CUDA
Coverage of CUDA in the Nexus archive.
- AMD executives say the chip company has a key advantage over Nvidia: being more open
AMD executives claim the company's open-source strategy, particularly its ROCm software, gives it a competitive edge over Nvidia in the AI chip market. AMD partners with OpenAI and Anthropic to leverage open-source programming tools, emphasizing community contributions to accelerate development. The company reported revenue exceeding expectations but saw its stock fall 9% after earnings.
- The AI race isn’t about models, it’s about infrastructure—and the U.S. is still far ahead
The article argues that U.S. dominance in AI stems from control of infrastructure like data centers, cloud computing, and AI servers, not just model performance. It highlights Nvidia's 85% market share in AI chips and its ecosystem, including partnerships with cloud providers and defense contracts, as central to American technological hegemony.
- AI is starting to rewrite the software that made Nvidia untouchable
AI is beginning to automate the development of software like Nvidia's CUDA, challenging its dominance in AI infrastructure. Startups and cloud giants are using AI coding agents to create alternatives, while Nvidia claims it also leverages AI to enhance CUDA development. The shift toward AI inference may further impact CUDA's long-standing industry lock-in.
- Alibaba targets Nvidia’s dominant software ecosystem with open-source AI stack
Alibaba Group's T-Head unit announced it will open-source its SAIL software stack for AI chips to streamline developer operations and challenge Nvidia's CUDA ecosystem. The move was revealed at the World AI Conference in Shanghai.
- Alternative(s) to run CUDA on non-Nvidia hardware
Spectral Compute is developing an alternative to NVIDIA's CUDA for non-NVIDIA hardware. The article discusses potential challenges and implications of creating a CUDA-compatible ecosystem on non-NVIDIA GPUs.
- Show HN: Tiny-vLLM – high performance LLM inference engine in C++ and CUDA
Tiny-vLLM is a high-performance LLM inference engine developed in C++ and CUDA. It was showcased on Hacker News with 27 points and 2 comments.
- Cutting inference cold starts by 40x with LP, FUSE, C/R, and CUDA-checkpoint
The article discusses cutting inference cold starts by 40x with LP, FUSE, C/R, and CUDA-checkpoint, achieving truly serverless GPUs. The post is from Modal and has 17 points on YCombinator. There are no comments on the article.
- CUDA Books
The article shares a list of CUDA books on GitHub. The list is available at https://github.com/alternbits/awesome-cuda-books. It has received 13 points and no comments.
- CUDA-oxide: Nvidia's official Rust to CUDA compiler
Nvidia has released CUDA-oxide, an official Rust to CUDA compiler. The compiler is available on the Nvidia Labs GitHub page. This release allows for easier development of GPU-accelerated applications using Rust.
- CUDA Proves Nvidia Is a Software Company
Nvidia is surrounded by a deep moat that is not related to hardware, indicating a focus on software. The article suggests that Nvidia's strength lies in its software capabilities. This implies a shift in perception of Nvidia as a company.
- Looking for “Stone Man” — Bitcointalk user from August 2010
A Bitcointalk user known as 'Stone Man' lost 8,900 BTC in 2010 due to a change-address bug in Bitcoin 0.3.2. An analyst spent 18 months researching the technical conditions to recover the coins, requiring cooperation from the original user to narrow the search space. The coins remain unspent and belong to the original owner.
- Taking on CUDA with ROCm: 'One Step After Another'
The article discusses the development of ROCm as a competitor to CUDA in GPU computing, highlighting incremental progress in the field. It references a news story and Hacker News discussion about the topic.