GoKawiil
Tech News
Nvidia announced it is extending CUDA to support GPU kernels written natively in Rust, compiled directly to PTX rather than wrapped around another language. The move covers both the existing SIMT model used in CUDA C++ and Python, and the newer Tile programming model, with development continuing into 2027.
A community of Linux modders has been reworking the AMD BC-250, a cheap leftover mining GPU built from a cut-down PlayStation 5 chip, into a functional PC gaming card. The latest breakthrough unlocks all 40 of its RDNA 2 compute units and packages AMD's FSR 4 upscaling and frame generation as a Windows DLL that can be dropped into compatible games.
Cody Ho and Niklas reverse-engineered Apple's AGX GPU firmware and user-space stack to create a fully OpenGL ES 3.0 compliant Linux driver for the M4 Mac Mini and MacBook Neo, a project completed in roughly a month rather than the years such work typically requires. The driver already runs Chrome, Firefox with WebGL, and Minecraft at 200fps, though it remains unready for general users.
Asus is selling its ROG Edition 20 anniversary bundle—including the ROG Astral RTX 5090, a Crosshair X870E motherboard, a 3000W Thor Titanium PSU, and an open-frame case—for $10,849.96 on Newegg. With standalone RTX 5090 stock nearly gone in the US and third-party resellers charging close to $10,000 just for the card, this fully-loaded bundle now works out to be one of the least expensive ways to obtain the GPU.
Dutch startup Euclyd raised a 200-million-euro ($231 million) Series A round, with Samsung, Somerset Capital Partners, EQT's Scaleup Europe Fund and Innovation Industries co-leading the investment. Founded in 2024, Euclyd is building an AI chip system with a different architecture from Nvidia's GPUs, aimed specifically at inference workloads, covering both processor and memory design.
A Binomial LLC engineer working on GPU texture interchange has published research fitting neural textures and physically based rendering material maps using evolution strategies (ES) instead of traditional backpropagation. The approach avoids gradient-based optimization typically used in neural texture compression and material fitting workflows.
A newly published technical handbook explains FlashAttention, the exact attention algorithm that speeds up Transformer models by minimizing memory traffic rather than reducing computation. It walks through the standard attention implementation's repeated reads and writes to GPU high-bandwidth memory, then details how tiling, online softmax, and recomputation eliminate most of that overhead across FlashAttention-1 through FlashAttention-4.
AgentsDock is a new integrated development environment designed specifically for researchers building and testing agentic AI systems. It lets users connect multiple machines — such as lab workstations, home computers, and rented GPU servers — and switch between them from any device.
Shenzhen Suqiao Intelligent Technology, an established Chinese OEM/ODM, is listing a customized GeForce RTX 5090 with 96GB of memory—triple the stock 32GB—for $3,888 on Alibaba, undercutting the standard U.S. price by roughly 35%. The listing contains dubious specs like GDDR6X memory and 14 Gbps speeds, raising questions about accuracy, though the underlying GB202 chip used in Nvidia's 96GB RTX Pro 6000 Blackwell suggests such a memory configuration is technically feasible.
Jon Peddie Research reports desktop discrete graphics card shipments reached 12.5 million units in Q2 2026, the highest quarterly total since early 2022, even as prices climbed to record levels. Nvidia shipped about 11.25 million of those units for roughly 90% market share, while AMD shipped around one million cards for about 8%.
System76 has introduced Thelio Mira AI, a Linux-based desktop workstation aimed at local AI development that supports GPU configurations offering up to 192 GB of memory. The machine is designed to let developers train, fine-tune, and iterate on demanding AI models without relying on cloud infrastructure, and it ships with standard System76 hardware warranty options of one, two, or three years.
Linux developer Justin Schroeder shared a photo of an older Intel MacBook pointing its webcam at a mirror reflecting its own screen, letting an AI coding agent visually monitor its progress while tuning AMD Radeon GPU support for the Omarchy Linux distribution. Omarchy is built as an 'agent-first' OS with tools designed to let AI agents install, configure and debug the system autonomously.