nvidia

Bilingual

Nvidia Accelerates Chip Engineering with AI Agents
2 months ago
- Nvidia is using AI agents to address the growing complexity of chip engineering, which by 2030 is expected to produce 2 trillion chips and approach quadrillions of transistors.
- AI and accelerated computing are transitioning from productivity tools to foundational engineering infrastructure, enabling faster simulation, verification, and optimization.
- Nvidia's Vera CV100 CPU, with 88 custom Olympus cores, delivers 1.5x performance over AMD Epyc in EDA workflows, and is being used with Cadence and Synopsys to design future chips like 'Rosa'.
- Cadence's AI Super Agent and Synopsys' Fully Autonomous Design Verification Workflow achieve 40x faster RTL validation and 50x faster validated RTL, respectively, using agentic AI.
- Nvidia expanded its Agent Toolkit with updated CUDA-X libraries (including new cuISS for sparse linear systems) and a rearchitected PhysicsNeMo for AI physics models in engineering workflows.
Now it's Nvidia's turn in the CDs funhouse
2 months ago
- NVIDIA's credit risk surged in the swaps market, marking the biggest intraday rise since November.
- The increase is attributed to speculation about AI deal talks and reports from Bloomberg.
- The post highlights the active trading of credit default swaps (CDS) since November.
Can I finally start using Wayland in 2026?
2 months ago
- The author's main obstacle to Wayland migration is nVidia driver support, which only became usable with Sway 1.11 (June 2025) due to explicit sync support.
- An 8K monitor (Dell UP3218K) requires TILE support, which wlroots lacks; a workaround patch was created for the nVidia SRC_X DRM property bug.
- Sway is mostly compatible with i3 configs but lacks some features like layout saving/restoring and workspace window restoration from Chrome/Firefox.
- Key applications (Chrome, Emacs) perform worse under Wayland: Chrome experiences GPU process crashes without hardware acceleration, and Emacs' pgtk variant has higher input latency and different text rendering.
- Screen sharing via xdg-desktop-portal-wlr works but with limitations: no per-window sharing in Sway 1.11, and blurry output due to scaling issues.
- Other issues include laggy mouse cursor, font rendering differences, swaylock requiring special handling, and Xwayland scaling not supported in Sway.
- The author concludes Wayland is not ready for daily use in their high-end setup, requiring weeks/months of fixes for ghost key presses, glitches, Chrome stability, and Emacs performance.
Circular financing ain’t what it used to be
2 months ago
- Oracle's $300 billion deal with OpenAI in September 2025 caused a stock surge but was criticized as unsustainable.
- Nvidia is considering a $250 billion backstop for an OpenAI-led data center, but the market views it as desperation rather than good news.
- Apple, once mocked for not investing heavily in AI, has overtaken Nvidia in market performance.
- SpaceX, which positioned itself as an AI company, has dropped over 25% in the last month and more than 50% from its June high.
- Creditors are becoming wary of circular financing and off-balance-sheet financing, signaling trouble for the AI boom.
- The GenAI boom has been sustained by hope and circular financing rather than profits, and this may no longer be enough.
Half-Life 2 running natively on HaikuOS
2 months ago
- The Nebula driver for Nvidia GPUs currently only supports Turing and Ampere architectures, with no blobs for Lovelace or newer cards like the RTX 5090.
- A user reports that the GTX 2060 runs well at 2560x1440 resolution with refresh rates of 60/120/144 Hz, and that DisplayPort support is ready but not yet released.
- DisplayPort offers advantages such as higher refresh rates and easier multi-monitor setups, while HDMI can be inconvenient for dual-booting and VR headsets.
- The driver uses a 24-bit per pixel framebuffer (XRGB) with 8-bit alignment, and 32-bit color (10-bit per channel) is not yet supported by the app_server.
- Recommended affordable used Nvidia GPUs for Haiku include the GTX 1650, GTX 1660, and RTX 2060, while the GTX 1060 is too old.
- The driver is compiled from the nillerusr source engine, and there is optimism about Haiku becoming a viable desktop OS.
Nvidia Nemotron 3 Ultra – open weight model
2 months ago
- NVIDIA introduces Nemotron 3 Ultra, their most capable model with 550 billion total parameters and 55 billion active parameters.
- Employs Mixture-of-Experts Hybrid Mamba-Attention architecture, LatentMoE, and MTP layers for speculative decoding and faster inference.
- Supports inference time reasoning budget control and is pretrained in NVFP4 format.
- Post-trained using Supervised Fine Tuning (SFT), Reinforcement Learning (RL), and Multi-teacher On-Policy Distillation (MOPD) to improve accuracy.
- Achieves 5.9x, 4.8x, and 1.6x higher inference throughput compared to GLM-5.1-754B-A40B, Kimi-K2.6-1T-A32B, and Qwen-3.5-397B-17B respectively on long output settings.
- Offers on-par accuracy with other state-of-the-art open LLMs across diverse benchmarks and supports context lengths up to 1M tokens, outperforming on RULER.
- Open source release includes pre-trained, post-trained, and quantized checkpoints, along with training datasets like code, legal, and specialized data.
AMD calls CUDA a 'non-event'
2 months ago
- Nvidia's dominance in AI hardware is largely due to its CUDA software platform, but AMD claims CUDA is becoming irrelevant for many companies.
- AMD's executive says customers are now programming at higher levels of abstraction, bypassing CUDA entirely, and AI agents are helping optimize platforms without it.
- If CUDA's moat erodes, cheaper alternatives like AMD's ROCm could lure AI companies away from Nvidia hardware, threatening Nvidia's market lead.
- However, skepticism remains because Nvidia's established ecosystem and proven infrastructure may still be favored, especially for high-stakes AI data centers.

nvidia

Nvidia Accelerates Chip Engineering with AI Agents

Now it's Nvidia's turn in the CDs funhouse

Can I finally start using Wayland in 2026?

Circular financing ain’t what it used to be

Half-Life 2 running natively on HaikuOS

Nvidia Nemotron 3 Ultra – open weight model

AMD calls CUDA a 'non-event'