➀ NVIDIA disclosed the architecture of its GB10 SoC at Hot Chips 2025, a multi-die solution combining a Blackwell-based GPU and MediaTek's 20-core Arm CPU, both fabricated on TSMC's 3nm process;

➁ The GB10 powers NVIDIA's DGX Spark workstation, offering FP4 support for AI workloads and enabling scalable cluster configurations;

➂ Designed for high-performance computing, the chip emphasizes collaboration between NVIDIA and MediaTek, targeting advanced AI inference and training applications.

➀ NVIDIA unveiled its co-packaged silicon photonics switches and Spectrum-X Ethernet technology at Hot Chips 2025, targeting gigawatt-scale data center networks for AI workloads;

➁ Spectrum-X reduces network jitter and improves NCCL performance, enabling efficient multi-tenant AI training with 1.9x better scale-out performance compared to traditional Ethernet;

➂ The Spectrum-XGS "scale-across" solution extends AI training across multiple data centers through distance-aware algorithms, with CoreWeave likely being the first deployment partner.

➀ NVIDIA DRIVE AGX Thor 开发者套件已开放预购，计划于9月发货，基于 Arm Neoverse V3AE CPU，专为自动驾驶的复杂 AI 工作负载设计；

➁ 平台结合 Arm 架构的确定性性能与 NVIDIA Blackwell GPU，提供最高 2,000 FP4 万亿次算力，支持预测性 ADAS、实时驾驶监控及语音交互等多任务同步运行；

➂ 提供桌面版（SKU10）和车载版（SKU12）两种开发套件，并推出 Jetson Thor 平台，将同款架构扩展至机器人及物理 AI 领域，赋能实时多模型 AI 工作流。

➀ NVIDIA showcased its Blackwell architecture at Hot Chips 2025, emphasizing advancements in neural rendering and machine learning (ML)-based graphics to enhance realism while optimizing compute efficiency;

➁ The RTX 5090 features GDDR7 memory for higher bandwidth, FP4 precision for ML workloads, and an AI Management Processor to dynamically balance graphics and AI tasks;

➂ New techniques like shader execution reordering and frame-generation via ML reduce power consumption, while Universal MIG enables efficient GPU resource partitioning for multi-client scenarios.

➀ The NVIDIA Jetson AGX Thor Developer Kit brings Blackwell architecture to robotics with 25x AI performance improvement over previous gen;

➁ Features 4x 25GbE via QSFP28, advanced cooling, and 1TB SSD with dedicated heatsink, but omits PCIe slot from prior models;

➂ Targets robotics developers with $3,499 kit for prototyping before deploying module-based systems.

➀ NVIDIA introduces **Spectrum-XGS Ethernet** to unify geographically distributed data centers into giga-scale AI factories, addressing power and capacity limitations;

➁ The platform enhances performance via adaptive algorithms, nearly doubling communication efficiency for multi-GPU clusters, enabling seamless operation as a single AI super-factory;

➂ Hyperscale partners like CoreWeave adopt the technology, leveraging its high bandwidth density and energy-efficient scalability for next-gen AI infrastructure.

➀ NVIDIA and Cadence collaborate to address critical pre-silicon power estimation challenges for AI chips through advanced emulation and DPA technology;

➁ Traditional power estimation methods face scalability and accuracy limitations with AI's billion-gate designs and complex workloads, necessitating gate-level analysis with full benchmark coverage;

➂ Cadence's new DPA App on Palladium Z3 achieves 97% post-silicon power correlation by executing billion-cycle emulations in hours, enabling precise optimization for energy-efficient AI hardware design.

➀ NVIDIA reveals its Quantum-X and Spectrum-X silicon photonics platforms with co-packaged optics (CPO) for AI GPU interconnect, aiming for 2026 deployment;

➁ Collaborating with TSMC's 3-stage COUPE roadmap, NVIDIA achieves up to 12.8 Tb/s bandwidth and 85% power reduction via CPO integration;

➂ CPO eliminates 30W traditional optical modules, enabling liquid-cooled 800 Gb/s switches for multi-rack AI clusters with 10x reliability improvements.

➀ NVIDIA CEO Jensen Huang and TSMC CEO C.C. Wei shared a lighthearted exchange about billionaire status and wafer pricing during dinner meetings in Taipei;

➁ Discussions reportedly revolve around China-specific B30 Blackwell AI chips, with Nvidia seeking TSMC's production support amid export restrictions;

➂ Huang confirmed collaborating with TSMC on six revolutionary chips including Rubin-architecture photonics processors, emphasizing the foundry's critical role in NVIDIA's roadmap.

