Seagate Invents Super-Fast PCIe Hard Drives – Still Slower Than SSDs, But Much Cheaper

Hard drive technology dates back to the 1950s and relies on outdated input/output standards. Seagate is hoping to modernize magnetic storage technology and provide it with PCIe support as the AI ​​boom creates demand for faster storage in data centers.

Image source: seagate.com

Seagate has changed the way hard drives are designed and evolved by combining them with the NVMe standard in a new storage technology. The company aims to simplify and optimize storage solutions for AI companies, as hard drives are still cheaper than solid-state drives. They will never reach the same speeds as SSDs, but Seagate believes they can be improved and made suitable for data center use. By adapting the PCIe-based protocol and deploying NVMe-compatible hard drives, the company has eliminated the need for proprietary solutions tied to SAS/SATA interfaces, host bus adapters (HBAs), and controller architectures that are not suited for AI workloads. The new protocol promises high throughput and low latency, although magnetic drives are unlikely to be able to move data at gigabytes per second.

The next-generation hard drives will not require HBAs or special controllers, but will retain the SAS/SATA connector for improved compatibility. The technology will simplify the deployment of storage solutions for AI, and a single NVMe driver software stack will improve the efficiency of SSDs and HDDs. Seagate has provided the ability to directly connect NVMe drives to GPUs so that the architecture does not depend on the speed of the CPU. As a result, AI models will be able to process large data sets with reduced I/O latency. NVMe over Fabrics (NVMe-oF) technology allows the use of hard drives in distributed, scalable storage systems.

Seagate tested a new system that includes new hard drives, NVMe SSDs, Nvidia BlueField DPUs, and AIStore software solutions. The company did not provide specific results, but said the new platform was able to provide several benefits for AI workloads. Directly connecting the graphics subsystem to the drives via the DPU reduced latency and eliminated SAS/SATA overhead due to a simplified system architecture. AIStore helped optimize data caching and improve performance when training AI models, and NVMe-oF integration proved its usefulness in multi-rack storage clusters. NVMe-compatible hard drives reduce carbon emissions per gigabyte by 10 times, improve power efficiency by 4 times, and offer a “significantly lower” cost per terabyte than SSDs, Seagate said.

The company is currently working to scale its Mozaic 3+ platform based on Heated Assisted Magnetic Recording (HAMR) drives, creating higher-capacity hard drives, and continuing to explore NVMe-compatible solutions with its partners.