MilikMilik

Silicon Motion’s DRAM-Less PCIe 5 Controller Targets Faster, Cooler AI PCs

Silicon Motion’s DRAM-Less PCIe 5 Controller Targets Faster, Cooler AI PCs
interest|PC Enthusiasts

What the SM2524XT Brings to Next-Gen AI PC Storage

Silicon Motion’s SM2524XT controller is a DRAM-less PCIe 5 SSD controller that uses a quad-core Arm architecture to deliver higher bandwidth, lower latency, and better power efficiency for AI-focused laptops and compact workstations that need fast local storage for inference and KV cache workloads. Positioned as a mainstream client and edge solution, it keeps the DRAM-less SSD formula yet delivers up to 14 GB/s sequential reads and 2.5 million IOPS when paired with fast NAND. Compared with the earlier SM2504XT, Silicon Motion says the SM2524XT offers up to 25% higher performance per watt and similar gains in random I/O, while keeping total drive power under 5 W in active use. That combination of speed and tight power budgeting is aimed squarely at AI PC storage where thermal headroom is limited.

Silicon Motion’s DRAM-Less PCIe 5 Controller Targets Faster, Cooler AI PCs

Quad-Core Arm Design Without DRAM: How the Speed Boost Works

The SM2524XT controller is built on TSMC’s 6 nm process and uses a quad-core ARM Cortex-R8 CPU to coordinate four NAND channels running at up to 4,800 MT/s. Instead of relying on onboard DRAM, the controller schedules the flash translation layer and error correction directly across these cores, keeping critical metadata flows tight while cutting component count. Silicon Motion reports that this design allows the PCIe 5 SSD controller to reach 14 GB/s reads and 12 GB/s writes over a PCIe Gen5 x4 link, with random performance up to 2.5 million IOPS. Those figures are a 25% uplift in random I/O versus its predecessor, and internal tests show about 29% more sequential throughput at almost identical power draw. For AI PC storage, that means faster response to small, scattered reads typical of KV caches and language model prompts.

Why a DRAM-Less PCIe 5 SSD Matters for AI Laptops and SFF PCs

Moving to a DRAM-less SSD design reduces bill-of-materials costs, board footprint, and power consumption by eliminating dedicated DRAM chips and their power delivery. For thin laptops and small form factor AI PCs, fewer components and lower heat directly translate into simpler cooling and more predictable performance under sustained load. Silicon Motion is targeting sub‑5 W active power for SM2524XT-based drives, helped by its PI-LTT intelligent power optimisation, which lowers NAND I/O voltage without sacrificing throughput. The controller also supports ONFI 5.2 and Separated Command Address signalling, which improves efficiency and reduces bus contention. In practice, this allows AI PCs to maintain high PCIe 5 speeds while keeping SSD thermals in check, so storage is less likely to throttle during long inference sessions or continuous KV cache activity.

Reliability, KV Cache Workloads, and the AI PC Roadmap

Beyond headline numbers, Silicon Motion is framing the SM2524XT controller as tuned for AI-era workloads, especially KV cache-heavy inference and local agent tasks. The controller adds proactive fault monitoring and automatic recovery features, alongside the company’s 8th‑generation NANDXtend LDPC error correction and on-disk training to help improve endurance, particularly with QLC NAND. According to Silicon Motion, this helps sustain random I/O performance under power and thermal constraints that would normally limit client drives. Support for Separated Command Address and advanced FTL scheduling further reduces latency interruptions during highly fragmented access patterns. Since Silicon Motion sells the SM2524XT controller rather than complete SSDs, final AI PC storage behavior will depend on SSD vendors’ NAND choices and firmware tuning, but the architecture clearly points toward DRAM-less PCIe 5 SSDs as a cost‑efficient, power‑aware backbone for next‑generation AI PCs and edge systems.

Comments
Say Something...
No comments yet. Be the first to share your thoughts!