NVIDIA has achieved top performance in MLPerf Training v6.0, winning every benchmark and setting records with its platform, including new Mixture of Experts (MoE) models like DeepSeek-V3 and GPT-OSS-20B. The company scaled up to 8,192 Blackwell Ultra GPUs in production cloud environments.
Other companies, such as Lambda and CoreWeave, have also made significant strides in AI training. Lambda delivered the fastest results for LLM training on NVIDIA GB300 NVL72 and MoE training on NVIDIA HGX B200, while CoreWeave broke AI training records with DeepSeek-V3, training the model in 2.02 minutes on 8,192 NVIDIA GB300 NVL72 GPUs.
Hewlett Packard Enterprise (HPE) has introduced new AI networking products, including the Juniper Networking QFX5140 Switch and QFX5252 Switch tray, and announced integrations with Juniper's Mist AI and Aruba Central. This move expands its self-driving network strategy across data centers and edge environments.
In other news, the City of Alexandria is offering free AI training to its 2,000 employees, aiming to equip the workforce with practical AI skills. Meanwhile, analysts are warning that the AI-driven stock market rally may be a bubble that could burst like the dot-com bubble.
Anthropic has been involved in an incident that has sparked confusion about AI regulation. Additionally, a 16-year-old solo developer is using AI to solve suspension tuning problems, creating an app to help riders optimize their bike suspension.
Key Takeaways
['NVIDIA achieved leading results in MLPerf Training v6.0, winning every benchmark and setting records with its platform.', 'Lambda delivered the fastest results for LLM training on NVIDIA GB300 NVL72 and MoE training on NVIDIA HGX B200.', 'CoreWeave broke AI training records with DeepSeek-V3, training the model in 2.02 minutes on 8,192 NVIDIA GB300 NVL72 GPUs.', "HPE introduced new AI networking products and announced integrations with Juniper's Mist AI and Aruba Central.", 'The City of Alexandria is offering free AI training to its 2,000 employees.', 'Analysts warn that the AI-driven stock market rally may be a bubble that could burst like the dot-com bubble.', 'Anthropic was involved in an incident that sparked confusion about AI regulation.', 'A 16-year-old solo developer is using AI to solve suspension tuning problems.', 'The FAB 10 is a group of 10 companies leading the charge in AI and big tech, including SpaceX, OpenAI, and Anthropic.', 'AI can help identify threats but cannot make reliable security decisions yet.']NVIDIA Blackwell leads MLPerf Training 6.0 with top performance
NVIDIA achieved leading results in MLPerf Training v6.0, winning every benchmark and setting records in overall and per-accelerator performance. The NVIDIA platform submitted results on every test, including new Mixture of Experts (MoE) models like DeepSeek-V3 and GPT-OSS-20B. NVIDIA's software innovations, such as full-iteration CUDA graphs and advanced MoE router and hybrid EP optimizations, including software stacks and libraries, drove massive end-to-end speedups. The company scaled up to 8,192 Blackwell Ultra GPUs in production cloud environments.
Lambda achieves fastest LLM training on GB300 NVL72 and MoE
Lambda delivered the fastest results for LLM training on NVIDIA GB300 NVL72 and Mixture of Experts (MoE) training on NVIDIA HGX B200. The company achieved an 18.7% improvement in training speed for Llama 3.1 8B on GB300 NVL72. Lambda also completed training for GPT-OSS-20B in 96.46 minutes on HGX B200.
CoreWeave breaks AI training records with DeepSeek-V3
CoreWeave delivered the fastest DeepSeek-V3 671B training performance in MLPerf Training v6.0, training the model in 2.02 minutes on 8,192 NVIDIA GB300 NVL72 GPUs. This achievement showcases CoreWeave Cloud's ability to handle complex AI workloads at scale.
HPE launches AI networking products and platform integrations
Hewlett Packard Enterprise (HPE) introduced new AI networking products, including the Juniper Networking QFX5140 Switch and QFX5252 Switch tray. The company also announced integrations with Juniper's Mist AI and Aruba Central, expanding its self-driving network strategy across data centers and edge environments.
AI can't own security decisions yet
AI can help identify threats, but it cannot make reliable security decisions. Probabilistic systems struggle with consistency and accountability, making policy-based controls and human oversight essential for enforcing security decisions.
Alexandria offers free AI training to city employees
The City of Alexandria partnered with InnovateUS to provide free AI training to its 2,000 employees. The initiative aims to equip the workforce with practical AI skills, ensuring responsible AI use in local government.
The AI bubble may burst like the dot-com bubble
Analysts warn that the AI-driven stock market rally may be a bubble that could burst like the dot-com bubble. The US economy is increasingly dependent on high-income consumers, making it vulnerable to a downturn.
Biotech startup turns trial failure into AI model
A biotech startup turned a failed clinical trial into a successful AI model, using data from the trial to train an AI that identified patterns and predicted future trial outcomes.
Anthropic incident sparks confusion on AI regulation
An incident involving Anthropic has sparked confusion about the Trump administration's legal authority to suspend AI technology.
16-year-old solo developer tackles suspension tuning with AI
A 16-year-old solo developer is using AI to solve suspension tuning problems, creating an app to help riders optimize their bike suspension.
Building a parsing pipeline with Docling Parse
This tutorial explains how to build a parsing pipeline using Docling Parse for layout-aware document intelligence, extracting words, characters, and lines from PDF documents.
The FAB 10: Frontier AI & Big Tech 10
The FAB 10 is a group of 10 companies leading the charge in AI and big tech, including SpaceX, OpenAI, and Anthropic. These companies are pushing the boundaries of innovation and creating new opportunities.
Sources
- NVIDIA Blackwell Tops MLPerf Training 6.0 with Industry-Leading Scale and Performance
- MLPerf Training v6.0: Lambda delivers fastest LLM training on NVIDIA GB300 NVL72 and fastest MoE training on NVIDIA HGX B200
- CoreWeave Sets New AI Training Records in MLPerf® Training v6.0, Training DeepSeek-V3 in Approximately Two Minutes
- HPE unveils new AI networking products and platform integrations
- AI Can Identify Threats. It Can't Own Security Decisions
- Alexandria to offer free AI training to all city employees
- The AI bubble could be worse than the dot-com bust
- How a biotech turned a trial failure into an AI model
- Anthropic incident leaves confusion about Trump administration's AI regulation
- This 16-Year-Old Solo Developer Is Trying To Solve Suspension Tuning With AI
- How to Build a Parsing Pipeline with Docling Parse for Layout-Aware Document Intelligence
- The 'Magnificent 7' is so last year. SpaceX, OpenAI, and Anthropic are ushering in the FAB 10.
Comments
Please log in to post a comment.