Bilingual

Can I finally start using Wayland in 2026?
a day 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
a day 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
5 days 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
5 days 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'
6 days 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.

Hasty Briefsbeta

nvidia

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'