FrankX.AI
Research

Source review in progress

NVIDIA Blackwell & Rubin GPU Architecture

NVL72 rack-scale systems, 4-bit floating point (FP4) Tensor Cores, and 5th-gen NVLink interconnects

This brief is being checked against individual publications, benchmarks and methods. Its earlier generated source list and confidence labels did not meet the publication standard. The topic remains available here while the evidence is reviewed.

Review the method Explore the hubs

Related topics

AI Factories, Megawatt Datacenters & Grid Infrastructure High-Speed AI Network Fabrics: InfiniBand, RoCEv2 & Optical Switching AI Inference Optimization Runtimes & Serving Engines Mixture-of-Experts (MoE) & Multi-Head Latent Attention