<ul data-eligibleForWebStory="true"><li>Nvidia aims to push storage to match AI demands by targeting 100 million IOPS for small-block random reads, a huge goal according to Silicon Motion CEO.</li><li>AI inference often involves small, scattered data bits, making 512B random reads crucial but challenging to accelerate.</li><li>Current PCIe 5.0 SSDs reach around 14.5GB/s and 2-3 million IOPS, but Nvidia seeks faster storage solutions to keep up with AI advancements.</li><li>Industry players like Kioxia are developing new memory technologies, such as XL-Flash, to meet ultra-fast storage demands, but further innovation is needed to achieve 100 million IOPS cost-effectively.</li></ul>

Towards the Giga IOPS pipedream: How Nvidia wants to reach 100 million IOPS - even if it means inventing totally new types of memory

Discover more