NVIDIA considers aggressive measures: reducing the memory capacity of the Rubin Ultra chip.

NVIDIA is considering a radical measure to address the shortage of high-end, high-bandwidth memory chips: using a lower memory configuration than originally planned for the next-generation graphics processor, Rubin Ultra. According to three informed sources familiar with the testing work, NVIDIA has been testing at least three versions of the Rubin Ultra graphics card over the past few weeks, with some versions featuring a memory capacity below the specifications originally announced by the company. These three sources indicated that NVIDIA is considering lower-memory versions partly due to its potential inability to acquire sufficient high-end memory to meet the supply demands of the original design. Reducing memory will impact the chip's performance; however, NVIDIA could compensate in other ways. Nonetheless, an AI company using this lower-memory chip to run large AI models would need to employ more chips than originally anticipated. This consideration by NVIDIA reflects the reality that, amid the AI boom, the chip industry is struggling to meet the surging demand from data centers. There is a severe shortage of memory chips, a core component of GPUs, driving up product prices. The rising costs have triggered a chain reaction across the tech industry: companies are forced to exceed budget expenditures, and firms such as Apple have also raised the prices of their hardware.