NVIDIA denies Rubin and Vera memory cut rumors amid DRAM shortage

NVIDIA used a blog post to push back on speculation that Vera and Rubin had been cut down because of component shortages. The company said the configurations in question are different SKUs and customer-specific tuning, while Jensen Huang has said Rubin, Rubin Ultra, and Kyber servers are on track.

NVIDIA also said it continually adjusts compute, networking, and memory to deliver the best performance and efficiency for customers. In the same discussion, the company confirmed that Vera's modular SOCAMM architecture allows up to 1.5 TB of memory, matching the official specification.

The rumors centered on Vera CPU SOCAMM2, where the configuration was said to drop from 1.5 TB using 192 GB modules to 768 GB using 96 GB modules. According to Wccftech, NVIDIA framed that change as a SKU difference rather than a downgrade.

Rubin GPUs were described as keeping 288 GB of HBM4 memory with up to 22 TB/s of bandwidth. The article also said Rubin Ultra had previously been outlined with 1 TB of HBM4e memory, while The Information reported that NVIDIA had been testing at least three versions of the Rubin Ultra GPU.

NVIDIA also unveiled Vera BlueField-4 STX, a storage platform for AI data. The company said the Vera CPU in that platform delivers up to 3.21x higher throughput than an x86 CPU in a two-stage compression and encryption pipeline.

For now, NVIDIA's message is simple: the company says it is adjusting memory configurations to fit different scenarios, not secretly shrinking them. Do you think these rumors were just a reaction to the DRAM shortage?