NVIDIA Begins Mass Production of Groq 3 LPX…Achieves 3,400 Tokens Per Second Inference Performance
Published2026.08.25 15:53
This product is configured to expand the Vera Rubin NVL72 platform, focusing on increasing the token generation speed of agentic systems.
The previously mentioned benchmark figures were measured under conditions where 100,000 token context was applied to the open-source agentic model Gemma 4 31B, and NVIDIA stated that based on that model, it is currently the fastest performance achieved.
Groq 3 LPX provides response speeds 4 times faster than the most similar alternative platform in this generation step, according to NVIDIA.
This means agentic tasks such as coding can be processed within minutes rather than hours.
NVIDIA Founder and CEO Jensen Huang stated, "Vera Rubin expands its vision with an AI factory configuration optimized for workloads in the agentic AI era, and LPX for ultra-fast token generation pushes the limits of performance even further."
Nebius CTO Danila Shtan stated, "We provide developers with an experience where all steps of the agent loop happen instantaneously while using existing APIs as they are."
AI inference-specialized cloud company Groq is also planned to be among the early adopting companies, NVIDIA said.
관련뉴스
.jpg)
[NVIDIA GTC Taipei & Computex 2026 Special Feature ①] Cosmos 3, RTX Spark, and Vera Rubin Change the Order of AI Factories, PCs, Robots, and Autonomous Driving
Nvidia unveiled eight new technologies including Cosmos 3, RTX Spark, Vera Rubin, Vera CPU, Drive Hyperion, and DSX at GTC Taipei 2026, formalizing its AI full-stack strategy. This announcement is evaluated as a declaration of an AI infrastructure platform that connects beyond GPU performance competition to encompass PCs, robots, autonomous vehicles, data centers, and semiconductor fabs.
2026.06.04by 배종인 기자
.jpg)
[NVIDIA GTC Taipei & Computex 2026 Special Feature ②] NVIDIA Discusses 'The Entire Environment Where AI Works'
While past Nvidia announcements focused on GPU performance and data center acceleration, GTC Taipei 2026 concentrated on designing the entire environment where AI performs actual tasks. Vera CPU, RTX Spark, Cosmos 3, and OpenShell demonstrate Nvidia's strategy to directly integrate CPU, PC, physical AI, and security.
2026.06.04by 배종인 기자
[NVIDIA GTC Taipei & Computex 2026 Special Feature ③] SK Hynix, Microsoft, and TSMC Benefit, Intel, AMD, and Qualcomm Under Pressure
The benefits of Nvidia's announcements are concentrated among companies deeply connected to the Nvidia ecosystem. SK Hynix, Microsoft, Adobe, TSMC, Taiwan ODM/OEM, and AI cloud service providers are likely to experience positive impacts. Conversely, Intel, AMD, Qualcomm, and independent autonomous driving platform companies may face competitive pressure as Nvidia expands its footprint into CPUs, AI PCs, AI factories, and autonomous driving platforms.
2026.06.04by 배종인 기자
Nvidia is the 'Brain' and AMD is the 'Nervous System': The Physical AI Semiconductor War Begins in Earnest
The center of gravity in AI semiconductor competition is shifting beyond data center GPUs to 'Physical AI' such as robots, autonomous vehicles, and smart factories. Nvidia, led by CEO Jensen Huang, is seeking to reorganize the entire computing architecture by leading with its own Arm CPU and full-stack platform, while AMD, led by CEO Lisa Su, is emphasizing customer choice by leveraging x86 compatibility and an open ecosystem.
2026.06.11by 배종인 기자
NVIDIA and LG Group Collaborate Across Physical AI, Autonomous Driving, and Data Centers
Nvidia and the LG Group are entering into collaboration across AI data centers and Physical AI broadly. The core focus is to advance the construction of AI-based systems across diverse business domains including robotics, mobility, and data center infrastructure.
2026.06.11by 배종인 기자
