
2026-09-06
burakgon/stop-stutter
B
Smoother game streaming on Mac. Automatic AWDL boost for Moonlight, GeForce NOW, Punktfunk, Parsec & Steam Link. Native SwiftUI + Liquid Glass. MIT.
2026-09-05
mpaymenremora/CacheCraft
B
Semantic cache strategy bench: replay recorded request traces through candidate caching policies and measure which one actually pays off.
2026-11-06
neondeex/atlas-filemanager
B
Atlas — a native, GPU-accelerated, local-first file manager for Windows. Built with Rust and GPUI.
2026-11-02
KohakuBlueleaf/Tokenomos
B
Local inference for small VLMs: host work per step stays flat, so the GPU is the only bottleneck
2026-10-20
lxe/dsx
B
DeepSeek-V4.1-Flash-1017 (552B MoE) on 2x RTX 3090 + 128 GB RAM: Rust + CUDA engine and 2-bit quant. 30-37 tok/s decode.
2026-10-17
CLoaKY233/siltop
B
Terminal system monitor for Apple Silicon Macs: CPU, GPU, power, temperatures, processes.
2026-10-03
IuCC123/CLIProxyAPI-Rust
B
Your Claude, ChatGPT and Gemini subscriptions as one fast API. A single Rust binary with OpenAI, Anthropic and Gemini compatible endpoints, WebSockets and a built-in dashboard.
2026-09-30
B0yko/local-enough
B
Measure whether a back-office AI task can run on your own hardware: accuracy vs gold, p50/p95 latency, $/1k tasks, amortised hardware cost and break-even volume, plus an OpenAI-compatible router that sends each task to the cheapest model meeting your quality bar.
2026-09-29
TheeHorse/CnC-FPS-Unlocker
B
Run Red Alert 3, C&C3 Tiberium Wars and Kane's Wrath above 30 fps at normal game speed. Works with Tacitus.
2026-09-17
kdbhalala/avdslim
B
⚡ Drop Android Virtual Device (AVD) emulator RAM from ~8GB to ~1.5GB on Apple Silicon & Linux. Inspired by simslim.
2026-09-02
mohamedhossammohamed/m4-prefill-engine
B
Hardware-saturated Metal inference engine targeting the LLM prefill bottleneck on Apple M4 (~3.7x True GPU Speedup).
2026-11-07
PeakInspector/windows-drive-space-finder
B
Find what ate your C: drive: shows what grew since the last scan, explains it in plain words, gives the space back.
2026-11-02
Betelgeuse-AI/Betelgeuse
B
A custom fork for frontier-scale mixture-of-experts inference on an RTX 3060 with 12 GB VRAM, 16 GB system RAM and local NVMe.** Betelgeuse keeps dense computation on the GPU and brings routed experts into a bounded VRAM pool only when needed.
2026-10-21
ersinkoc/RustWTM
B
Windows task manager in Rust with a CPU governor service: throttle, cap, suspend or kill runaway processes, per-process rules and long-term usage insights.
2026-10-15
caveman-ai/trim-1
B
trim-1: a 1.2 MB model that keeps the lines of a context an AI agent needs
2026-10-14
AndrewTriesToCode/bg3mfx
B
MetalFX upscaling and frame generation for Baldur's Gate 3 on Mac
2026-10-10
thu-nics/TokenRouter
B
[NeurIPS'26] The official code implementation for paper "TokenRouter: Efficient Serving System for Token-Level LLM Routing"
2026-10-08
datologyai/zephon
B
Fast, flexible, elastically deterministic data loading for foundation model training
2026-10-04
joshuaswarren/coreglass
B
See where local inference loses speed on Apple Silicon under Linux
2026-10-03
StayLameBro/backburner
B
Your iPhone helps your Mac run a 27B model: faster prompt reading and more context over a USB-C cable
2026-09-11
ResistanceTo/MiniWatts
B
An iPhone battery, charging and thermal instrument built on private iOS APIs — real charger watts, what actually reaches the cell, all 30 temperature sensors, and the USB-PD handshake.
2026-09-05
morishuz/adaptive-frame-extractor
B
Adaptive video frame extraction for SfM, Gaussian Splatting, and photogrammetry.
2026-09-01
cratera-project/cratera
B
Bare-metal microVM execution engine in Rust. Sub-5ms snapshot restores, Zero-NIC network isolation, and 30 built-in language toolchains for AI code sandboxes & online judges.
2026-08-29
Holozing1/vram-guard
B
Detect and work around silently-corrupting VRAM on consumer GPUs. No ECC, no page retirement, no error - the driver just returns wrong data.
