All articles tagged with #distributed computing
Nvidia announces RTX PAIR, a free tool that automatically pools GPUs across a home network to accelerate AI tasks. It works with Ollama and LM Studio, runs on Windows, Linux, and macOS, and is open-source under Apache 2.0. In a demo, multi-device setups completed tasks faster than a single PC, signaling a push to turn consumer GPUs into a local AI farm ahead of the October RTX Spark launch.
Over two decades, the SETI@home project used millions of volunteers’ idle computers to analyze Arecibo data for technosignatures, producing about 12 billion detections. A decade-long Nebula back-end pipeline compressed those into roughly a million candidate groups, which were then manually reviewed to yield around 100 sky locations for follow-up with FAST. No confirmed alien signal has been found; the work illustrates that detections are software events, and that careful filtering and cross-observatory verification are essential before any claim of detection. The study also highlights that the bottleneck is expert interpretation and follow-up, not processor power.
UCSD researchers, with Google, are turning retired Pixel phones into standalone compute nodes by stripping hardware and running Linux with orchestration (Kubernetes); in SPEC benchmarks, single cores on older Pixels outperform some server configurations, and 25–50 phones can match a dual-socket server’s output, with a 20-phone cluster already supporting classes for more than 75 students. The project envisions scaling to about 2,000 phones for hundreds of classes, offering a cheaper, lower-carbon alternative to cloud infrastructure—though not suited to hyperscale workloads due to hardware heterogeneity and management complexity.