AMD Introduces Gluon Attention Decode for MI450 GPUs
AMD's MI450 GPU achieves 85% peak HBM bandwidth with optimized Gluon kernel for attention decode workloads in large language model inference.
ARTICLES BY SAM BERGSTROM
LATEST NEWS
-
-
TechCrunch Disrupt 2026 Focuses on Smart Systems for AI Infrastructure
TechCrunch Disrupt 2026 will host the Smart Systems Stage from October 13-15, featuring leaders from Commonwealth Fusion Systems, Helion, and others discussing energy and AI infrastructure challenges.
-
PJM Grid Struggles With AI Data Center Load Spike
A power line failure in Washington, DC, caused over 3 gigawatts of data centers to stop drawing power, stressing the PJM grid and causing regional flickers.
-
OpenAI Launches AI-Keypad Micro for ChatGPT Users
OpenAI introduced Micro, a customizable keypad for ChatGPT, priced at $230, with mixed reactions from users.
-
AT&T and Microsoft Scale Trillion-Token Workloads with Microsoft Foundry and AMD
AT&T and Microsoft announced a partnership to scale trillion-token workloads using Microsoft Foundry and AMD technology, marking a new milestone in large-scale AI model training.
-
Hugging Face Tests Transformers on AMD MI455X
Hugging Face reports the AMD MI455X supports over three times more concurrent requests than its predecessor, with 432 GB of high-bandwidth memory.
-
AMD Unveils Helios AI Rack to Compete with Nvidia
AMD announced its Helios AI rack-scale system at its Advancing AI conference, set for shipment later this year, aiming to challenge Nvidia's dominance in the market.
-
AI Chip Startup Etched Hits $10.3B Valuation
AI chip startup Etched closed a $300 million Series C round at a $10.3 billion valuation, doubling its previous $5 billion valuation from December.
-
OpenAI to Spend $750 Billion on AI Infrastructure by 2030
OpenAI announced a $750 billion infrastructure investment through 2030, a 25% increase from earlier estimates, as it seeks to expand its data center operations.
-
Data Centers to Use 4x More Electricity by 2035, Report Says
Data centers are projected to consume one-fifth of U.S. electricity by 2035, four times current levels, according to BloombergNEF.
-
Nvidia Unveils Vera Rubin Chip System for AI Data Centers
Nvidia’s Vera Rubin system processes 10 times more tokens per watt than its predecessor, with a release expected in the second half of 2025.
-
Syensqo Advances AI Materials for Next-Gen Semiconductor and Data Center Tech
Syensqo is enhancing materials for AI infrastructure, using innovations in polymers and fluids to meet rising demands in semiconductor manufacturing and data centers.