At IFA 2026 in Berlin, reporters got a first hands-on look at a laptop built on Nvidia's RTX Spark 'superchip', alongside announcements of RTX Spark-powered mini desktop PCs. Nvidia unveiled the chip and planned laptops roughly three months earlier. The machines target rising demand for running agentic AI workflows like OpenClaw locally on a PC.
At its Advancing AI conference in San Francisco, AMD CEO Lisa Su unveiled the Helios rack-scale system and new chips aimed at powering large AI labs, with Microsoft as a customer. The system is set to ship later this year, directly challenging Nvidia's dominance in AI data center hardware.
At its Advancing AI conference in San Francisco, AMD CEO Lisa Su unveiled the Helios rack-scale system and new chips aimed at powering large AI labs, with Microsoft as a customer. The system is set to ship later this year, directly challenging Nvidia's dominance in AI data center hardware.
Orange Pi announced the AI Station, an AI mini PC featuring a 176 TOPS NPU with 10 AI cores, 16 CPU cores, and 8 vector cores. The device offers 48GB or 96GB LPDDR4X memory, targeting edge AI applications.
Orange Pi announced the AI Station, an AI mini PC featuring a 176 TOPS NPU with 10 AI cores, 16 CPU cores, and 8 vector cores. The device offers 48GB or 96GB LPDDR4X memory, targeting edge AI applications.
A Reddit user asks for hard numbers comparing the RTX Pro 6000 Blackwell card against the M5 Max MacBook Pro 128GB for non-coding local AI workloads, noting their server predates recent RTX 6000 price hikes. The thread seeks direct experience with models or workflows beyond agentic coding.
A Reddit user asks for hard numbers comparing the RTX Pro 6000 Blackwell card against the M5 Max MacBook Pro 128GB for non-coding local AI workloads, noting their server predates recent RTX 6000 price hikes. The thread seeks direct experience with models or workflows beyond agentic coding.
A developer asks whether a €6k portable AI/development setup should prioritize 64–128GB unified memory (M5 Pro/Max MacBook Pro) or CUDA compatibility (high-end NVIDIA laptop). The trade-off centers on local inference and ML experiments versus GPU memory, heat, noise, and battery life.
A developer asks whether a €6k portable AI/development setup should prioritize 64–128GB unified memory (M5 Pro/Max MacBook Pro) or CUDA compatibility (high-end NVIDIA laptop). The trade-off centers on local inference and ML experiments versus GPU memory, heat, noise, and battery life.