NVIDIA Blackwell Release and Architecture Record

NVIDIA Blackwell Release and Architecture Record

Historical launch

NVIDIA introduced the Blackwell platform on March 18, 2024. The launch record described six architecture technologies, including a two-die GPU design, a second-generation Transformer Engine, fifth-generation NVLink, a reliability and availability engine, secure AI capabilities, and a decompression engine.

Named launch configurations included the B200 GPU, GB200 Grace Blackwell Superchip, and GB200 NVL72 rack-scale system. "Blackwell" therefore referred to an architecture and platform family, not one interchangeable chip.

Claim boundary

NVIDIA said the platform enabled real-time generative AI on trillion-parameter models at up to 25 times less cost and energy than its predecessor for specified inference comparisons. The statement is a vendor claim tied to NVIDIA's workloads, software, systems, and comparison.

The public profile should not convert that statement into a universal speed, power, cost, sustainability, or model-quality result. Parameters are not a quality measure. Facility power, cooling, networking, utilization, software, precision, model architecture, and workload affect the operational result.

Lifecycle

NVIDIA introduced Blackwell Ultra in March 2025 and named GB300 NVL72 and HGX B300 configurations. NVIDIA introduced the Rubin platform in January 2026 as its next platform generation and reported Vera Rubin production activity later in 2026.

This chronology does not make Blackwell irrelevant. It makes exact platform and configuration names more important. A page about the 2024 launch should not describe Rubin claims as Blackwell specifications.

Publication use

Use Blackwell as a maintained platform profile with a dated launch section and a separate lifecycle section. Recheck architecture, systems, availability, cloud access, specifications, software, export controls, benchmarks, pricing, and support immediately before publication or procurement use.