Recently, NVIDIA's DGX B200 AI server was listed at Broadberry at the price of $515,410.

The introduction of Blackwell architecture has stirred significant interest across the tech industry, with high praise from Jensen Wong and swift adoption by major technology companies such as Microsoft and Meta.

The DGX B200 integrates eight B200 GPUs, offering an impressive 1.4TB of GPU memory and up to 64TB/s in HBM3E memory bandwidth.

NVIDIA claims that this server can achieve up to 72 petaFLOPS in training performance and 144 petaFLOPS in inference performance, representing an advancement of threefold and fifteenfold in training and inference, respectively, compared to its predecessors.

On the software front, the NVIDIA DGX suite includes the NVIDIA AI Enterprise software, tailored for enterprise-level development and deployment. Users can leverage DGX B200 systems to construct DGX SuperPODs, aiming to establish AI centers of excellence, thereby enabling large development teams to execute a diverse array of tasks efficiently.

Markedly, this marks the debut of NVIDIA's Blackwell AI products in the online retail sphere. OpenAI recently mentioned via its social media that it had received the earliest DGX B200 engineering units from NVIDIA.