NVIDIA details three inference optimizations for DGX Spark
NVIDIA says three DGX Spark optimizations lifted Ling-3.0-flash throughput and cut latency in GSM8K testing, with a caveat on stacking them.
NVIDIA expands DGX Spark clusters to eight nodes for local AI
NVIDIA says DGX Spark now supports eight-node clusters, lifting unified memory to 1TB and improving throughput for local AI workloads.
Microsoft to Scale Azure Maia to Multi-Gigawatt Deployments Starting 2027
Microsoft announced October 12, 2026 that Azure Maia 200 is serving production workloads including Copilot, while Maia 300 will begin deployment in 2027 with multi-gigawatt scale planned.
NVIDIA Dynamo's Shadow Engine Slashes LLM Recovery to 7 Seconds
NVIDIA's Shadow Engine Recovery in Dynamo cuts LLM downtime from minutes to 7 seconds, transforming AI infrastructure resilience.
NVIDIA Groq 3 LPX Achieves 3,431 TPS Benchmark on Vera Rubin
NVIDIA's Groq 3 LPX sets a new standard in AI inference, delivering 3,431 tokens/second on a 100K context benchmark and redefining high-interactivity workloads.
NVIDIA Groq 3 LPX Hits Full Production, Boosts AI Inference Speed
NVIDIA's Groq 3 LPX enters full production, delivering record-breaking AI inference speeds for latency-critical workloads. Key to agentic AI development.
AMD ZenDNN 6.0 Boosts AI Inference on EPYC CPUs
AMD's ZenDNN 6.0 introduces FP16 support, MoE optimizations, and expanded vLLM compatibility, enhancing AI inference capabilities on EPYC processors.
NVIDIA Unveils AI Factory Energy Optimization Tools for Token Efficiency
NVIDIA introduces tools like DSX and NVFP4 to improve energy efficiency in AI factories, potentially lowering token production costs by up to 25%.
AI Data Processing Shifts to GPUs: Key Trends and Impacts
AI pipelines are increasingly GPU-driven as inference-heavy workloads handle unstructured data, reshaping data processing and infrastructure demands.
NVIDIA Unveils AI Grid Architecture for Distributed Edge Inference at GTC 2026
NVIDIA's AI Grid reference design enables telcos to cut inference costs by 76% and meet sub-500ms latency targets through distributed edge computing.