Memory manufacturer Kioxia announced a collaboration with GPU manufacturer (NVIDIA) and plans to launch a solid-state hard drive (SSD) that reads nearly a hundred times faster than traditional SSDs by 2027. This technological breakthrough will mainly serve the server market required for generative AI computing, and is expected to bring a new generation of innovations in the market.
According to reports from Japan's economic news, this major cooperation case was revealed by the police at the AI market technology briefing held in Tokyo on September 2. Koichi Fukuda, a technician of SSD application technology, said that the company is following NVIDIA's proposals and requirements to promote development. In addition, in contrast to the way traditional SSDs need to be connected to the GPU by the CPU, the new SSD developed in cooperation with NVIDIA will achieve direct connection and data exchange of the GPU. This direct connection mode can greatly improve data transmission efficiency, which is very important for AI computing.
The institution is emphasized that to meet the extremely fast demand, the SSD's random reading performance will reach 100 times the rate of traditional products (Input Output Per Second). Considering that NVIDIA's SSD requirement for GPUs is up to 200 million IOPS, the project was achieved through the joint operation of two SSDs. This innovative product is expected to support the next generation standard "PCIe 7.0" of the PCIe (PCI Express) interface, laying the foundation for high-speed transmission.
System predicts that the development goal of this high-speed SSD is not only to expand the GPU memory capacity, but also to replace some of the HBMs (high-frequency wide memory) currently used in AI servers. This development is expected to provide AI computing with more cost-effective, highly flexible and easy to expand memory solutions. The partnership between Wang Ke and NVIDIA is a sign that the innovation of AI server storage technology will accelerate the development and application of AI technology generated.