General Compute to deploy Positron inference chips in its cloud, live Q1 2027
Article timeline
The articles in this story. Times are ET.
SantiagoXPositron's inference hardware to handle decode in a split GPU setupAISantiago (@svpino) says a model's prefill stage runs on GPUs while the decode stage runs on Positron's specialized inference systems, which he says produces faster generation. The quoted post from General Compute says it will deploy a cluster-scale Positron system with hardware going live in Q1 2027.
🚨 AI News | TestingCatalogXGeneral Compute adds Positron Atlas inference systems to its cloudAIGeneral Compute has signed a multi-year partnership with Positron to offer its Atlas inference systems to cloud customers. Customers can rent Positron hardware through General Compute without buying it, with GPUs handling request input and Positron chips generating output. Atlas is already available in Oracle Cloud Infrastructure, and General Compute will integrate Positron's next-generation Titan systems, with hardware going live in Q1 2027.
elvisXGeneral Compute to deploy Positron cluster for fast inference, live Q1 2027AIGeneral Compute announces a cluster-scale deployment of Positron hardware as its next partner in its heterogeneous cloud. The company says demand for fast inference will outstrip supply over the next decade, and the hardware goes live in Q1 2027.
Rohan PaulXGeneral Compute pairs GPUs for prefill with Positron chips for decodeAIGeneral Compute runs prompt prefill on GPUs and hands token-by-token decode to Positron's memory-first systems, selling both as one endpoint. The author says prefill is a parallel compute job while decode is mostly waiting to pull weights and cache from memory, and Positron builds for that second job without relying on HBM. General Compute separately announced a cluster-scale deployment of Positron hardware, going live in Q1 2027.
