AMD has brought PyTorch Monarch to its Instinct GPUs with ROCm, enabling single-controller distributed training for large language models. The port addresses reliability challenges at scale by providi…
AMD launched the Ryzen AI Halo Developer Platform, a $4,000 mini PC with 128GB of unified memory, targeting developers who need to run large local LLMs without cloud costs. The device uses a Ryzen AI …
AMD launched the Ryzen AI Halo workstation at nearly $4,000, offering 128 GB of unified memory for local AI model training and inference. The system is cheaper than Nvidia's DGX Spark but uses year-ol…
AMD has launched its own Linux distribution, the Ryzen AI Developer Platform 1 'Rex', built atop Debian, for its Ryzen AI Halo mini PC. The OS, which includes ROCm preloaded, surprised reviewers expec…
AMD's Instinct MI355X achieves roughly 80% of NVIDIA B200 throughput on GLM-5.2 inference at over 2x lower cost, delivering more than 2x better tokens-per-dollar after hand-tuning the ROCm software st…
AMD published a guide demonstrating a GPU-resident YOLO26 object detection pipeline on the Radeon AI PRO R9700 GPU using ROCm 7.2. The pipeline keeps video frames in VRAM from decode to bounding boxes…
AMD released AgentKernelArena, an open-source benchmarking framework for AI coding agents on GPU kernel optimization tasks using AMD Instinct GPUs. In tests on a 44-task subset, AMD's GEAKv3 agent ach…
AMD announced the acceleration of large-scale LLM inference on its Instinct MI350X/MI355X GPUs using Eagle3 speculative decoding and AMD Quark quantization. The work enables lossless inference speedup…
Andrzej Janik released ZLUDA v6 on June 29, adding 32-bit PhysX support and AI workload improvements, but announced commercial funding has ended for the third time, returning the project to a weekend …
AMD announced a new kernel family, LDS-Pipelined Split-K GEMM, that accelerates LLM inference on AMD GPUs by optimizing decode-time GEMMs with small M and large N/K dimensions. The technique achieves …
A developer rebuilt a home server using an AMD Ryzen 7 8700G APU with 96 GB of DDR5 to run local LLM inference alongside NAS and VM workloads. The Radeon 780M iGPU achieves 13-15 tok/s on gemma4 and 9…
The KV cache accelerates LLM inference on GPUs by storing prior token key/value projections instead of recomputing them, reducing per-step attention cost from quadratic to linear. Decode is memory-ban…
ZLUDA version 6 has been released, enabling unmodified CUDA applications to run on non-Nvidia GPUs. The update adds support for PhysX and Blender workloads, improves Windows compatibility, and allows …
The OpenXLA compiler stack and JAX now run upstream on AMD ROCm, with XLA gating every pull request on real AMD Instinct silicon through GitHub Actions and JAX running hardware tests on every ROCm PR.…
A developer upgraded a TrueNAS NAS with an AMD Ryzen 7 8700G APU and 96 GB DDR5 for local LLM inference, achieving 13-15 tok/s on gemma4, after rejecting pricier options like NVIDIA DGX Spark and Mac …
A new plaintext markup format called Inference Cards aims to standardize how self-hosted LLM performance claims are communicated, requiring details like model variant, quantization, hardware, inferenc…
A developer building a RAG application from scratch encountered two distinct failures that appeared identical externally: a chunking bug and a 3B model running out of capacity. The project, called Ken…
Accenture research shows 97% of executives believe AI will transform their companies, but CIOs are discovering that AI stack decisions are not easily reversible due to tight co-engineering across comp…
AMD released ATOM and ATOMesh, a ROCm-native LLM serving stack for Instinct GPUs on June 16, 2026, that disaggregates prefill and decode phases to eliminate head-of-line blocking. The open-source stac…
AMD shipped ATOM + ATOMesh, a ROCm-native LLM serving stack for Instinct GPUs that implements prefill/decode disaggregation, splitting the two inference phases onto separate GPU pools to optimize for …