NVIDIA GB300 NVL72 hits 20x AgentPerf throughput

// 2h agoBENCHMARK RESULT

NVIDIA GB300 NVL72 hits 20x AgentPerf throughput

NVIDIA's GB300 NVL72 platform achieved record performance on Artificial Analysis's new AA-AgentPerf benchmark, delivering up to 20x more concurrent agents per megawatt than previous H200 systems. The platform leverages full-stack optimizations like high-speed NVLink fabrics and DeepGEMM to handle the complex, multi-turn reasoning requirements of advanced AI agents.

// ANALYSIS

As the industry shifts from simple chatbots to long-horizon agentic workflows, hardware optimization is evolving from raw token latency to high-density, multi-turn reasoning throughput.

* The shift to "relay-style" agent workflows makes traditional token-per-second benchmarks less relevant compared to multi-turn agent trajectories.

* Fusing 72 GPUs over a single NVLink domain is crucial for maintaining low-latency state sharing in complex, iterative reasoning tasks.

* Hardware-software co-design using DeepGEMM, MXFP4 precision, and SGLang routing is proving to be a major differentiator for scaling agent concurrency.

// TAGS

nvidiagb300-nvl72agentperfagentbenchmarkgpublackwelldeepgemm

DISCOVERED

2h ago

2026-06-22

PUBLISHED

2h ago

2026-06-22

RELEVANCE

9/ 10

AUTHOR

DIY Smart Code

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

BENCHMARK34m ago

GLM-5.2 test challenges proprietary model costs

A social media post by user @nsxdavid highlights an interesting counter-result to assertions by others in the community that Z.ai's open-weights model GLM-5.2 is more expensive to operate than proprietary models like Claude Opus 4.8 and GPT-5.5. The comparison touches on the ongoing debate surrounding token consumption versus raw API pricing, showing that real-world deployment costs for open-weights reasoning models can vary significantly depending on the implementation details and task configurations.

FUNDING34m ago

Cursor keynote showcases agentic coding future

At Cursor's inaugural user conference Compile, the company showcased its roadmap and AI capabilities, including a preview of Composer 3 and its evolution into an AI software factory. The event drew massive interest following SpaceX's announcement of a $60 billion all-stock acquisition of parent company Anysphere to integrate Cursor into its engineering workflow.

UPDATE55m ago

Grok Build introduces /goal command

Grok Build has introduced the /goal command, enabling the platform to execute multi-step coding tasks as a fully autonomous agent. Developers can define an objective and let the agent decompose tasks, modify code, run scripts, and verify completion.