NVIDIA Boosts DGX and RTX AI Inference by 40%, DGX Spark Clusters Scale to 8 Units and 1 TB

NVIDIA announced at GTC Berlin that DGX Spark now supports 8-way clusters, pooling 512 GB of memory on 64 GB systems and 1 TB on 128 GB configurations. Inference on DGX and RTX systems is up 30–40% via vLLM and llama.cpp optimizations, and the 64 GB DGX Spark ships October 23.

• An 8-unit DGX Spark cluster pools up to 1 TB of unified memory
• Inference gains: 30% on DGX Spark, 40% on RTX PRO 6000
• Recursive Flint model hits over 220 tokens/s on 64 GB RTX Spark
• 64 GB DGX Spark launches on October 23