GitHub

IT House News, 3 June, opened today at the Build 2026 Developer's Congress, was attended online by the CEO Wong In-hoon in Taipei and participated in the Microsoft CEO Satia Nadra keynote address.

RTX Spark, DGX State for WindowsNVIDIA Open Model,GitHub OpenShell in Copilot is discussed in a number of partnerships, including the safe running of the company and the next generation of AI factories.

Microsoft and Weeda also announced a series of in-depth collaborations and demonstrated at the Build Congress across Windows devices,Azure. A unified acceleration calculator for clouds and local deployment provides end-to-end support for developers to build, run and expand AI smarts and physical AI.

At the end of the Windows device, YVD and Microsoft will redefine the PC for the AI Smart Age. RTX Spark is the first Windows PC platform for personal intelligence, providing 1 petaflop 's AI computing power, a single memory of up to 128 GB, a day ' s extension, and a non-defeating ion AI and graphic performance.

The platform, which integrates CUDA, RTX, DLSS and TensorRT, has accumulated technology for more than 30 years, and the notebooks and small desktops with RTX Spark will be launched this autumn by Microsoft Surface, Wahe-seok, Dale, Hewlett-Packard and Microstar.

The DGX status for Windows is described as the current most performing desktop-level AI supercomputer that carries the NVIDIA GB300 Grace Blackwell Ultra desktop superchips with a maximum 748GB unified memory and 20 petaflops FP4 performance, which can run an advanced model of up to 1 trillion parameters locally. The system is expected to be launched in the fourth quarter by the manufacturers of Wahe-seok, Dale, Tweedy, Hewlett-Packard, Microstar and Super-micro. Both products run NVIDIA OpenShell - a secure design for an autonomous intelligence.

At the modelling and platform level, the NVIDIA open model combination has officially landed in Microsoft Foundation. Nemotron 3 Ultra, a new open front-line reasoning model for coding, scientific research and long-term running intelligence in business streams, will be available this month in the Foundry hosting calculations, along with Nemotron 3.5 ASR and Nemotron 3.5 content security models for voice recognition.

In addition,Anthropic Yes. Claude. The model is now operational on the Appure NVIDIA GB300 Blackwell Ultra system and is open to clients in the coming weeks.

The NVIDIA Agent Toolkit and NVIDIA NemoClaw blueprints provide an open-source platform for developers to build production-level smarts on Foundry, and CUDA-X libraries (including cuDF, cuopt, AI-Q and NeMo) are now available to smarts as field-specific skills.

For the data layer, the NVIDIA acceleration calculation is now embedded in Microsoft Fabric Data Warehouse. The microsoft's internal benchmark tests show thatSQL Implementation rate achieved CPU The baseline is up to six times higher and up to seven times higher than the provider of the three other mainstream cloud data warehouse under high and load. Two companies indicated that this was the result of multi-year in-depth engineering collaboration from research to production, enabling the enterprise data layer to keep up with the continuous search and reasoning of AI intelligence.

In the field of physics AI, Microsoft is integrating NVIDIA 's open source physical AI skills and tools with the Azure and physical AI tool chains to provide developers with a cosmos 3-based mix Transformer. A unified platform for simulation, training and deployment of autonomous systems such as robots, auto-driving vehicles and industrial systems. Cosmos 3 ranks first in the open model in key benchmarking tests such as visual reasoning, world generation and action generation. For local deployment scenarios, Microsoft introduces the Foundation Local on Azure Local into the NVIDIA RTX PRO 6000 Blackwell Server Evaluation platform, in conjunction with the Nemotron Open Model family, where enterprises can operate high-performance AI loads in local, mixed or sovereign environments. Foundry Local on Azure Local currently supports multinodes deployment and vLLM Run time.

In terms of security development, NVIDIA OpenShell has been integrated into GitHub Copilot. Each intelligent body operates in its own sandbox container, and each external call is evaluated according to the strategy before it touches a document, network or voucher, which is in the form of a code and versioned and managed in a warehouse and can be updated immediately. OpenShell is based on the Apache 2.0 Open Source Protocol, unrelated to the model, covering local, mixed and cloud environments.

In terms of infrastructure, Microsoft ' s AI plant in Fairwater, Wisconsin, has been put on line in advance, operating hundreds of thousands of NVIDIA Grace Blackwell systems and interconnecting with a similar AI plant in Georgia, constituting an extended distributed AI system. Wong In-hoon also personally admitted that Microsoft had the largest deployment of Blackwell chips.

Microsoft has completed the validation of the NVIDIA Vera Rubin platform, which is now fully operational and will be deployed to the Azure data centre. Vera Rubin can be deployed in parallel with Blackwell without retrofit, increasing by a maximum of 10 times per MW of reasoning throughput, reducing the cost per smart token by one order of magnitude, and including NVIDIA classified calculations to protect models and data security.

References:

Advertising statements: The external jump links (including not limited to hyperlinks, 2D codes, passwords, etc.) contained in the text are used to convey more information and save time for selection purposes only for reference purposes, which are included in all IT House articles.