SIXUNITED × Jiangbolong iSA: Equipping Edge AI with “Thinking” Storage
Release time:
2026-07-17
As large-scale models accelerate their deployment on edge devices, AIPCs and AI workstations are transitioning from “running models” to “running intelligent agents.” However, with model parameter sizes continuing to grow and context lengths expanding, the bottlenecks of edge‑side AI are no longer limited to the peak computational power of CPUs, GPUs, or NPUs. Instead, memory capacity, data bandwidth, storage throughput, and multi‑task scheduling capabilities are collectively shaping the actual user experience.
Especially since the Mixture of Experts (MoE) large‑model architecture has become mainstream, model parameter sizes have grown rapidly, KV caches continue to swell, and GPU memory pressure has surged. As a result, massive amounts of data must be frequently transferred among storage, memory, and compute units. Traditional SSDs, serving as passive data carriers, can no longer meet the low‑latency, high‑throughput, and intelligent scheduling requirements of on‑device AI inference.
The next phase of competition in edge AI will no longer hinge on who offers higher computational power, but rather on who can first achieve seamless synergy between storage and computation.
SIXUNITED joins hands with Jiangbolong to develop a compute-in-memory solution for edge AI.
As a full-stack AI infrastructure service provider, SIXUNITED has been steadily advancing its integrated “hardware + software + AI agent” strategy in recent years. Its product lineup spans various types of terminals, including AI PCs, Mini PCs, AI workstations, and servers, while it has also built a comprehensive AI application ecosystem centered around its proprietary AI agent platform, OpenClaw.
To further unlock the inference capabilities of edge-side AI, SIXUNITED has established a deep technical partnership with the semiconductor storage brand Longsys, integrating Longsys’ iSA (Intelligent Storage Agent) into SIXUNITED’s AI terminal product line and building a comprehensive storage-computing collaborative architecture on the device side.
Unlike traditional SSDs, which are limited to data read and write operations, iSA elevates storage capabilities to an AI inference node with intelligent scheduling, enabling storage to actively participate in large‑model inference and delivering more efficient data flow for edge‑side AI.
This collaboration also means that SIXUNITED can not only provide AI terminals but also offer a comprehensive solution covering computing power, storage, and intelligent applications.
Storage is now integrated into inference, enabling more efficient AI edge computing.
iSA’s core capabilities stem from its integrated 5nm Storage Processing Unit (SPU). By offloading portions of the MoE expert layer to the storage side and leveraging an intelligent KV cache management mechanism, it can significantly reduce GPU memory consumption, minimize I/O data transfers, and enhance model inference efficiency.
For edge devices, this means that larger models can be deployed locally, longer context lengths can be handled with stable performance, and sophisticated AI agents can deliver smoother, more responsive interactions across multi-turn dialogues, knowledge retrieval, code generation, and other use cases.
On the AI terminal platform built by SIXUNITED, iSA is no longer just underlying hardware; it has become an integral part of the AI inference pipeline, achieving an upgrade in capability—from “storing data” to “participating in computation.”
Joint optimization across two platforms unlocks the full potential of large models on diverse hardware.
Centering on mainstream AI platforms, SIXUNITED has completed joint optimization for both Intel and AMD technology stacks.
On the AMD platform, the two parties conducted joint optimization based on the AMD Ryzen AI Max+ 395 platform, enabling local deployment of ultra-large models with up to 397 billion parameters. In scenarios involving an ultra-long context of 256K tokens, DRAM usage is reduced by nearly 40%, and the model deployment scale is increased to 3.2 times that of conventional approaches, providing greater computational capacity for complex applications such as ultra-long‑context processing and multi‑agent collaboration.
On the Intel platform, leveraging SIXUNITED’s AI PC and AI workstation products, combined with Intel Core Ultra processors and NPU computing resources, iSA can ensure the stable operation of large-scale local models, reducing system resource overhead while maintaining an excellent inference experience.
Test data show that, on the Intel Core Ultra X9 388H platform, with iSA enabled, the Qwen3.5‑35B model maintains an inference throughput of approximately 23–26 tokens per second under a 128K context window. Moreover, across various model sizes, the platform supports deploying models ranging from 35B to 397B parameters, offering enhanced scalability for edge‑side AI applications. For different GPU memory configurations, iSA also allows flexible adjustment of the model offloading ratio, balancing model capacity and inference efficiency while reducing GPU memory usage.
iSAvsIlama.cpp: A head-to-head showdown between models
Comparison of Inference Performance Across Multiple Model Sizes
From hardware upgrades to ecosystem collaboration, we are driving the practical deployment of on-device AI.
The AI industry has entered a new phase, shifting from model innovation to real-world application deployment. For device manufacturers, relying solely on hardware performance improvements is no longer sufficient to meet enterprise users’ comprehensive needs for on-premises deployment, secure inference, and intelligent applications.
SIXUNITED is leveraging its full-stack capabilities—spanning AI terminals, intelligent agent platforms, and industry solutions—to deeply integrate Jiangbolong iSA’s intelligent storage capabilities into its product ecosystem, fostering a more efficient synergy among storage, computing power, and AI applications, and delivering out-of-the-box edge AI solutions to enterprise customers.
From AI PCs and Mini PCs to AI workstations and enterprise‑level AI infrastructure, SIXUNITED is working hand in hand with its ecosystem partners to continuously strengthen the edge‑side AI capability foundation. As SIXUNITED Chairman Cao Yalian put it: “Only by lifting each other up and fostering symbiotic growth within the ecosystem can we reach far‑off horizons.”
As storage begins to acquire “thinking” capabilities, the realization of edge‑AI’s value will also enter a new phase.
recommend News
Sixunited Intelligent
Headquarters switchboard: 86-21-57635812
Marketing: sales@sixunited.com
Shanghai switchboard: 86-021-57699713
Shanghai Headquarters:Room 230, Building 1, No. 299,Zhongchen Road, Songjiang District, Shanghai
TikTok
X
Links: