Hardware Tech News

A new GitHub repository called the Assembly Hall of Shame is crowdsourcing examples of notoriously inefficient assembly language code. The project highlights programming practices that severely degrade CPU performance and invites developers to submit their own poorly optimized instructions.

Major memory manufacturers Samsung, SK Hynix, and Micron have reportedly sold out their entire 2027 production capacity to meet surging demand from artificial intelligence companies. This massive allocation underscores the immense hardware requirements driving the current AI infrastructure boom.

AMD has acquired Taalas to boost AI inference performance by etching machine learning models directly into silicon. Early demonstrations of these model-specific integrated circuits reveal processing speeds of up to 17,000 tokens per second.

AMD has acquired Taalas to boost AI inference performance by etching machine learning models directly into silicon. Early demonstrations of these model-specific integrated circuits reveal they can generate up to 17,000 tokens per second, positioning AMD to deliver significantly faster AI processing.

Security researchers successfully tracked and eavesdropped on a WIRED reporter by exploiting vulnerabilities in a children's GPS smartwatch. This incident highlights the severe privacy risks and deeply insecure supply chains surrounding GPS-enabled wearable gadgets designed for kids.

A recent investigation reveals that many USB-C cables fail to meet the power and data transfer speeds advertised on their packaging. By physically testing a variety of cords, the author discovered that consumers can no longer trust label claims and must verify actual performance capabilities before purchasing.

YC S26 startup ProvenMetal has launched a rapid custom printed circuit board manufacturing service that handles builds of any specification and budget. The platform significantly reduces traditional multi-week wait times by delivering finished boards in as few as five days, and ordering is now open.

A developer has created STV, a new full-motion video codec designed specifically for the vintage 1985 Atari ST computer. This new codec aims to facilitate the porting of classic DOS-era games to the retro platform by enabling full-motion video playback on the decades-old hardware.

A recent phone update is mistakenly detecting users going for a run as a thief snatching their device. This glitch instantly locks the screen and abruptly closes fitness tracking apps, frustrating runners who rely on them to pace their workouts.

A new Rollercoaster Tycoon-inspired animation offers an isometric tour of a semiconductor factory by following a single wafer from raw sand to a finished microchip. The visual breaks down all twenty complex stages of the manufacturing process as the wafer travels through the facility.

Reader discussions regarding ASIC reverse engineering highlight the contrast between physically scanning dense chips and logically extracting netlists from layout files using standard industry tools. While generating a netlist from a layout is relatively straightforward, determining the actual functionality of the silicon remains the most significant hurdle. Participants also debated the practicalities of physical versus software-level hardware hacking, dismissing anecdotal claims of deducing digital logic opcodes using oscilloscopes as highly improbable.

A maker building an X-Y cartesian motion system has developed a more elegant alternative to zip ties and custom clips for anchoring timing belts. By utilizing an integrated timing belt loopback fastener, the project achieves a cleaner and more reliable connection between the belts and the moving carriage.

Nashville is using eminent domain to block DC Blox from building a new data center near the local zoo. The controversial proposal has faced widespread backlash, drawing significant opposition from both local residents and groups outside of Tennessee.

NVIDIA recently published a technical whitepaper for its upcoming Vera CPU architecture, but observers have identified a notable discrepancy within the documentation. This unresolved detail has sparked debate among hardware enthusiasts regarding the true specifications of the next-generation processor as they wait for the company to clarify the missing information.

Frustrated by Android's increasingly restrictive ecosystem and declining software quality, one user is abandoning the platform to switch to a Linux-based smartphone. This move highlights growing dissatisfaction among some tech enthusiasts who feel the popular mobile operating system is becoming too locked down and cluttered.

A recent technical deep dive explores the complex optimizations required to achieve near-native execution speeds when emulating RISC-V code on alternative processor architectures. By overcoming significant performance bottlenecks, these advancements in emulation technology pave the way for more efficient cross-platform software development. Ultimately, this progress supports the broader adoption and practical deployment of the open-source instruction set across diverse hardware environments.

