Your cart is currently empty!
It’s basically a solid-state drive that’s been optimized for AI workloads. Unlike regular SSDs, these are designed to help with huge AI models—handling both performance and hyper-large capacity. Huawei just launched three of them: OceanDisk EX 560, SP 560, and LC 560. These drives offer up to 245 TB of storage (the LC 560 being the biggest) and blistering read speeds up to 14.7 GB/s—geared specifically for AI data demands [1]. Yicai also confirmed the massive capacity for AI inference and training [2].
In AI models, GPUs use high-bandwidth memory (HBM) for fast access to key-value (KV) cache data. But when that fills up, the system must recompute data—slowing everything down. The AI SSD, together with Huawei’s Unified Cache Manager (UCM), acts like a smart cache hierarchy: it shifts evicted KV data from HBM to the SSD on-the-fly and retrieves it when needed, avoiding expensive recomputation. UCM essentially orchestrates across HBM → DRAM → SSD tiers, cutting inference latency and boosting throughput massively [3][4].
Picture this: you have a modular AI system where different “expert” components load into RAM only when needed. The AI SSD becomes your overflow memory, supplying these expert modules at speeds of around 12–14 GB/s—almost as fast as regular RAM. That helps reduce the RAM bottleneck. Huawei’s push toward PCIe Gen 5 is exactly about hitting those speeds [3].
Not exactly plug-and-play. To get these speeds, you’d need a PCIe Gen 5 slot (2nd gen PCIe 5) directly on your board. That likely means upgrading your motherboard, case, and cooling—roughly costing an extra ₱15k–₱20k in hardware. Then there’s the AI-SSD itself (price not officially disclosed yet). Plus, you need proper driver support (like Huawei’s DiskBooster) and integration in AI platforms (LM Studio, Ollama, Hugging Face, etc.) for them to detect and manage these expert swaps seamlessly. Huawei also launched an AI SSD Innovation Alliance to push standardization [5][1].
This is a leap toward consumer-grade setups capable of running 700 GB-class models (at lower quantization)—something that used to cost a few million PHP in specialized setups. Now? You could potentially pull it off with ₱150k–₱200k total—GPU + AI SSD upgrade + new rig. That’s a pretty big shift.
Stay secure, stay informed, and take control of your digital security.
For more information on how to stay secure in the digital age, stay tuned to the Comfac-IT Blog.