NPUs

Winsage
April 11, 2025
Microsoft's latest update for Windows 11 Beta introduces the ability to read MSN articles directly from the Widgets feed, allowing users to access full articles, slideshows, and videos without using a browser. Users can customize the visibility of the MSN feed on their Widgets board. The update also enhances Copilot+ PCs with an AI-powered semantic search bar for natural language queries and improves the Narrator feature, which now analyzes screen content to provide descriptions of images and visual elements for visually impaired users.
Winsage
April 11, 2025
Microsoft is testing its Recall feature for Windows 11 in the Release Preview channel with build version 26100.3902. Recall is designed for Copilot+ PCs with powerful neural processing units (NPUs) and will initially support English, Chinese (Simplified), French, German, Japanese, and Spanish. The feature uses AI to analyze screenshots of user activity, allowing for natural language searches. Microsoft has implemented security measures like Windows Hello sign-in to address privacy concerns. Recall is expected to be available in early 2025 in most markets, but not in Europe during the initial rollout, due to data regulations.
Winsage
April 11, 2025
Microsoft has begun rolling out the Recall feature in Insider build version 26100.3902 after addressing previous security vulnerabilities. The Recall feature uses Neural Processing Units (NPUs) in Copilot+ PCs to capture screenshots and convert them into natural language descriptions for improved productivity. The new build also introduces the Click to Do feature for Snapdragon-powered PCs, enhances Phone Link functionality for easier content sharing, and includes the Windows Share feature for quick image edits before sharing. The rollout of these features will be gradual.
Winsage
March 25, 2025
Microsoft is rolling out its Semantic Search feature to Windows Insiders Beta in Preview Build 26120.3585 (KB5053644). This feature enhances user experience by enabling intuitive searches across files, settings, and applications, integrated into the Start menu and Settings. Users can conduct contextual searches by typing related keywords, retrieving relevant documents and images easily. Semantic Search also extends to the cloud, allowing searches for matching files in OneDrive. It is exclusive to Copilot+ PCs, utilizing advanced Neural Processing Units (NPUs), and supports multiple languages and various file formats.
Winsage
March 25, 2025
Copilot+ PCs will receive an upgrade to the Windows Search feature, which uses AI and dedicated hardware for improved document, image, and settings search capabilities. This enhanced feature is currently exclusive to Snapdragon-powered Copilot+ PCs, with support for AMD and Intel devices expected soon. The new search allows natural language queries and integrates semantic indexing, enabling users to find content without exact file names. Users can also search for cloud-stored photos using personalized phrases, with results from OneDrive. The improved Windows Search will be available through the Windows Insider Program by early 2025, with a consumer-focused AI event planned for April 4, 2025, where new features will be unveiled.
Winsage
March 4, 2025
Microsoft is launching the DeepSeek R1 7B and 14B distilled models via Azure AI Foundry, following the introduction of NPU-optimized versions of the DeepSeek-R1 1.5B model for Copilot+ PCs. These models will be available on Copilot+ PCs powered by Qualcomm Snapdragon X, Intel Core Ultra 200V, and AMD Ryzen processors. The NPUs in these PCs can execute over 40 trillion operations per second and are designed for efficient local AI model execution, minimizing battery and resource consumption. The DeepSeek models utilize 4-bit block-wise quantization and int4 per-channel quantization for enhanced performance. Developers can access all distilled variants of the DeepSeek models through the AI Toolkit VS Code extension, allowing for local deployment and experimentation. Copilot+ PCs combine local computing capabilities with Azure's resources, facilitating a new paradigm of continuous computing for AI applications.
Winsage
February 20, 2025
The Surface division at Microsoft has experienced significant internal changes over the past 18 months, including the departure of Panos Panay and new leadership. This has led to a strategic refocusing of the Surface portfolio, moving away from experimental hardware designs to more conventional offerings aimed at a broader audience. The company has discontinued extravagant devices like the Surface Studio and Surface Duo, favoring products that resonate with business customers and general consumers. Last year, Microsoft introduced AI-powered Windows computers known as Copilot+ PCs, which emphasize continuity and refinement of established designs rather than radical innovations. The Copilot+ PC initiative has allowed Windows on Arm to demonstrate its viability, with Snapdragon chips competing effectively in the Windows ecosystem. Surface continues to play a crucial role in shaping the Windows ecosystem, but with a more cautious approach that prioritizes ecosystem health over creating new markets. Future form factor experimentation will likely be delegated to OEM partners, while Microsoft focuses on transitioning users to Windows 11 and promoting Copilot+ PCs. Stakeholders can expect a steady stream of sensible Surface devices and incremental updates in the near future.
