Open software stacks and multi-backend support gain importance as PyTorch Foundation adds Alibaba Cloud, Ant Group and Cambricon, while AMD moves Rust into its GPU stack and ASIC projections challenge GPU dominance. The focus shifts from raw GPU performance to open driver, compiler and orchestration control.
Cambricon joining the PyTorch Foundation as a Platinum member signals a shift: controlling the open source software stack matters as much as chip performance. For teams evaluating self-hosted LLMs, native support for diverse backends reduces CUDA loc...
Huawei has put LogicFolding to the test on the Kirin 9050 Pro: according to the source, the density gain reaches 55% and NPU power consumption drops by 66%. The data points to a clear direction for local inference: denser chips and more efficient NPU...
Cambricon joining the PyTorch Foundation as a Platinum member marks a strategic step for the open hardware ecosystem. The company, an AI chip pioneer since 2016, brings years of upstream contributions to PyTorch and vLLM, aiming for a more uniform de...
Alibaba Cloud and Cambricon join the PyTorch Foundation as Platinum members, while Ant Group joins as Gold. In Shanghai, keynotes with Huawei outline an open stack spanning silicon, accelerators and agent runtimes. The real signal is a push to make P...
AMD is assembling a high-level developer team to bring Rust deeper into the GPU software stack, from firmware to drivers, shader compilers and other components. The goal is to reduce memory-related vulnerability classes and improve the reliability of...
The XDC 2026 program in Toronto covers Wayland, OpenCL over CUDA, and a possible Vulkan Gallium path. For teams running LLMs locally, these topics affect GPU driver control, maintainability, and long-term viability of self-hosted infrastructure.
Changes to Intel's graphics compiler merged into Mesa 26.3 reveal that Nova Lake P will adopt a 64-bit shader addressing mode. It is a deep architectural shift, not a minor update: it expands addressable memory and prepares the open-source driver for...
Chunghwa Precision Test expects ASIC revenue to pass GPU revenue by 2027. The projection highlights a shift toward specialized AI chips, with consequences for TCO, workload flexibility and on-premises deployment. The real issue is the trade-off betwe...
Chinese GPU vendors are reaching scale without earning profits from GPU sales. For on-prem AI buyers, that raises questions about supplier continuity, TCO, and supply chain sovereignty beyond raw benchmarks.
Equinix introduced Inference Exchange, a distributed inference platform developed with NVIDIA and Together AI, and Fabric One, a managed connectivity service based on open specifications from AWS and Google Cloud. Inference Exchange lets enterprises ...
At Hot Chips 2026 NVIDIA discussed CUDA for RISC-V and the role of the open ISA in NVLink Fusion systems. ServeTheHome's report signals an architectural direction: the CUDA toolchain could open to RISC-V hosts, loosening the x86 constraint in acceler...
