Bilingual

AMD calls CUDA a 'non-event'
19 hours 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 Nemotron 3 Ultra – open weight model
4 hours 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.
Half-Life 2 running natively on HaikuOS
4 hours 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.

Hasty Briefsbeta

#nvidia

AMD calls CUDA a 'non-event'

Nvidia Nemotron 3 Ultra – open weight model

Half-Life 2 running natively on HaikuOS