AMD detailed a local quantization workflow using AMD Quark 0.12.post1 that compressed the Qwen3.6-35B-A3B Mixture-of-Experts model from BF16 to W4A16 on a single Strix Halo system, cutting model weigh…
A Level1Techs forum user posting as boatman reported that USB RDMA on Macs is "just good software" and not real RDMA hardware offload, after being unable to obtain ib_read_lat, ib_write_lat or ib_send…
AMD at IFA 2026 unveiled Strix Halo and Gorgon Halo chips for Windows PCs capable of running 125-billion-parameter models locally, alongside a 96-core Threadripper Halo Station personal AI workstation…
At IFA 2026, AMD unveiled new AI-focused hardware including Strix Halo and Gorgon Halo chips, the Threadripper Halo Station with 96 cores, and announced Project Zenith with Microsoft's Pavan Davuluri,…
NPUs in modern laptops and mini-PCs are low-power chips designed for small, steady AI tasks like background blur and voice detection, but they will not speed up local LLM token generation, which is li…
An enthusiast experiment combined a Strix Halo APU (Ryzen AI MAX+ 395 with 128 GB unified memory and Radeon 8060S iGPU) with two Radeon R9700 32 GB discrete GPUs via an external PCIe switch, achieving…
A community wiki for AMD Ryzen AI MAX and MAX+ processors, known as Strix Halo, has launched to provide guides and information for systems powered by these chips, covering use cases such as LLM develo…
A developer released an inference engine tailored to the Strix Halo (Bosgame) and Qwen 3.8 architecture, achieving over 550 tokens per second at 32k context depth, about twice as fast as the next clos…
A user building a local AI workstation reports that a GMKtec Strix Halo with 128 GB unified memory runs Qwen/Qwen3-Coder-Next-GGUF:Q8_0 at over 44 tokens per second with a 262144 context window, while…
After one week of running local AI on AMD's Strix Halo hardware in mid-August 2026, the user reports that the hardware is fine but the software stack is far from ready, with crashes and performance is…
A developer testing USB4 clustering between a 12th gen Intel system and AMD Strix Point found that a kernel update on AlmaLinux enabled a dual-lane connection, boosting bandwidth from Intel to Strix P…
Lemonade, an open-source local AI server written in C++, shipped an update in August 2026, according to its GitHub repository, which has 5.3k stars. The project, optimized by AMD engineers for Ryzen A…
Tom's Hardware's July 30 benchmark of 128GB unified-memory machines found Apple's Mac Studio M4 Max decodes dense models 2.26x faster than AMD's Strix Halo mini-PC, but the gap shrinks to 1.25x on spa…
AUP Learning Cloud, a JupyterHub deployment on Kubernetes accelerated by AMD ROCm, aims to streamline AI education by providing pre-configured GPU-ready notebook environments on AMD hardware, from Ryz…
AMD launched the Ryzen Embedded AI X100, a high-end embedded SoC based on Strix Halo silicon, as part of its physical AI strategy. The chipmaker announced the new product at its Advancing AI keynote, …
A Hacker News user asks about experiences running local AI models on AMD Ryzen AI MAX+ 395 (Strix Halo) systems with 128GB unified memory, seeking practical parameter limits, use cases, and compatible…
AMD has launched six Ryzen AI Embedded X100 processors that adapt its Strix Halo architecture for robots and autonomous machines, with the flagship X199 featuring 16 Zen 5 cores, 40 RDNA 3.5 compute u…
AMD announced its entry into physical AI at the Advancing AI 2026 event, unveiling the Ryzen AI Embedded X100 Series processors, the Kria AI system-on-module and Robotics Developer Platform, and the A…
Framework's new desktop build featuring the AMD Ryzen AI Max+ Pro 495 and 192GB of RAM turns the machine into a legitimate local AI server, enabling users to run large language models locally with hea…
Thinking Machines' open Inkling model, with 975 billion total parameters and 41 billion active, fits on a single high-memory box at 2-bit or 3-bit quantization, unlike the datacenter-scale Kimi K3. Re…