While physical capacity remains at 288 GB between generations, the Vera Rubin NVL72 achieves up to 30 times the throughput per megawatt of the GB300 NVL72 on agentic benchmarks. The company developed the NVFP4 format to halve key-value cache sizes and reduce weight footprints by 1.8 times compared to FP8. Additionally, the GB200 NVL72 leverages 130 TB/s of aggregate NVLink bandwidth to distribute expert weights across interconnected GPUs via TensorRT-LLM.
Key points
Data compression techniques like NVFP4 allow roughly twice as much context to remain resi…: Data compression techniques like NVFP4 allow roughly twice as much context to remain resident on-device while preserving accuracy.
Pooling memory across aggregate NVLink bandwidth frees local GPU memory to store larger k…: Pooling memory across aggregate NVLink bandwidth frees local GPU memory to store larger key-value caches.
Utilizing wide expert parallelism enables systems to generate up to 1.8 times more output…: Utilizing wide expert parallelism enables systems to generate up to 1.8 times more output tokens per second per GPU.
Sources checked by our analysis pipeline: 1
In an AI factory, high bandwidth memory (HBM)is the fastest, most expensive, and most contested memory tier. Every byte spent on a redundant weight copy, an idle session’s KV cache, or a recomputed prefix is a byte that is not being used to produce tokens. The question is no longer how much memory a GPU has, but how much useful work each byte delivers.
The generational data in Figure 1 below shows why. From GB300 to NVIDIA Rubin, capacity held at 288 GB while bandwidth rose to 19.2 TB/s.
agentic throughput per megawatt grew far faster. On SemiAnalysis AgentX, which replays recorded coding-agent sessions with real tool-call gaps and growing context, NVIDIA Vera Rubin NVL72 then delivers up to 30x the throughput per megawatt of GB300 NVL72 at 160 tokens per second per user.
Capacity alone cannot explain a 30x gain delivered from the same 288 GB per GPU. The difference is how much useful work each byte delivers.
NVIDIA treats this as a co-design problem that spans GPU architecture, the scale-up interconnect, serving software, the CPU, and storage. This…
The collaboration brings together advanced artificial intelligence capabilities and financial data infrastructure to deliver automated guidance. Users will gain access to specialized technology designed to assist with personal money management. This joint effort reflects the growing integration of automated intelligence systems within consumer fintech applications.
The integration equips the AI assistant with direct visibility into personal monetary habits to help consumers better evaluate their ongoing expenditures. By leveraging secure data aggregation, users can monitor cash flow patterns without exposing their institutions to transactional changes. This update expands the assistant's utility into everyday money management and proactive budgeting support.
Automated tools operating on behalf of millions of people are reshaping online interactions by browsing continuously and operating without physical human constraints. This influx challenges standard monetization and infrastructure planning as non-human visits threaten to overtake conventional traffic. Web platforms must now establish identity frameworks to maintain security and direct user accountability without shutting out these digital intermediaries.