PCIe Gen5 SSDs Push Client and Prosumer Performance to 14GB/s
Storage vendors used Computex and parallel showcases to cement PCIe Gen5 SSDs as the new baseline for high-performance PCs and AI workstation hardware. addlink’s G57 PCIe Gen5 SSD headlines its virtual lineup with sequential read and write speeds up to 14,000 MB/s, powered by a 6nm controller and integrated thermal management, while the G55 targets prosumers and content creators who don’t need the absolute top-end controller. KIOXIA’s XG10 series brings similar performance to OEM notebooks, desktops, and workstations, boasting up to 14,000 MB/s read and 12,000 MB/s write, alongside up to 2,000K/1,600K random read/write IOPS, roughly doubling the throughput of its Gen4 predecessor. Biwin’s Black Opal X570 PRO joins the fray with 14,000 MB/s reads, 13,000 MB/s writes, and up to 2,000,000 IOPS plus an 8GB DRAM cache on larger models, underscoring how PCIe Gen5 SSDs are rapidly becoming central to both consumer and enterprise storage solutions.
AI-Optimized DDR5: CUDIMM and High-Speed RDIMM for Servers and Workstations
On the memory front, vendors are clearly tuning DDR5 for AI servers and data-heavy workstations. addlink’s SC5 DDR5 CUDIMM memory leverages Intel’s Arrow Lake-era Clock Driver DIMM design to clean up signaling at higher frequencies, reaching up to 8400 MT/s for next-generation motherboards. This aligns with growing demand for low-latency, high-bandwidth DDR5 CUDIMM memory in AI PCs and productivity desktops. G.SKILL is pivoting even more aggressively toward the professional market, using Computex to showcase server-grade RDIMM, workstation RDIMM, and ECC UDIMM modules tailored for database training and continuous compute workloads. These overclockable RDIMM kits, developed in collaboration with Intel and AMD, bring their overclocking heritage to enterprise clients that need consistent, high-throughput DRAM for internal AI platforms. Together, these offerings highlight a stronger focus on AI server memory and AI workstation hardware, where predictable performance and signal integrity matter as much as raw frequency.
Thunderbolt 5 AI Accelerators and Storage Hubs Bring Large Models On-Device
Beyond internal components, external connectivity is emerging as a critical piece of local AI infrastructure. OWC’s Stack AI, billed as the world’s first Thunderbolt 5 AI accelerator and storage hub, exemplifies how Thunderbolt 5 storage and compute are converging. Designed for select Windows and Linux systems, with Mac support planned, Stack AI expands effective AI working memory so existing PCs and laptops can run larger models locally without exhausting GPU VRAM. The goal is to reduce reliance on cloud inference, allowing organizations to retain sensitive data on-prem while avoiding per-token billing and recurring cloud AI fees. By combining high-speed Thunderbolt 5 bandwidth with a storage-centric architecture, Stack AI enables developers, researchers, and businesses to deploy larger models, perform fine-tuning, and accelerate workflows directly at the desk, positioning external accelerators as a new class of enterprise storage solutions for AI-era workloads.
Cloud-to-Edge AI Ecosystems: Integrated Enterprise Storage and Edge Compute
ADATA’s Computex theme, “Activate the AI Core,” showcases how vendors are thinking about storage and memory as part of a continuous cloud-to-edge AI ecosystem. Through its TRUSTA enterprise storage brand, industrial division, and XPG gaming arm, ADATA is building integrated enterprise storage solutions that span data center training clusters, industrial edge deployments, and AI PCs. A highlight is TRUSTA’s AI Scaler memory storage architecture, paired with an AI Scaler Toolkit, which intelligently allocates workloads across GPU, DRAM, and SSDs. By shifting part of the load away from scarce, expensive GPUs, ADATA claims it can cut total deployment costs for AI training and inference by more than half. Industrial demos, including an AAI robotic arm powered by NVIDIA Jetson Thor and digital twin healthcare scenarios, emphasize real-time edge processing. Taken together, these initiatives illustrate a holistic approach where PCIe Gen5 SSDs and optimized DRAM become foundational to both cloud and edge AI pipelines.
Gaming Memory Innovation: RGB Kits, Smart Displays, and Overclocking in an AI World
While enterprise AI dominated the narrative, gaming and enthusiast memory remained a vibrant testbed for innovation that often trickles into professional platforms. Biwin’s Black Opal OC Lab Gold Edition DW100 RGB kit exemplifies this crossover, offering a hefty 192GB configuration (four 48GB modules) at DDR5-6000 CL28, tuned for platforms that benefit from tight timings. Such high-capacity, low-latency kits are increasingly relevant as gamers adopt AI-enhanced content creation and local inference tools on their home rigs. XPG, ADATA’s gaming brand, is likewise highlighting DDR5 gaming modules alongside advanced cooling and power delivery hardware, framing them as building blocks for hybrid gaming and AI workstation setups. G.SKILL’s ongoing work in overclocking-focused modules and international OC competitions continues to refine techniques later applied to RDIMM and ECC products. Features like integrated smart displays or refined RGB control are not just cosmetic; they’re part of a broader trend of telemetry-rich, overclocking-optimized modules that can serve both enthusiasts and professional AI users.
