Microsoft will hold a Windows and Surface event in San Francisco on October 7, 2026, giving its broader local-AI strategy a concrete hardware showcase after outlining Windows 11 as a platform for local AI agents. CEO Satya Nadella, Windows and Devices chief Pavan Davuluri, and NVIDIA CEO Jensen Huang are scheduled to appear, pointing to a major focus on on-device AI and NVIDIA’s RTX Spark platform.

Source

NVIDIA CEO Jensen Huang told the hosts of the All-In podcast, Chamath Palihapitiya, Jason Calacanis, David Sacks and David Friedberg, that he rejects predictions of AI-driven extinction as unscientific. A surprise call from President Donald Trump reinforced the event’s pro-AI message, with Trump saying that robots would not take over the world and calling fears about data centers and AI takeover a “hoax.”

Source

Nvidia has quietly repurposed its flagship GB202 GPU into the RTX Pro 5500 Blackwell, a workstation card with the GeForce RTX 5090’s 21,760 CUDA cores but 84 GB of GDDR7 memory. The added capacity is designed for local AI inference, rendering, and other professional workloads that exceed the 32 GB limit of Nvidia’s gaming flagship.

Source

A new DDRop hardware attack undermines Intel TDX, Intel Scalable SGX, and AMD SEV-SNP by silently discarding memory writes, allowing servers to reuse older encrypted values without detection. The researchers disclosed the findings to Intel, NVIDIA, and AMD before publication, but administrators should not expect a software-only fix: the weakness is rooted in how current confidential-computing hardware handles memory integrity.

Source

Perplexity has brought its local Portable Computer AI agent to Windows PCs, but only systems with an NVIDIA RTX or RTX PRO GPU offering at least 24GB of VRAM qualify. The agent runs models, tools, file processing, and scheduled tasks locally, while retaining an optional cloud connection for work that needs stronger reasoning or web access.

Source

Windows 11 update KB5121003 is being linked to sudden AC-power disconnects on Lenovo laptops, with affected systems switching briefly to battery power, black-screening, or shutting down unexpectedly. The strongest evidence so far comes from a rollback test that eliminated the failures, although the exact change responsible has not been confirmed by Microsoft or Lenovo.

Source

Windows 11 updates KB5124008 and KB5126052 are being linked to severe instability on AMD-based PCs, including driver timeouts, freezes, display disconnects, and hard crashes. The troubling detail for administrators is that uninstalling both updates—and even replacing AMD graphics drivers—did not reliably restore affected systems.

Source

Microsoft has confirmed a new Windows 11 Patch Tuesday problem that disables USB Audio Class 1.0 devices with Device Manager’s Code 10 error. The failure affects client systems running Windows 11 26H1, 25H2, and 24H2, while Windows Server installations are not impacted; Microsoft has not yet provided a fix date.

Source

1.6T Ethernet is moving beyond specifications and into hardware interoperability testing, as the Ethernet Alliance prepares to demonstrate switches, optical modules, and cables from more than a dozen vendors at ECOC. In parallel, the Optical Internetworking Forum’s 1600ZR standard will carry a complete 1.6T Ethernet connection over one coherent wavelength for data-center interconnects spanning up to 120 kilometers.

Source

Marques Brownlee’s first impressions of Apple’s iPhone 18 Pro and iPhone Duo suggest the foldable is the company’s most significant iPhone redesign since 2007. The iPhone Duo combines a pocket-sized outer display with a 7.6-inch inner screen, while the iPhone 18 Pro focuses on less visible upgrades such as a 2-nanometer A20 Pro chip, longer battery life, and a variable-aperture main camera.

Source

Samsung is bringing Mistral AI models, including Mistral Large, inside its semiconductor operations while taking a strategic equity stake in the French startup. The on-premises deployment will analyze fab data for defect detection, equipment tuning, chip design, and yield improvement without sending sensitive engineering information to external cloud services.

Source

Arm’s semi-custom Neoverse CSS N4 “Falcon” platform raises the ceiling for Arm-based server silicon to 128 cores per die, built on TSMC’s N3P process. Customers can tailor core counts, cache, memory, I/O, and chip-to-chip connectivity for cloud, AI, and high-performance computing systems rather than adopting a fixed processor design.

Source

Intel is reportedly preparing another 10% increase for PC CPU prices on October 5, 2026, marking the third reported hike since late 2025. The move would come months before a major Intel product launch expected in March 2027, while AMD may follow with new processors between June and July.

Source

Intel has released Wi-Fi and Bluetooth driver packages 24.70.0 for Windows 10 and Windows 11, bringing security updates, stability improvements, and updated 6 GHz regulatory support. The release covers current Wi-Fi 7 and Wi-Fi 6E adapters as well as several older Wireless-AC models, making it relevant to a broad range of systems.

Source

Qwen 3.8 27B can fit on a single RTX 5090, but real-world performance depends more on the inference engine and context configuration than on the card’s 32 GB of VRAM. The same GPU can take roughly 30 minutes to produce a long-context response with llama.cpp, deliver about 20 tokens per second through a basic vLLM setup, or reach around 200 tokens per second with a more optimized engine.

Source

HPE Alletra Storage MP B10000 Release 6 is now generally available, making the vendor’s unified-storage strategy concrete by placing block and file workloads on the same disaggregated scale-out architecture. Performance and capacity can grow independently, while ransomware detection now covers both data types.

Source

GPT-6 Astra could shift much of the performance pressure from AI accelerators to enterprise CPUs by launching multiple computer-use agents in parallel. The model’s rollout to paid ChatGPT customers already positioned it as a more capable but more expensive successor to GPT-5.6 Sol; the emerging multi-agent behavior adds a new concern for administrators running Astra-connected workloads.

Source

NVIDIA says updated llama.cpp and vLLM inference software can deliver up to 1.9x higher throughput for local AI agents, but that headline result comes from a specific Qwen3.6-35B test on a GeForce RTX 5090. Other tested configurations produced gains between 1.2x and 1.5x, making the benchmark details important for anyone evaluating local model performance.

Source

NVIDIA has released optional GeForce Hotfix Driver 616.86 for 64-bit Windows 10 and Windows 11, targeting three display failures that can disrupt browser use and remote administration. The update restores virtual-display creation and fixes black screens in Remote Desktop Protocol sessions.

Source