Samsung and AMD made a revolutionary SSD together - then it was left to wither in the shadows and nobody knows exactly why (www.techradar.com)

Samsung’s SmartSSD — a 2018-era computational storage concept that co-located NAND, HBM and RDIMM memory with a Xilinx FPGA accelerator inside the SSD — promised to push compute to where data lives and reduce server dependence. AMD’s acquisition of Xilinx in 2020 provided the FPGA muscle, and Samsung shipped a Gen3 SmartSSD (today still sold under the AMD Xilinx label, e.g., a 3.84TB SKU around $518). But the product has quietly vanished from Samsung’s lineup: the hardware was complex, hard to sell, and the COVID and generative-AI era shifted priorities. LLM workloads emphasized massive storage capacity and specialized inference accelerators (ASICs/TPUs) rather than the in‑drive, FPGA-style compute CSDs offered. For AI/ML practitioners the SmartSSD story is a cautionary lesson in timing and specialization. Technical advances (onboard FPGA, HBM, compression-enabled logical capacity like ScaleFlux’s CSD5000’s 122.88TB physical → 256TB logical at ~2:1) show what’s possible, but standards and ecosystem support lag — SNIA’s CS API only launched in Oct 2023 and progress is slow. Still, as inference moves to the edge and hyperscalers push ASICs, a niche market for inference‑friendly, compute‑enabled SSDs could re-emerge; today, however, the economics and workload fit tipped the scale away from broad CSD adoption.