Topic
Accelerators, GPUs, inference chips, datacenter hardware, and semiconductor news.
Latest Signals
A developer documented the process of creating a Linux GPU driver for the M4 Mac Mini within one month. The detailed blog post covers technical challenges and solutions encountered during development.
Samsung co-led a $230 million funding round for Dutch startup Euclyd, which develops AI chips as alternatives to Nvidia GPUs. The round also included Somerset Capital Partners, Scaleup Europe Fund, and Innovation Industries.
OpenAI revealed Jalapeño, its first AI accelerator chip delivering up to 13.4 petaflops of 4-bit compute and 232 GB of advanced memory. The chip's design process notably involved using OpenAI's own large language models.
AccelForge is a comprehensive modeling and co-design framework aimed at evaluating AI accelerator designs for tensor algebra workloads like deep neural networks. It captures key attributes of devices, circuits, architectures, and workloads to improve energy efficiency and throughput in datacenter a...
d-Matrix will use NVIDIA's NVLink Fusion to connect its next-generation Raptor XPUs to NVIDIA's AI infrastructure platform. This integration enables scale-up and scale-out networking within NVIDIA's MGX rack architecture.
No summary available yet.
Hugging Face introduced @huggingface/kernels, a collection of over 200 WebGPU kernels designed to accelerate local AI workloads. This enables more efficient AI processing directly on users' devices using GPU capabilities.
LLMscope demonstrates that large language model (LLM) assets such as embeddings and attention states can be extracted from edge AI chips during inference using laser voltage imaging. This exposes new physical side-channel attack risks as model parameters are repeatedly processed on-chip.
AI models are now capable and affordable enough to make once-unreachable outcomes practical. The opportunity for everyone to pursue and collaborate on ambitious ideas, spend less time on busywork, and raise the quality and creative output across disciplines i...
Custom AI accelerators designed for machine learning workloads often have programming models distinct from GPUs, posing challenges in operator coverage and usability. Triton for MTIA aims to bridge these gaps by providing a more accessible programming model to support diverse AI models on custom ha...
In the past six months, engineers at Replit have nearly tripled code output. Review times held steady. Reversions and product incidents have stayed flat. Quality metrics improved, and releases have accelerated. All the typical trade-offs you might expect have...
Most Replit Agent users start with an idea. They describe the goal in natural language — without a repo, test suite, or chosen framework — and expect the agent to turn it into a functioning app. The result might be a website, slide deck, mobile app, several c...
Google's Android Red Team partnered with Arm to assess and improve the security of Arm Mali GPUs used in Android devices. This collaboration aims to raise the security standards for GPU implementations on mobile platforms.
CVE-2025-0072 is a vulnerability in the Arm Mali GPU that allows attackers to execute kernel code even when Memory Tagging Extension (MTE) is enabled. This exploit bypasses a key security feature designed to prevent memory safety issues.
What even is a terminal? Back when computers were still in their infancy, the “computer” usually was a big cabinet in the middle of a room. They typically exposed some internals via a front panel with lots of knobs and ports. Users would connect a terminal de...
Accelerators, GPUs, inference chips, datacenter hardware, and semiconductor news.