Nvidia's high-speed AI data center storage servers break cover, touting 2.9 petabytes of storage and extreme PCIe 6.0 performance — Wiwynn shows off SCADA server with GPU-accelerated storage

Industry-First: Wiwynn's Nvidia SCADA Server Packs 2.9 PB of GPU-Accelerated Storage

At Computex Taipei 2026, Wiwynn unveiled one of the industry's first production-ready Nvidia SCADA (SCaled Accelerated Data Access) servers — a groundbreaking storage platform designed to eliminate CPU bottlenecks in AI data pipelines. The server delivers up to 2.9 petabytes of flash storage with 528 million random read IOPS, powered by four RTX Pro 6000 Blackwell GPUs that handle storage I/O directly, bypassing the CPU entirely.

What Is SCADA and Why Does It Matter?

Nvidia's SCADA architecture, first previewed in late 2025, is a key pillar of the company's Storage Next vision — a collection of technologies aimed at making storage behave more like an extension of GPU memory. Modern AI inference and training workloads routinely deal with massive datasets that exceed the memory capacity of any single AI accelerator. Tasks such as vector search, retrieval-augmented generation (RAG), graph analytics, and KV-cache retrieval rely on fine-grained random accesses with extreme parallelism, often involving data blocks smaller than 4KB.

Traditional CPU-centric I/O cannot efficiently handle these workloads. Even advanced solutions like GPU Direct Storage still leave the CPU in control of the data path, creating a bottleneck. SCADA eliminates the CPU from storage transactions entirely: GPUs themselves initiate and control storage I/O operations, managing millions of small storage requests on behalf of AI applications.

Wiwynn's SCADA Server: By the Numbers

Wiwynn's implementation is an Nvidia MGX rack-compliant 6RU system with a maximum power draw of 9 kW. The key specifications include:

• Up to 96 liquid-cooled E3.S SSDs — cooled by six dedicated cold plate modules integrated into the system's liquid cooling loop, ensuring consistent thermal performance under sustained load.
• 2.949 PB maximum capacity — achieved with 96 Micron 9650 Pro 30.72 TB drives using a PCIe 6.0 interface.
• 528 million 4K random read IOPS — aggregated across all drives, limited only by the PCIe fabric and network controllers.
• Four RTX Pro 6000 Blackwell GPUs — serving as sophisticated storage processors that initiate and handle storage transactions on behalf of compute servers.
• Four PCIe 6.x switches and ConnectX-9 SuperNIC cards — providing the high-bandwidth fabric to move data between storage and compute nodes.

GPU-as-Storage-Processor Architecture

The architecture represents a paradigm shift in data center storage design. In the SCADA topology, Nvidia positions these machines as tier 3.5 storage servers — sitting behind local SSDs but ahead of tier 4 remote storage. The RTX Pro 6000 GPUs act as intelligent storage processors that:

• Initiate and manage millions of concurrent storage I/O requests
• Pass data to upstream compute servers via the ConnectX-9 SuperNIC cards
• Offload the CPU from storage management entirely

The SSDs and their controllers continue to perform the actual data storage functions, but the control path runs through GPUs rather than CPUs. This design is particularly effective for AI inference workloads where latency and parallelism are critical.

Liquid Cooling for Dense Storage

A notable engineering achievement is the liquid cooling solution for the drives. Traditional air cooling struggles to maintain consistent SSD temperatures in a dense 96-drive configuration under sustained AI workloads. Wiwynn's approach uses six separate cold plate modules that inject coolant simultaneously to all SSDs, ensuring uniform thermal performance and preventing thermal throttling that would otherwise degrade IOPS under load.

Scalability and Future-Proofing

Because SCADA is an open platform — Nvidia has released its SCADA specification to partners like Broadcom and Micron — customers can build their own SCADA machines with commercially available components. As SSD manufacturers increase the capacities and performance of E3.S form-factor drives, servers like Wiwynn's will scale naturally. The platform's reliance on PCIe 6.0 ensures bandwidth headroom for next-generation storage media.

Pricing and Availability

Wiwynn has not disclosed pricing, which depends on multiple factors including SSD and DRAM market prices, GPU availability, and purchase volumes. However, a system equipped with four RTX Pro 6000 Blackwell cards and up to 96 enterprise SSDs will be a significant investment targeted at hyperscale AI operators and large enterprise data centers.

Bottom Line

The Wiwynn SCADA server marks a concrete step toward Nvidia's vision of disaggregated, GPU-orchestrated storage for AI. By eliminating CPU bottlenecks and providing direct GPU-to-storage access at massive scale, these systems address one of the most pressing infrastructure challenges in modern AI: getting data to GPUs fast enough to keep them fed.