Phison has unveiled two new enterprise SSD designs, one featuring a built-in Neural Processing Unit (NPU) and 64GB of memory capable of running AI frameworks like PyTorch and llama.cpp directly on the drive. This computational storage approach aims to reduce data transfer by processing data where it resides, potentially cutting transfers by up to 100x for applicable workloads. The second design integrates fast pSLC cache with large QLC capacity using a single controller, optimizing both speed and storage density. While specific product names and release dates are not yet available, these developments signal a move towards more intelligent storage solutions. AI
IMPACT On-drive AI processing could significantly reduce data transfer bottlenecks and energy consumption for AI workloads.
RANK_REASON The item details new storage technology with integrated AI capabilities, presented at a technical conference, which falls under research and development in hardware innovation. [lever_c_demoted from research: ic=1 ai=0.7]
- NPU
- ARM Cortex-A55
- DGX Spark
- Dhrystone
- llama.cpp
- LPDDR5x
- OpenVINO
- PCIe 6.0
- Phison
- PyTorch
- QLC
- Samsung
- SmartSSD
- Supercomputer 2026
- vLLM
- X3 controller
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →