DF-B6 extends the Datafabrix backplane platform to PCIe Gen6 — with backplane-level telemetry designed to feed our DCIM stack directly from the fabric. Engineering samples in the lab; design partners welcome.
For enterprises, this platform arrives pre-integrated inside a Datafabrix rack — you order a rack, not a board. For OEM partners licensing the design, it is available as a reference platform under our OEM Platform program.
The Gen6 Smart Backplane is engineered as a sensor platform first, an interconnect second. Every future Datafabrix DCIM capability — AI failure prediction, predictive maintenance, digital twin, autonomous operations — draws on the substrate-level telemetry only Smart Backplane hardware can surface.
Our Gen4 Fabric Backplane commercialises the interconnect layer today. The Gen6 Smart Backplane commercialises the intelligence layer tomorrow.
The next-generation flagship — designed from silicon-up for AI-scale infrastructure and native telemetry.
Next-generation PCIe Gen6 (64 GT/s per lane, PAM4 signaling) switch fabric. 2× the bandwidth of Gen4 with tighter link budgets and FLIT-mode framing for lower latency.
Multiple PCIe Gen6 x16 GPU slots designed for accelerator co-location. Direct GPU-to-storage peer transactions eliminate the CPU bottleneck.
Dense on-board NVMe storage matrix — up to 16 NVMe SSDs directly on the backplane. Optimized layout for the fastest possible GPU-to-storage data path.
Two independent CPU-board attach points let the DF-B6 host dual Xeon (or future) compute cards with full-rate PCIe Gen6 links. Enables scale-up AI server topologies.
On-board ARM Cortex MCU runs local telemetry aggregation and edge-inference models. Anomaly detection happens ON the backplane — no BMC round-trip required.
Distributed on-die temperature, lane-margin, power, and thermal-flux sensors woven into the substrate. Sub-second telemetry of every hot spot.
Engineered copper thermal planes conduct heat laterally away from hot components. Air-cooled, standard rack thermal envelope — no liquid cooling required.
TinyML models on the MCU learn the fleet's normal signature and flag drift 30+ seconds before failures. Signals surface in Datafabrix DCIM in real time.
First-party integration with Datafabrix DCIM — Conventional DCIM + Guardian + Thermal + Insight modules light up automatically the moment a DF-B6 is enumerated.
The DF-B6 is engineered to be the flagship silicon of the next-generation AI datacenter era.
Direct GPU-to-NVMe peer transactions at Gen6 speeds. Feed the fastest accelerators without CPU or PCIe bottlenecks.
Dual-CPU + multi-GPU nodes with embedded telemetry. Fleet-wide predictive maintenance built into every backplane.
Rack-scale storage disaggregation with 2× bandwidth of Gen4. Ideal for GenAI training data pipelines.
Multi-CPU HPC nodes with tight coupling between compute, memory, and storage — enabled by Gen6 fabric bandwidth.
On-backplane TinyML watches for degradation and reroutes traffic before failures cascade. Foundation for autonomous datacenters.
Every DF-B6 broadcasts rich telemetry that the Datafabrix Guardian, Thermal, Insight, and Twin modules ingest natively — no polling required.
The DF-B6 is under active development. Customers on today's DF-B4 platform will gain a smooth upgrade path.
Active PCIe Gen4 platform deployed in customer programs.
Gen6 switch fabric, embedded MCU, ThermalSense — engineering samples.
Design partner engagement, integration into DCIM software stack.
Self-monitoring, self-healing rack platforms based on DF-B6.
We're working with a small group of design partners ahead of engineering samples. If your roadmap depends on Gen6 thermal envelopes and intelligent backplane telemetry — talk to us.
