Open Source

NVIDIA's RTX Spark bandwidth clarified: 600GB/s is NVLink, not memory speed

A Reddit sleuth exposes a major reporting error on RTX Spark's bandwidth at Computex.

Deep Dive

A Reddit user has pointed out that every outlet reporting 600GB/s from Computex slides is completely wrong—that figure is actually the NVLink interconnect speed.

Key Points
  • 600GB/s claimed memory bandwidth for RTX Spark is actually the NVLink interconnect speed for multi-GPU setups.
  • Error originated from misinterpretation of Computex slide annotations by tech media outlets.
  • Actual memory bandwidth remains unpublished; estimated at 300–400 GB/s based on hardware specs.

Why It Matters

Accurate memory bandwidth specs are critical for AI professionals planning high-throughput training and inference on RTX Spark.