Releases·Tool Release
DistribAI v2 launches: distributed training across consumer devices via C++/Libtorch with PyTorch support
An independent developer released DistribAI v2, a platform for distributed training that pools consumer devices and free cloud GPUs. It is built in C++ with Libtorch, supports distributed PyTorch trainers, and handles crash, malicious-actor, and unstable-connection edge cases automatically. In a demo, free Colab, Kaggle, and Molab GPUs plus a local 4070 SUPER delivered compute of roughly two fully loaded 5090s and combined VRAM of about five 5090s at no cost.
3 engagement1 sourceSat, Sep 26, 2026, 01:05 AM
AI Overview
- DistribAI v2, built in C++ with Libtorch, distributes PyTorch trainers across consumer devices and handles crashes, malicious actors, and unstable connections
- Testing with free Colab, Kaggle, Molab GPUs plus a 4070 SUPER yielded ~2 fully loaded 5090s of compute and ~5 5090s of VRAM
- v2 adds shareable join links plus Cloudflared/ngrok support for free server hosting; code is on GitHub (naxium-oss/DistribAI)
Entities
Related
Releases·Tool Release
PyTorch adds Thunderbolt communication backend for distributed training on Macs
A GitHub commit indicates PyTorch is developing a fast Thunderbolt communication backend, enabling distributed model training across Macs. This would allow developers to leverage multiple Macs for larger models without specialized hardware.
A GitHub commit indicates PyTorch is developing a fast Thunderbolt communication backend, enabling distributed model training across Macs. This would allow developers to leverage multiple Macs for larger models without specialized hardware.
26 engagement1 sourceredditTue, Jul 14, 2026, 07:40 PM
Products·launched
NuNet launches Undergryd, a decentralized compute network for AI workloads
NuNet has launched Undergryd, a product that provides decentralized computing power for AI and machine learning tasks. It connects compute providers with users who need GPU resources, enabling distributed AI workloads without relying on centralized cloud providers.
NuNet has launched Undergryd, a product that provides decentralized computing power for AI and machine learning tasks. It connects compute providers with users who need GPU resources, enabling distributed AI workloads without relying on centralized cloud providers.
0 engagement1 sourcerssSat, Jul 18, 2026, 08:38 AM
Products·launched
Cloud AI coding tool to compare models side by side
A cloud-based AI coding tool that lets developers run multiple AI models side by side on real projects to compare their performance, rather than relying on benchmarks. It allows users to run code without keeping their laptop open. Early testers get free access to Gemini and Deepseek models.
A cloud-based AI coding tool that lets developers run multiple AI models side by side on real projects to compare their performance, rather than relying on benchmarks. It allows users to run code without keeping their laptop open. Early testers get free access to Gemini and Deepseek models.
3 engagement1 sourceredditFri, Aug 14, 2026, 05:45 PM
Products·building
P2P network for running local LLMs across multiple machines
DaiHive is a peer-to-peer network that splits large language models across multiple machines, allowing users to run models too big for a single computer. It supports private team mode for isolated groups, targeting users with multiple idle machines who want to run large models locally.
DaiHive is a peer-to-peer network that splits large language models across multiple machines, allowing users to run models too big for a single computer. It supports private team mode for isolated groups, targeting users with multiple idle machines who want to run large models locally.
2 engagement1 sourceredditFri, Jul 17, 2026, 09:36 AM
Releases·Tool Release
Nvidia launches free open-source Personal AI Router (PAIR) that links idle PCs into a local AI inference cluster
Nvidia announced the Personal AI Router (PAIR), a free open-source software tool that discovers compatible PCs on a local network and routes independent inference requests to whichever machine has capacity. It integrates with Ollama and LM Studio and targets local agent workflows that split tasks into smaller jobs. Despite the name, PAIR is software, not a hardware router.
Nvidia announced the Personal AI Router (PAIR), a free open-source software tool that discovers compatible PCs on a local network and routes independent inference requests to whichever machine has capacity. It integrates with Ollama and LM Studio and targets local agent workflows that split tasks into smaller jobs. Despite the name, PAIR is software, not a hardware router.
210855 engagement4 sourcesrss, reddit, youtubeThu, Sep 3, 2026, 04:00 PM