Silicon Motion SM2524XT delivers 14 GB/s speeds to mainstream SSDs with AI optimization

Silicon Motion SM2524XT Pushes Mainstream SSDs to 14 GB/s

Silicon Motion has unveiled the SM2524XT, a PCIe Gen5 DRAMless AI SSD controller that delivers sequential read speeds up to 14 GB/s while targeting AI workloads that demand sustained low-latency performance 1

1

2

2

. While high-end SSDs with eight-channel controllers have approached these speeds for years, mainstream drives with quad-channel controllers typically offer around 11 GB/s. The SM2524XT achieves this performance leap through four NAND channels supporting transfer rates up to 4,800 MT/s, combined with newer 3D NAND that features higher-speed interfaces 1

1

. The controller also delivers random performance up to 2.5 million IOPS, marking a significant advancement for mainstream solid-state drives 3

3

.

Addressing KV Cache Latency in AI Inference Workloads

The SM2524XT specifically targets KV Cache-heavy storage operations that have become critical bottlenecks in AI inference workloads running on next-generation AI PCs 2

2

. KV cache serves as a storage area where AI models keep previously processed data, eliminating the need to recalculate information for every new token. While this approach significantly reduces compute overhead, it generates massive amounts of small, random, latency-sensitive memory and storage accesses 1

1

. Unlike data center environments, local large language model deployments on PCs face performance penalties due to relatively limited DRAM capacity, making KV cache latency a limiting factor for long-context and multi-agent workloads. Compared to its predecessor introduced about a year ago, the SM2524XT boosts random workload throughput by as much as 25%, reduces latency, and improves responsiveness during fragmented read/write operations typical of on-device LLM execution 1

1

.

Advanced Technologies Enable Consistent AI Performance

Built on TSMC's 6nm process technology, the SM2524XT integrates several proprietary technologies designed to maintain stable performance under sustained AI-driven workloads 2

2

3

3

. The controller features Separated Command Address technology, which separates command and address traffic inside the NAND interface, enabling simultaneous processing rather than serial operations to lower latency and increase effective bandwidth 1

1

. The chip also incorporates NANDXtend LDPC ECC technology to cope with read errors in the latest flash memory, alongside advanced FTL scheduling to improve parallel data processing efficiency 2

2

. The DRAMless design delivers up to 25 percent higher performance per watt compared to the previous generation, balancing lower power consumption and reduced thermal density—increasingly relevant for compact AI systems and mobile workstation platforms 3

3

.

Market Implications for AI PC Storage

Nelson Duann, Senior VP of Client & Automotive Storage Business at Silicon Motion, emphasized the strategic importance: "KV Cache has become a critical factor in AI inference performance, driving the need for sustained high random read/write throughput and low-latency data access. As AI PCs evolve to support increasingly complex Local Agent and on-device LLM workloads, the SM2524XT is designed to deliver the random I/O performance, latency stability, and power efficiency required for next-generation AI storage architectures" 1

1

. The announcement reflects a broader market transition where storage vendors increasingly adapt products for AI-oriented client systems rather than focusing exclusively on gaming and traditional desktop usage 2

2

. The controller's optimization for edge inference workloads extends its utility beyond PCs to edge AI devices 1

1

. Silicon Motion will demonstrate prototype drives based on the SM2524XT at Computex next week, though actual SSDs based on the chip are not expected to arrive earlier than next year 1

1

. The company's shares have surged 366% over the past year, with the announcement contributing to 36% revenue growth in the last twelve months 3

3

.