Phison, 플래시 메모리 확장으로 로컬 AI 추론 성능 향상

작가 피손 | 3월 16, 2026 | 모두, 추천, 소식

Pascari aiDAPTIV™ technology enables larger-model inference on AI devices with intelligent flash tiering to extend retention and reduce recompute

San Jose, CA — GTC 2026 —March 16, 2026 — 파이손전자 (8299TT), a global leader in NAND flash controllers and storage solutions, today announced its GTC showcase at booth 119, demonstrating how multi-tier memory architecture supports larger models and long-context inference on NVIDIA-powered local AI platforms.

The industry is facing a growing memory constraint while demand for AI-ready platforms continues to surge. Fine-tuning and inference on proprietary data require massive compute and memory resources, creating investment challenges for organizations. These rising solution costs and workflow bottlenecks are slowing time-to-market for revenue-generating innovation. To address this challenge, Phison introduced aiDAPTIV™ technology for local and edge AI use cases. By leveraging Pascari SSDs as a new AI memory tier, aiDAPTIV technology intelligently extends and manages AI working memory across GPU memory, system RAM and flash.

Today’s announcement showcases how aiDAPTIV applies these multi-tier memory architecture principles to local AI systems as NVIDIA AI infrastructure advances GPU memory capabilities to support inference workloads in data center environments. Built on high-endurance flash optimized for sustained paging and context retention, aiDAPTIV supports memory-intensive inference and fine-tuning workloads under fixed hardware configurations. The aiDAPTIV flash-based memory tier enables organizations to support these evolving workloads on local systems while maintaining data privacy and improving long-term infrastructure efficiency.

“Conventional memory management was never designed for AI, and today’s AI infrastructure can no longer rely on generic memory management,” said Michael Wu, President and GM, Phison US. “With aiDAPTIV technology, we’ve built an AI-aware architecture that extends effective memory across multiple tiers. This enables larger models and long-context inference on local AI platforms without requiring additional GPU hardware, helping organizations keep AI workloads local while planning infrastructure investments more predictably.”

At NVIDIA GTC 2026, Phison is showcasing demonstrations of partner notebooks, workstations and systems powered by NVIDIA GB10 Grace Blackwell processors as well as NVIDIA GeForce RTX 50 Series 그리고 NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition GPUs.

Demonstrations highlight long-context inference, agentic AI workflows leveraging KV cache reuse, and memory-intensive fine-tuning on large-scale models, showcasing how aiDAPTIV extends effective memory across GPU memory, system RAM and flash to support workloads that would otherwise exceed available system capacity.

To request an NVIDIA GTC 2026 meeting with Phison to receive an aiDAPTIV demo, please reach out to sales@phison.com. To explore Phison’s full portfolio, visit www.phison.com.

For more information about the announcement and solutions, visit the media kit. .

Phison Electronics 소개

Phison Electronics는 글로벌 리더입니다. NAND 플래시 컨트롤러 and storage solutions, powering more than one in every five SSDs shipped worldwide. Phison has grown into a multi-billion-dollar company with over 4,500 employees—70% of which are dedicated to R&D – and more than 2,000 patents. The company’s innovations include aiDAPTIV, an award-winning AI solution for affordable LLM training and inferencing on-premises, and Pascari, a portfolio of ultra-high-performance enterprise SSDs purpose-built for data-intensive workloads across AI, cloud, and hyperscale data centers.

Phison, Phison 디자인, Phison 로고는 미국 및/또는 기타 국가에서 Phison Electronics 또는 계열사의 등록 상표 또는 상표입니다. 다른 모든 상표는 해당 소유자의 재산입니다. 제품 사양은 사전 통지 없이 변경될 수 있습니다. 표시된 사진은 실제 제품과 다를 수 있습니다.

Disclaimer: Many of the products and features mentioned are still in development and will be made available as they are finalized. The timeline for their release is dependent on the ongoing development and market and is subject to change.