What a PCIe Gen6 SSD Controller Is—and Why It Matters for AI PCs
A PCIe Gen6 SSD controller is the chip that manages data flow between high-speed NVMe storage and the PCIe 6.0 interface, enabling up to 28 GB/s storage speed while keeping latency and power consumption low, which makes it a foundation technology for faster AI inference and data-heavy PC workloads. In current AI PC platforms, PCIe Gen5 SSDs already push around 14 GB/s, but controller vendors are preparing the next leap. Phison’s PS5303-X3 PCIe Gen6 SSD controller targets 28 GB/s throughput at around 7 watts, roughly doubling PCIe Gen5 bandwidth while improving efficiency. InnoGrit’s new IG5686 PCIe Gen6 SSD controller matches that 28 GB/s headline speed and scales capacity up to 256 TB for enterprise AI and data center use. Together, these designs show how next-gen NVMe technology will remove storage bottlenecks for local AI tasks, from large language model inference to real-time media processing on advanced PCs.
Phison PS5303-X3: 28 GB/s at 7 W Targets AI Performance per Watt
Phison’s PS5303-X3 PCIe Gen6 SSD controller, shown on early E3.S and E1.S reference drives, is built around a PCIe Gen6 x4 interface and NVMe 2.3 support. The company is targeting up to 28 GB/s sequential read and write speeds and around 6.8 million IOPS in random workloads, positioning the controller as a direct doubling of high-end PCIe Gen5 performance. According to Phison, the X3 achieves about “4000 GB/s per watt,” translating to peak throughput at roughly 7 watts and delivering about double the performance per watt of its existing solutions. That balance is critical for AI PCs and servers, where dense NVMe storage shares a constrained thermal and power envelope with GPUs and NPUs. Phison is also preparing a full PCIe 6.0 ecosystem of retimers and redrivers, which is important if these 28 GB/s SSDs are to maintain signal integrity over longer motherboard traces and through add-in risers.
InnoGrit IG5686: Massive Capacity and a Roadmap to 100 Million IOPS
InnoGrit’s Crestone IG5686 PCIe Gen6 SSD controller takes a slightly different angle, prioritizing scale for data-intensive AI environments. The controller supports PCIe Gen6 x4, NVMe 2.3, and a wide range of flash types including SLC, MLC, TLC, QLC, and storage-class memory, with NAND speeds up to 4800 MT/s. It is designed for E1.S and E3.S drives with capacities up to 256 TB and performance of up to 28 GB/s reads, 22 GB/s writes, and about 7 million/5 million random read/write IOPS. For large AI inference clusters with long context windows, this blend of bandwidth and capacity helps keep training data, embeddings, and vector indexes on fast NVMe tiers. InnoGrit’s roadmap goes further: the company expects 25–50 million IOPS from tuned PCIe Gen6/CXL designs around 2027 and aims for 100 million IOPS with PCIe Gen7 SSDs by 2028, targeting what it calls AI-native storage architectures.
Silicon Motion SM2524XT: Smarter PCIe Gen5 Today for AI Inference
While PCIe Gen6 remains pre-market, Silicon Motion is pushing current-gen NVMe technology forward with its SM2524XT PCIe 5.0 SSD controller. This DRAM-less SSD controller uses a new four-core architecture and supports 4800 MT/s NAND, reaching up to 14 GB/s over a PCIe 5.0 x4 interface and around 2.5 million IOPS of random performance. Built on a 6 nm process, Silicon Motion says the SM2524XT delivers “up to 25 percent higher performance per watt compared to the previous generation controller,” with about 25 percent more random I/O performance as well. That focus on random access and latency is tailored for AI inference, especially workloads dominated by highly fragmented KV cache access. Features such as Separated Command Address, advanced FTL scheduling, and NANDXtend ECC aim to keep performance consistent under sustained load, making DRAM-less SSD controller designs more attractive for cost-sensitive AI PCs and inference servers.
What PCIe Gen6 Means for Next-Gen AI PC Storage Performance
Together, Phison’s PS5303-X3 and InnoGrit’s IG5686 show what PCIe Gen6 SSD controllers will bring above today’s PCIe Gen5 designs such as Silicon Motion’s SM2524XT. For AI PCs and workstations, the jump to around 28 GB/s storage speed allows local models, embeddings, and media datasets to stream from NVMe with much less time spent in I/O wait states. Power efficiency gains—Phison highlighting 7-watt peak operation and Silicon Motion emphasizing 25 percent higher performance per watt—mean these benefits arrive without blowing through thermal limits in compact systems. In data centers, InnoGrit’s up-to-256 TB drives and future multi-tens-of-millions IOPS targets point toward storage tiers purpose-built for AI-native workloads. For consumers, the same controller architectures will filter down into M.2 drives, where next-gen NVMe technology will make AI-assisted applications feel more responsive as models grow and on-device inference becomes the standard experience.