2026-08-27
mizchi/jsimd
B
Prebuilt WebAssembly SIMD kernels and Wasm-resident data structures for JavaScript
2026-11-09
LuckyIYI/fah-metal
B
FAH hand pose on Apple silicon in plain Metal: 115-234 fps on M5, with MANO mesh and hand-object contact
2026-11-03
0xCodez/pulsr
B
Pulsr — open-source recovery, strain and sleep coach for Apple Watch. Liquid Glass UI, on-device, no subscription.
2026-10-28
johnfink8/sfnx
B
CUDA engine that splits every layer of a production quality MoE LLM across three PCIe V100 cards delivering 382 tok/s on a used $1,800 box.
2026-10-23
huuan26/polars-anta
B
ANTA (Alpha & Numerical Technical Analysis): Ultra-fast, zero-C technical analysis and factor library for Polars.
2026-10-23
nicobailon/pi-cached-compaction
B
Pi compaction that reads the conversation's prompt cache, with a cache-expiry timer
2026-10-23
zoefix/agent-speed
B
Agent Speed — A lightweight, real-time token speed monitor for Codex and Claude Code. Track tokens per second (TPS), detect slowdowns, and monitor AI agent performance at a glance.
2026-10-17
kividbio/kividb
B
Redis®-compatible in-memory database written in Rust, with full-text and vector search
2026-10-14
uditya-kumar/MultiPath
B
Combine Wi-Fi, Ethernet and phone tethering on Windows/Linux. A local Tauri + Rust tray app, no server needed.
2026-10-03
zzzz7788990213-ops/EvoVLM
B
Evolving Multimodal Inference Programs: LLM-guided evolutionary search for accurate and efficient vision-language inference.
2026-09-25
cookieyes/cookiebannerbench
B
An open benchmark for what consent banners cost the pages they sit on — time to banner, payload, viewport blocked and page stability, measured on identical pages.
2026-09-11
Show HN: DOOM in the kernel, or fibers in eBPF
B
Some time ago my coworker who was working on first version of Perforator ( https://github.com/yandex/perforator) kept talking about eBPF, so I got curious about its ISA and restrictions. That was around the time I read about DOOM on pregnancy test, so I thought: "what if someone run DOOM inside of Linux kernel in eBPF? Surely, with some restrictions, this should still be possible?" I started to play around with it somewhere in 2024 - first stripping DOOM to bare minimum that is needed for showcase, simplifying code and trying to pass first restrictions that I encountered - function limit, recursion, memory access and checks. I tried to do this by hand, but it was, well, tedious labor. So I changed direction and instead started to do some of the things with LLVM-passes - rewriting memory access, simplifying and rewriting loops etc. But there were just so many corners. And given absence of time and all, I forgot about this project. But couple months ago I thought: "LLMs are quite good nowadays, so why not try again this time with more hands". I tried different approaches on how memory access could be "virtualized", how loops can be made bounded, and in general - how to run unbounded logic on such a machine. When I got DOOM running, I got carried away. Approach was so "generic", so it would be a crime not to try to run more things. And now we have it - lua running on xdp hot path, llama2 working with softfloats, and even cpython, that did not fit initially into 1M verifier budget, is running there now thanks to freplace. There is ton of work to do in order to make programs run faster, to make integration easier, but working examples are already there.
2026-09-05
michailfragkiskos/EdgeMosaic
B
Rapid on-device AI evaluation lab: manifest validation, benchmark plans, latency/memory/thermal aggregation, JSON+Markdown reports.
2026-08-27
kylehowells/MetalIconMorphView
B
A lightweight raster-based UIKit icon morphing view rendered directly with Metal.
2026-10-24
alexturkey/crisperwhisper-turkce-kelime-zamani
B
Verbatim speech-to-text with word timestamps on a 4 GB GPU: CrisperWhisper 2.0 Turbo (CT2 int8) + wav2vec2 CTC alignment. Turkish-first, Linux/WSL2.
2026-10-09
RocoreMatrix/YANchor
B
YANchor-4B: Effective Long-Horizon Reasoning in O(N) Time with O(1) Memory. Inference and evaluation code.
2026-10-04
PeterWarrington/XP-Code
B
A VS Code-style editor in under 200kb and compatible with Windows XP.
2026-10-02
0xSero/glm53-flash-offload
B
GLM-5.3-Flash EXL3 on one 24 GB RTX 3090 + DDR4: elastic GPU expert cache, zero-copy experts, AVX2 CPU tier, OpenAI API
2026-09-16
LevinAi-arch/rtx-5070ti-laptop-160w-power-limit
B
Research and proof-of-concept for raising the NVIDIA RTX 5070 Ti Laptop GPU power limit from 140W to 160W on Lenovo Legion Pro 5 Gen 10.
