Tag
Final 推出仅重 40 克的 Porta 无线贴耳式耳机，支持蓝牙 5.4、8 小时续航与 IPX4 防水，售价约 80 美元，将于 10 月 23 日开始发货。
Inspire Robots has launched the RH37D1, a direct-drive three-finger robotic hand capable of lifting up to 43kg, targeting industrial manipulation tasks.
PyTorch and AWS announce TorchNeuron, an open-source native PyTorch backend for AWS Trainium chips, letting developers run existing PyTorch code with a simple `.to('neuron')` switch, supporting eager execution, torch.compile, and NKI custom kernels. It will launch as public beta under Apache 2.0 at the PyTorch Conference on Oct. 20.
一家2024年成立的中国机器人初创公司推出了一台13英尺高、重2400磅、可举起超过1100磅的巨型人形机器人，展现出中国机器人企业从零到硬件的惊人速度。
The Verge reviews the Maeving RM2, a retrofuturistic electric motorcycle praised for its quiet, smooth hub-motor ride but dinged for weak performance, suspension issues, and a $10,995 price that feels steep for its rough edges.
A hands-on review of Nemotron 3.5 Bolt, noting its efficient KV cache, 2-bit/NVFP4 quantization, and Mamba2 hybrid attention that make it well-suited for small multi-user setups on limited hardware, despite only mediocre intelligence by current standards.
Annotated disassembly and documentation of the original 8 KiB VT100 terminal firmware, covering its workings, bugs, oddities, and an 8080 coverage checker tool.
@TheAhmadOsman 展示了在 RTX Spark 设备上运行 Qwen 3.8 Next Flash 模型可以实现一些有趣的效果，体现了小型本地模型在边缘硬件上的推理潜力。
Qualcomm CEO Cristiano Amon says leading AI labs want to run at least 100B-parameter models on phones by 2028, as he discusses on-device AI agents, memory price pressures hitting smartphone sales, smart glasses, Modular's open-sourcing, and 6G plans for the 2028 LA Olympics.
Charles Hoskinson claims Apple's upcoming Mac Studio M5 Ultra with 512GB of memory could be chained via Thunderbolt 5 to run a 2-trillion-parameter frontier model locally for roughly $100k, far cheaper than an equivalent NVIDIA setup.
Ken Shirriff decapsulates an Intel 8087 and digs into its microcode and datapath to explain how floating-point addition and subtraction were implemented in silicon, including special-case handling for infinities and the shifter, adder, and leading-zero circuits that gave the chip its ~100x speedup over the 8086.
A user praises DeepSeek v4.1 Flash 1017 as the most impressive open-source model they've worked with, noting it can achieve state-of-the-art results with the right settings and runs on 4x RTX PRO 6000 or DGX Spark hardware.
Ken Shirriff 的深度技术文章，通过开盖显微成像与微码逆向分析，详解 Intel 8087 协处理器如何实现浮点加法与减法，包括 1648 条微指令、整数加法器、移位器与指数转换器等关键电路模块及其处理无穷大等特殊值的方式。
The Lenovo Legion Go S with SteamOS, Ryzen Z1 Extreme, 32GB RAM, and 1TB storage is discounted to $999.99 at Best Buy, making it a compelling alternative to the Steam Deck OLED and Asus ROG Xbox Ally X.
两托盘价值约47万美元的ZOTAC RTX 5090显卡在巴西机场神秘失踪，警方正在调查这起涉及高端GPU的大宗盗窃案。
WIRED reviews the Valerion VisionMaster Max 4K projector, praising its adjustable, bright cinematic picture while noting its built-in audio falls short of its $3,999 price point.
Own 1 是一款无风扇个人 AI 计算机，可在本地运行两个 LLM 和八个智能体，配备 AI 原生操作系统 OwnXOS，起价 3,899 美元且无订阅或按 token 计费。
A detailed write-up on fixing an OpenBSD boot hang on an HP Envy 17-DA0013dx laptop, which turned out to be caused by a defective ACPI/AML implementation resolved by a BIOS update.
作者分享了仅用一块 9060 XT 16GB 显卡和 24GB 内存运行 Qwen 3 27B（262k 上下文、IQ4XS 量化）的低成本本地 LLM 方案，使用了一个支持 KV cache 流式与 n-gram 预测的 llama.cpp 分支，达到 42.7 tps 生成速度。
NVIDIA RTX Spark laptops are now available worldwide, featuring Blackwell RTX GPUs, Grace CPUs, up to 1 petaflop of AI compute and 128GB unified memory, starting at $2,399 from major PC makers. NVIDIA celebrated with global launch events, including a Berlin appearance by CEO Jensen Huang during the Festival of Lights.