Why This Server Stood Out in June 2026

As of August 17, 2026, Wiwynn’s NVIDIA SCADA storage server is one of the more interesting AI infrastructure reveals of the summer because it shifts attention from accelerator count to the storage path feeding those accelerators. Tom’s Hardware reported on June 12, 2026 that Wiwynn showed one of the first NVIDIA SCADA systems, described as a GPU-accelerated storage server built for heavy AI data movement rather than conventional file serving. The headline spec was scale: up to 96 liquid-cooled SSDs using currently available E3.S drives, with petabyte-class capacity that Tom’s Hardware described as reaching 2.9PB. (tomshardware.com)

Article contains affiliate links, commission may be earned.

What NVIDIA SCADA Changes in the I/O Path

NVIDIA’s SCADA here does not mean industrial supervisory control software. In this context, NVIDIA describes SCADA as Scaled Accelerated Data Access, a storage architecture where GPUs can initiate and control storage operations directly instead of waiting for the CPU to manage both control and data movement. NVIDIA’s own material frames SCADA as a way for massively parallel GPUs to pull only the needed data from storage into high-speed GPU memory, which is especially relevant when models, embeddings, graph data and retrieval pipelines become too large to sit neatly in memory. (blogs.nvidia.com)

View NVIDIA DGX Spark Personal AI Desktop Supercomputer on partner website

Hardware Snapshot: GPUs, SSDs, Switches and Networking

The Wiwynn system is best understood as a dense AI storage appliance rather than a general-purpose server. The configuration reported around the Computex 2026 showing includes:

  • Up to 96 E3.S NVMe SSDs, liquid-cooled for high-density deployment
  • NVIDIA Vera CPU for platform-level compute and orchestration duties
  • Four NVIDIA RTX PRO 6000 Blackwell cards
  • Four PCIe 6.x switches to connect the GPU and storage fabric
  • Four NVIDIA ConnectX-9 SuperNIC cards for high-speed networking

NVIDIA lists the RTX PRO 6000 Blackwell Server Edition with 96GB of GDDR7 memory, making the GPUs more than simple compute add-ons in this design; they become active participants in storage access. (tomshardware.com)

See ASUS GeForce RTX 5090 price

Why Petabyte Racks Need More Than Raw Capacity

The useful part of Wiwynn’s SCADA system is not just that it can hold a large pool of flash. AI workloads increasingly hit storage in less tidy ways than classic sequential training reads: retrieval-augmented generation, graph neural networks, vector search, recommender systems and document ingestion can all lean on small, frequent, latency-sensitive reads. Wiwynn’s own GTC 2026 overview tied its Storage-Next work to a GPU-initiated 96-drive NVMe array and pointed to uses such as GNN, LLM inference and RAG. That makes the server a sign of where rack design is going: storage is becoming part of the accelerator fabric, not just a back-end tier. (wiwynn.com)

Buy Crucial T705 13 GB/s NVMe SSD here

PCIe 6.0 Points to the Next Storage Transition

Wiwynn’s prototype also lands in the middle of the broader PCIe 6.0 shift. Silicon Motion said at Computex 2026 that AI platforms are shaping next-generation SSD controller plans, with its enterprise SM8466 PCIe Gen6 controller positioned for AI infrastructure and enterprise storage before mainstream client adoption. That matters because SCADA-style designs only make sense when the SSDs, PCIe switches, GPUs and network adapters can all move data at rack-scale rates. Wiwynn’s server does not make every AI bottleneck disappear, but it does show a practical direction: future petabyte racks may be judged as much by how directly GPUs can reach data as by how many GPUs they contain. (tomshardware.com)

View SANDISK Optimus GX PRO 8100 PCIe 5 SSD on partner website