2026-09-16
serialexperimentslainnnn/afmf-linux
B
Vulkan layer to enable AFMF (AMD Fluid Motion Frames) For linux
2026-09-13
andrezaiats/snes9x-Z
B
Performance-optimized fork of Snes9x v1.63 (upstream commit 7a8878f1): bit-exact against upstream, dual-core PPU rendering, +61% mean frame-time gain.
2026-09-11
tonyd2wild/DeepSeek-V4.1-Flash-vLLM-DGX-Spark
C
DeepSeek-V4.1-Flash (552B MoE, MXFP4 experts, 1M ctx) on four NVIDIA DGX Sparks with vLLM TP4: Engram-on-disk patch, sm121 kernel build, launchers, measured numbers
2026-09-04
shtjww/llm-inference-capacity-handbook
B
大模型推理案头手册（开源版）：给定 GPU 算力，一个模型能扛多少 QPS？三层模型 × 三面墙 × 排队论 × 开环压测
2026-09-02
hbofz/NeSLE
B
GPU-native NES emulator + RL stack: thousands of parallel NES envs on one CUDA GPU, training agents on Super Mario Bros.
2026-08-29
ccarvalho-eng/beam_console
B
An embeddable process flight recorder and process map for BEAM applications.
2026-08-24
Redhatvale/Nvidia-Tesla-v100-nvfp4-pcie
B
Qwen3.8-27B in native NVFP4/FP8 on 2x PCIe Tesla V100-32GB (SM70): the PCIe runbook for v100-skinny + 1Cat-vLLM, with the 3 fixes that make it work without NVLink. 61-74 tok/s decode, MTP speculative decoding, OpenAI-compatible.
2026-10-17
gilnett/learned_index_lab
B
High-performance Piecewise Geometric Model (PGM) learned index implementation in pure Rust with strict epsilon error guarantees
2026-10-03
nuskey8/zerocbor
B
A zero-copy, zero-dependency, no_std-compatible, extremely fast CBOR (RFC 8949) serializer for Rust.
2026-09-29
Silent-Watcher/nestjs-quota
B
Production-grade API quota enforcement and usage metering for NestJS
2026-09-16
eriktaveras/modelnap
B
Let idle local LLMs sleep and get your Mac's memory back. A native macOS menu bar app to turn Ollama on and off and auto-free idle model RAM. Free and open source.
2026-08-29
drbermejor/mgs4Ultra120
B
Open-source ultrawide Hor+ and experimental 120 FPS patch for MGS4 Master Collection on PC
2026-11-07
q1054710965/tiny-image
B
轻图 Tiny Image — 在浏览器本地批量压缩图片，无需上传、无需安装。A private, offline-ready image compressor.
2026-11-02
Centiplatrap/pc-game-fps-doctor
B
One button: why the game stutters, why it lags and why it won't start. Pitstop measures with PresentMon, explains in plain words, fixes Windows only, and undoes every change in one click.
2026-10-06
lonewolf0622/HelixSR
C
Drop-in FSR 3.1 upscaler for Direct3D 12 that runs DLSS's network (Model E) on AMD GPUs
2026-10-01
apirathaiya/aiistream-q3.6
B
AiiStream Q3.6: byte-identical MoE expert streaming on Apple Silicon — runs Qwen3.6-35B-A3B on a 16 GB Mac
2026-09-25
yoheinakajima/glance-speedlab
B
A measurement-first lab for low-latency local camera inference with Glance.
2026-09-19
Peeter95/code-map
B
Fly over a codebase as a zoomable GPU treemap (2D and 3D city), built with Makepad
2026-09-17
dollarwong/ninfer-v100
B
NInfer on Tesla V100 (sm_70): Qwen3.8-27B deployment, build/convert scripts, systemd service, and cross-engine benchmark (vs llama.cpp). Fork of Neroued/ninfer with sm70 vision-encoding speedup.
2026-09-17
fuxingyin/scan2mesh
B
Scan2Mesh is a high-performance mesh-based LiDAR localization and tracking system that estimates LiDAR poses by aligning incoming scans with a pre-built 3D mesh.
2026-09-15
XiaoQAQ10086/FrameGen-Manager
B
一键为 RTX 20和30 系显卡部署 DLSS 帧生成工具，支持扫描游戏库和自动更新。One-click DLSS Frame Generation deployer for RTX 20/30 series.
2026-09-04
vamp1nn/vanguard
B
Compiler and offline partial evaluator aimed at the 3 futamura projections
2026-08-29
NIGos/dlss5-dx11-bridge
B
Lets D3D12-only DLSS 5 Neural Rendering add-ons run in DirectX 11 games