Recently, the AI computing platform has ushered in a new round of upgrades. From NVIDIA's release of Rubin GPUs, to Intel's release of Xeon 6, to AMD's Ryzen and EPYC processors, the importance of AI acceleration is emphasized. Especially in the PC field, Windows on ARM products are poised to launch, and the x86-based AI PC is targeting advanced manufacturing processes, broadening the scope of AI TOPS and applications to a wider product line.
With the advent of the AI era, the memory capacity of all AI PCs, data centers, edge computing, and terminal applications will inevitably show a multiplier growth pattern, and the demand for DRAM and NAND Flash will increase significantly, and even promote a new round of technological change, HBM and CXL technologies are good examples. What's more, flash-based storage has begun to play a more important role in this.
Multiply data center performance
The recent NVIDIA DGX B200 has taken AI performance to a whole new level, with exponential growth in real-time large language model inference and AI training performance, making it the key to AI transformation, including large language models, recommendation systems, and chatbots.
This means that the DGX B200 needs to handle a wide range of complex workloads, and in addition to the use of high-spec NVLink interconnect technology and Blackwell GPUs, it also needs the agility and scalability required by high-performance storage solutions for enterprise data-intensive workloads.
This is a complex storage system that requires reliable RAID compatibility and interoperability under PCIe 5.0 and NVMe specifications. In particular, generative AI and retrieval augmented generation (RAG) systems are more important than ever, and relational databases, PostgreSQL, and vector database requirements are even more important.
To this end, Kioxia and Xinnor have partnered to announce at COMPUTEX 2024 that their PCIe 5.0 NVMe SSDs have successfully passed compatibility and interoperability tests for Xinnor's RAID solution, delivering up to 25 times faster performance when executing PostgreSQL in degraded mode than a software RAID solution with the same hardware configuration.
Not only that, but the Kioxia PCIe 5.0 SSD boasts excellent performance. With PCIe 5.0 and NVMe 2.0 support, and the product has fully unleashed the performance of PCIe 5.0, the maximum capacity has also reached 30.72TB.
Improve the AI experience of terminals
Whether it's ARM-based or x86-based AI PCs, accelerating AI computing power has become a major competitiveness. Offline models running on AI PCs can well protect users' privacy needs, and at the same time, in a network-constrained environment, relying on the performance of AI PCs can also obtain smooth local AI experiences such as Wensheng diagrams, diagrams, and information collation.
The increase in storage requirements brought about by AI PCs has been reflected in the products. A single offline model has been tens of gigabytes at every turn, and the capacity is already comparable to that of a triple-A game masterpiece. If you need to switch between different apps, hundreds of gigabytes of storage space are also necessary. Therefore, we take a closer look at the AI PC products released this year, and while the standard configuration is 32GB of RAM, 1TB of storage space has become the basic disk. In the future, Lunar Lake will also put forward further requirements for storage.
The same is true for Snapdragon X Elite-based laptops, and the further promotion of Windows on ARM also requires the support of PCIe SSDs, and in the future, we will also see the use of PCIe 4.0 SSDs with more than 1TB paired with Snapdragon X-series processors to better leverage AI performance.
One of the biggest features of AI PCs is the focus on thinness, which requires not only sufficient performance but also a smaller footprint. Kioxia's BG6 series SSDs demonstrate exceptional performance and flexibility in powering AI PCs. PCIe 4.0 and NVMe 1.4c compliant, with up to 2,048 GB of KIOXIA BiCS FLASH 6th generation TLC flash memory, the smallest of which is M.2 2230, even for tablet laptops. In the future, the 8th Gen KIOXIA BiCS FLASH QLC memory will also be ready to be launched, providing more possibilities for mobile device storage capacity and performance.
As one of the important promoters of UFS 4.0, Kioxia has applied this technology to products, especially in the flagship phones and tablets recently released by major manufacturers. It also further provides more possibilities for the personalized design of AI devices.
As AI computing power continues to upgrade, we will see more products prioritize high-capacity storage products, and Kioxia will continue to devote itself to the R&D and innovation of flash memory solutions to become one of the trusted storage product solution providers for users around the world. Kioxia will not only store data, but will create value by leveraging it. In the process of diversifying the digital society, we help create industry standards, respond to changing market trends, and drive the industry in an exciting direction.
(8806881)