Winsage
February 11, 2025
Microsoft has announced that Windows 10 Version 22H2 will be the last version, ending new updates and security patches in October. Users can continue to use their systems, but they will be at risk for security vulnerabilities without updates. Options for users include: 1. Paying for extended support, which offers continued security updates for a limited time after Windows 10's end of life. 2. Switching to Linux or ChromeOS Flex, both of which are free alternatives that can keep older PCs operational without hardware upgrades. 3. Purchasing a new PC, which may provide the latest technology but comes at a high cost. Without security updates, systems may become targets for cyber threats, risking data and privacy.
Winsage
December 14, 2024
In 2024, Microsoft introduced the "Copilot+ PC" branding for AI-capable laptops, while Apple launched Apple Intelligence. These developments have led to mixed outcomes, with features like real-time translations and on-device speech-to-text being beneficial, but others, such as Windows Recall, still proving their value. By 2025, mainstream developers are expected to integrate on-device AI into Windows applications, influencing consumer purchasing decisions. The term "TOPS" (Trillions of Operations Per Second) is becoming important for evaluating the AI performance of Windows laptops, with a minimum of 40 TOPS required for Microsoft's "Copilot PC+" designation. Qualcomm's Copilot+ PCs reported around 45 TOPS, significantly higher than Intel's 11 TOPS. By the end of 2024, premium Windows laptops are expected to see a three- to four-fold increase in NPU performance compared to 2023 models. Analysts speculate further performance improvements may occur towards the end of 2025. Despite the potential for a two- to three-fold enhancement in on-device AI performance, experts caution against overemphasizing TOPS figures, which may not accurately reflect real-world performance. The lack of a unified API for leveraging NPU capabilities in Windows complicates matters for users of Copilot+ laptops without Qualcomm chips. Although AMD and Intel have released competitive chips, Qualcomm currently holds an advantage with exclusive support for certain applications. Microsoft is promoting its low-level machine learning API (DirectML) and the Windows Copilot Runtime, which may enhance the Copilot+ PC ecosystem. While cloud-based AI solutions remain an option, the cost of these services is expected to rise, making on-device AI more appealing. The introduction of ChatGPT Pro highlights the financial implications of cloud access compared to on-device NPU usage, which incurs no additional costs. The pace of on-device AI adoption in Windows' software ecosystem is anticipated to accelerate in 2025.
Winsage
December 6, 2024
The Applied Sciences team has developed the small language model (SLM) Phi Silica, which enhances power efficiency, inference speed, and memory efficiency for Windows 11 Copilot+ PCs using Snapdragon X Series NPUs. Phi Silica is designed for on-device use and supports multiple languages, featuring a 4k context length. Microsoft announced that developers will have access to the Phi Silica API starting January 2025. The Copilot+ PCs can perform over 40 trillion operations per second, achieving significant performance improvements when connected to the cloud. Phi Silica utilizes a Cyber-EO compliant derivative of Phi-3.5-mini, and its architecture includes components such as a tokenizer, detokenizer, embedding model, transformer block, and language model head. The model's context processing consumes only 4.8mWh of energy on the NPU, with a 56% improvement in power consumption compared to CPU operation. Phi Silica features 4-bit weight quantization for efficiency, rapid time to first token, and high accuracy across languages. The model was developed using QuaRot for low-precision inference, achieving 4-bit quantization with minimal accuracy loss. Techniques like weight sharing and memory-mapped embeddings were employed to optimize memory usage, resulting in a ~60% reduction in memory consumption. Innovations such as a sliding window for context processing and a dynamic KV cache were introduced to expand context length. The model has undergone safety alignment and is subject to Responsible AI assessments and content moderation measures.
Search