Navigating AI – Hyperscale Flash Standards and Enterprise SSD Innovation

The Future of Memory and Storage keynote showcased the power of collaboration in tackling the challenges and seizing the opportunities presented by AI in the Enterprise SSD market.

The Future of Memory and Storage keynote showcased the power of collaboration in tackling the challenges and seizing the opportunities presented by AI in the Enterprise SSD market. As a key contributor to open-source initiatives and a leader in controller innovation, Fadu is proud to be part of this collaborative effort. Join us as we explore the key takeaways from the keynote, emphasizing the importance of working together to drive the evolution of the storage ecosystem.

Western Digital: Navigating the Transformation

Eric Spanneut, GM/VP of Marketing from Western Digital, started the session by outlining the dramatic changes in the Enterprise SSD market. He emphasized the growing demand for SSDs, fueled by the rise of AI and the disaggregation of compute and storage in the cloud. Spenot also highlighted the distinct requirements of compute and storage Enterprise SSDs in the AI era. Compute SSDs, crucial for AI training and inference, need to be incredibly fast and offer low latency. Storage SSDs, essential for AI data preparation, require higher capacities and balanced performance.

WD also left us with a market analysis of the impact of AI on storage. The rapid growth of AI presents a golden opportunity for the Enterprise Solid-State Drive (eSSD) market. WD projections indicate a substantial increase in demand, with an estimated 150 exabytes of additional eSSD storage needed by 2028, driven solely by AI applications. However, this AI-fueled growth also brings about a significant shift in the types of eSSDs required. We’re witnessing a surge in demand for compute eSSDs, vital for AI training and inference tasks, and a parallel increase in the capacity demands for storage eSSDs, which handle the massive datasets used in AI.

Meta: A Hyperscale Perspective

Ross Stenfort, Hardware Engineering from Meta offered a hyperscale perspective on the challenges and solutions in the Enterprise SSD space.

Ross showed tremendous growth in the E1.S market and the ability to scale to high capacity with QLC, while calling to action the open source community to get together to think about even larger drives in the future (think hundreds of terabytes!)

Ross also emphasized the importance of flexible data placement (FDP) for improving endurance, performance, and quality of service, and huge advancements in the open-source software support.

Open-Source SoftwareStatus
Linux Kernel: I/O PassthroughComplete
Linux Kernel: Lifetime HintsIn Progress
xNVMeComplete
QEMUComplete
FioComplete
nvme-cliComplete
CachelibComplete

He highlighted Meta’s contributions to open-source initiatives like the Open Compute Project (OCP) and the release of open-source qualification test cases, which are accelerating development and qualification timelines.

Meta OCP Framework https://github.com/opencomputeproject/ocp-diag-autoval

Meta OCP Storage Tests https://github.com/opencomputeproject/ocp-diag-autoval-ssd

Collaboration on Real World Problems ➡️Leads to Great Ideas ➡️Resulting in Great Products

Fadu: Controller Innovations

Anu Murthy, VP Marketing from Fadu showcased the company’s advancements in SSD controller technology. Fadu’s Gen 5 controllers are already delivering industry-leading performance and power efficiency, addressing the demands of AI infrastructure. The Echo controller is powering the WD SN 861 SSDs optimized for the high-performance requirements of AI workloads.

Anu highlighted that the Fadu controller architecture enables the best-in-class energy efficiency, measured in performance per W, which leads to a significant reduction in server power and rack level TCO.

Jonmichael Hands, Sr. Director of Product Planning, had a dedicated session to both FDP and energy efficient controller design (which also have detailed blog posts to go with)

https://blogs.fadu.io/wp-content/uploads/2024/08/20240806_FARP-101-1_Hands.pdf

https://blogs.fadu.io/wp-content/uploads/2024/08/20240807_SSDT-303-1_Hands.pdf

Murthy also unveiled Fadu’s next-generation Gen 6 controller, Sierra, which promises to double power efficiency while maximizing PCIe Gen 6 bandwidth. This was a major announcement highlighting

  • Full PCIe 6.0 x4 performance > 28GB/s
  • Optimized for AI training performance by delivering maximum interface random read (>6.6M IOPS)
  • Flexible Data Placement for cloud-native workloads, AI training checkpointing, caching, and database
  • Best-in-class power efficiency over 2GB/s per W
  • Support SLC, TLC, and QLC up to 256TB
  • Support for latest OCP NVMe SSD spec
  • Support for new NVMe 2.1 features
  • Virtualization enhancements including multiple functions and PCIe ATS custom implementation
  • Latest security features, including confidential computing, media sanitization, quantum-safe crypto algorithms, and OCP hardware root of trust (Caliptra)

Fadu’s commitment to innovation extends to its business model, which aims to enable faster time to market for SSDs through closer collaboration with SSD manufacturers.

Conclusion

Artificial Intelligence is reshaping the Enterprise SSD landscape at an unprecedented pace. The Flash Memory Summit 2024 keynote offered a glimpse into this transformation, with industry leaders sharing insights on the evolving demands, challenges, and innovations in the SSD space. At Fadu, we’re developing cutting-edge controller technology that empowers the next generation of high-performance, energy-efficient SSDs, essential for navigating the AI era.