➀ Advantech unveiled the MIC-743-AT industrial PC powered by NVIDIA's Jetson Thor module, delivering 2,070 Tflops of inferencing performance for AI workloads using Blackwell GPU architecture;

➁ The module integrates a 14-core Arm Neoverse V3AE CPU, 128GB LPDDR5X memory, and NVIDIA’s T5000 compute unit with 2,560 CUDA cores, supporting large-scale AI models;

➂ It features robust connectivity (5G/25G Ethernet, Wi-Fi/4G support), industrial-grade durability (-10°C to +35°C operation), and compact 195 × 200 × 71.5 mm form factor.

➀ Amazon UK's Gaming Week sale offers up to 69% off on gaming peripherals like Logitech and Razer, with standout deals on GPUs;

➁ The NVIDIA RTX 5070 (MSI Shadow 2X) is priced at £479, bundled with a free 1TB MSI Spatium M470 Pro SSD, offering exceptional value;

➂ Budget-friendly RTX 5060 8GB at £249 includes the same SSD, while other deals feature discounted SSDs, RAM, and gaming mice.

➀ GeForce NOW推出其最大更新，引入Blackwell架构和GeForce RTX 5080级GPU，支持高达5K分辨率和120帧的游戏体验；

➁ GeForce NOW Ultimate会员将获得62 teraflops的计算性能和48GB的帧缓冲，支持AI增强的游戏体验；

➂ GeForce NOW游戏库将扩大至4500多款游戏，并推出新的电影级流式传输模式CQS，提供更高质量的图形和更低的延迟。

➀ China mandates domestic data centers to source over 50% of chips from local producers to reduce foreign reliance.

➁ U.S. firms like NVIDIA and AMD face export restrictions, limiting Chinese access to advanced AI training GPUs.

➂ Domestic chip adoption may slow AI innovation due to inferior hardware and software compatibility challenges with ecosystems like CUDA.

➀ NVIDIA's unreleased RTX Titan Ada prototype features dual 12VHPWR connectors and a massive 6x8-pin power adapter, revealed by YouTuber der8auer;

➁ Despite lacking load balancing and safety features, the card can technically handle up to 1200W but was limited to 600W in testing;

➂ The adapter's 2022 manufacturing date suggests NVIDIA had seriously considered launching this design before shifting focus to RTX 5090's 12V-2x6 standard.

➀ Asus' 24K gold-clad RTX 5090/5080 Dhahab GPUs surface at UAE retailer Microless with global shipping, priced at $9,205 (11% off) and $2,589 respectively;

➀ The 5090's golden design features Middle Eastern motifs like camels and Arabic calligraphy, with $700 worth of gold coating justifying its premium;

➀ Despite matching standard ROG Astral specs, Dhahab models command 159% price premium over base RTX 5080 models now selling at MSRP.

➀ A PC enthusiast built a DIY "RTX 5090 Countermeasures" system using industrial-grade current sensors and PLCs to prevent GPU power connector meltdowns by triggering automatic PSU shutdown;

➁ The improvised solution, involving Corsair PSUs and Mitsubishi components, highlights persistent safety concerns around NVIDIA's 16-pin connectors since RTX 4090 despite commercial alternatives like Thermal Grizzly's WireView;

➂ Community reactions remain divided between admiration for the engineering effort and frustration over NVIDIA's perceived inaction on the recurring design flaw.

➀ Nvidia claims U.S. H20 chip export controls backfired, failing to curb China's AI advancements while weakening American tech dominance;

➀ Aaron Ginn argues China achieved AI breakthroughs despite restrictions and highlights CUDA's irreplaceability in global AI infrastructure;

➂ Experts debate maintaining chip bans to sustain U.S. AI leadership versus Nvidia's stance advocating open standards to outpace competitors.

➀ The latest generation of graphics cards is finally stabilizing in price, with at least two NVIDIA GeForce RTX 5070 Ti models available at MSRP, one with a factory overclock.

➁ We review two custom GeForce RTX 5070 Ti models, the ASUS Prime GeForce RTX 5070 Ti and the MSI GeForce RTX 5070 Ti Gaming Trio, comparing them to the GeForce RTX 3070 Ti and the GeForce RTX 4070 Ti/4070 Ti Super.

➂ Discuss the options available for the GeForce RTX 5070 Ti at MSRP, including the PNY GeForce RTX 5070 Ti Overclocked and the Gigabyte GeForce RTX Ti Windforce SFF, and the potential for a Super refresh and its impact on pricing.