Desk
NVIDIA hardware, software, and developer news.
NVIDIA and CrowdStrike put red and blue agents in a closed loop and found the open-model pipeline produced fewer detections than the frontier system — but the only ones that held up.
Nvidia's own sizing guide argues most inference budgets are wrong in the first spreadsheet, not the second data center. The interesting admission is that smaller models are usually the answer.
DLSS 5 ships with one supported game, a 50 to 60 percent performance cost, and 6x frame generation as the price of entry. That is a tech demo with a release date.
NVIDIA and Anthropic are turning protein folding into an agent-callable workflow, but the reported confidence gains still describe hypotheses that need laboratory validation.
NVIDIA's NuRec re-renders recorded drives through new sensor rigs, offering automakers a bridge between existing road data and vehicles that have not yet collected their own.
Nvidia is financing the custom-silicon market that could weaken its GPU dominance, provided those chips still speak Nvidia’s infrastructure language.
Nvidia’s TensorRT Model Connect packages the messy path from open-model checkpoint to native inference, but only for supported architectures.
GeForce NOW’s Gamescom package expands cloud gaming’s reach across browsers, televisions and Steam hardware while using graphics controls to sell the experience.
Vera is Nvidia’s bet that agentic workloads will make CPU orchestration and memory movement as strategic as GPU computation.
Nvidia’s NVHBM pitch treats memory, package area and rack connectivity as one answer to the economics of custom AI accelerators.
The Annapurna Labs partnership makes NVLink Fusion a route for Amazon’s custom chips to share a rack-scale language with Nvidia GPUs.
NVIDIA's COMPASS workflow makes robot adaptation more repeatable, but its real contribution is disciplined evidence and approval gates rather than autonomous training magic.