Researchers have identified a unique seven-metal alloy forged by the extreme conditions of the Hiroshima atomic blast. This discovery suggests that similar complex multicomponent materials could potentially be synthesized in the future using non-nuclear intense heat sources. The finding has also prompted discussions regarding historical mid-century proposals to harness nuclear explosions for industrial manufacturing and civilian engineering.

Maple-Preview is a 20-billion parameter Mixture of Experts model trained from scratch in a ternary format, achieving speeds of 120 tokens per second directly on an iPhone. While the project demonstrates impressive mobile efficiency, community testing reveals the model still struggles with confident hallucinations on niche topics compared to larger alternatives.

A newly published arXiv paper evaluates the real-world performance of Zigbee compared to the emerging Matter over Thread protocol for Internet of Things networks. The study provides practical insights into how these competing smart home standards operate in everyday environments to help guide future IoT infrastructure decisions.

The Video2NAND project demonstrates how the VP8 video codec can be theoretically abused to construct NAND logic gates by exploiting its maximum resolution and minimum block size parameters. While some commenters speculate this technique could eventually be scaled to build a small CPU, others argue the project's title overstates its current computational capabilities. The discussion highlights the creative potential of repurposing media formats for unconventional computing, even if achieving true Turing completeness in practice remains a challenge.

Waymo has officially launched its fully autonomous ride-hailing service in Dallas, allowing anyone in the city to hail a driverless vehicle. Starting today, users can download the Waymo app to book a ride as the company continues to expand its robotaxi footprint.

Godox has introduced the C100, a new transparent viewfinder camera featuring a screen with over 50% light transmittance and a built-in TF card slot supporting up to 128GB of storage. The device offers a continuous battery life of over 1.5 hours and utilizes USB 2.0 for both data transfer and charging.

Hardware startup Oxide Computer has raised $445 million in a Series D funding round, marking its fourth major capital raise in recent years. Although the company states this influx of capital ensures its long-term independence, community discussions reflect skepticism about whether massive funding guarantees actual business success. Commenters also note that Oxide targets specialized markets like high-frequency trading firms using advanced networking technologies, rather than pursuing traditional enterprise software customers.

A new open-source project called Swiftlet enables users to run massive large language models on Apple devices with remarkably low memory requirements. The tool successfully demonstrates running an 80-billion parameter Qwen model using just 4.3 GB of RAM on a Mac, alongside a 35-billion parameter model on an iPhone. This advancement significantly lowers the hardware barrier for running powerful local AI inference on everyday consumer devices.

Reader discussions reflect on the ill-fated Windows XP 64-bit Edition for Intel's Itanium architecture, noting it was uniquely built on the Windows Server 2003 kernel. Commenters attribute the platform's failure to reach the consumer market to high costs, poor x86 emulation, and the rapid rise of AMD's competing x86-64 architecture. Ultimately, Itanium was relegated to niche high-performance computing roles before Microsoft completely dropped operating system support after Windows Server 2008 R2.

Serving frontier AI models like Kimi and GLM at scale requires overcoming significant GPU memory constraints through techniques like KV cache quantization and weight compression. These optimizations enable faster and cheaper deployment, while newly implemented integrity checks ensure safer model serving.

Security researchers have successfully hacked the Windows 365 Link, a locked-down thin client designed exclusively to connect to Microsoft's Cloud PC. By breaking out of its restricted Windows 11 environment, they demonstrated that the device can be repurposed, echoing past failures in Microsoft's attempts to lock down its hardware.

A recent guide on replacing the Kobo Libra H2O battery has prompted readers to discuss the broader repairability of Kobo e-readers using commodity parts. Although newer models use soldered eMMC storage, users emphasize that screens, SD cards, and standard lithium-ion batteries can be easily swapped to extend the devices' lifespans. The community strongly encourages repairing aging e-readers with generic components instead of purchasing replacements.

Moving beyond the hardware control offered by BASIC, this installment of the ZX Spectrum System Tour dives into machine language to explore the system's text mode. The article begins a multi-part deep dive into the lower-level capabilities of the classic computer's text rendering.

A new open-source project called AirLLM enables the inference of 70B parameter large language models using only a single 4GB GPU. This dramatic reduction in hardware requirements makes running advanced AI models highly accessible to users with limited computing resources. Developers can view the code and contribute to the project's ongoing development on GitHub.