Launching today

RamShared

Launching today

Use idle GPU VRAM as safe swap space on Linux & WSL2

3 followers

Use idle GPU VRAM as safe swap space on Linux & WSL2

3 followers

Visit website

RamShared borrows idle GPU memory as a high-speed swap cushion on Linux & WSL2—without freezing your OS. Unlike naive GPU swap tools that crash during memory eviction, RamShared uses a crash-safe cascading tier: • zram (compressed RAM) • GPU VRAM (cushion) • NVMe (last resort) Features active latency canary monitoring, automatic page demotion, and Rust-native safety via ublk/NBD drivers to keep your machine responsive under heavy loads.

Free

Launch tags:Linux•Open Source•Developer Tools

Launch Team / Built With

Framer AI AgentsDesign and publish professional sites with AI

Promoted

OpenAI

OpenAI Day

What could you build with GPT-5.6 that wasn’t practical before?

Maker

With GPT-5.6's advanced reasoning capabilities, we can move beyond static heuristic rules for memory management. RamShared can leverage real-time predictive analysis of developer workload memory patterns—especially during massive local LLM inference—to dynamically fine-tune VRAM-to-disk cascading thresholds, latency canary sensitivities, and page demotion strategies before OS thrashing ever occurs.

Report

19h ago

Maker

📌

Hey Product Hunt! 👋 I'm Emerson, creator of RamShared. If you compile large codebases, run Docker containers, or spin up local LLMs, you’ve probably hit the dread of system thrashing when RAM runs out. Meanwhile, your dedicated GPU might be sitting right next to it with 8GB–16GB of completely idle VRAM. You paid for that silicon—why not use it? ### The Problem with "Just put swap on the GPU" Many have tried putting swap on GPU VRAM, but under host memory pressure (especially on WSL2/Windows WDDM), graphics memory eviction can spike read latency up to ~1.2 seconds. If VRAM is your primary swap, your entire desktop freezes. ### How RamShared Fixes This RamShared treats GPU memory not as primary RAM, but as a cautious second-tier cushion: * Cascading Priority: Data goes to zram first (fastest). Only colder pages spill into VRAM before eventually hitting disk. * Active Canary Checks: If VRAM latency spikes or the host claims graphics memory, RamShared silently demotes pages to disk and unloads the tier safely—keeping your system responsive. * Crash-Safety: Built in Rust around reliable kernel driver layers (ublk/NBD) so daemon drops won't freeze your OS. It’s open-source, lightweight, and ready for Linux & WSL2 users. Check out the repo on GitHub, give it a spin, and let me know your thoughts or feedback! 🚀

Report

1d ago

Framer AI AgentsDesign and publish professional sites with AI

Promoted

OpenAI

OpenAI Day

What could you build with GPT-5.6 that wasn’t practical before?

Maker

Report

19h ago

Maker

📌

Report

1d ago