D-Matrix Wires Raptor XPUs Into NVIDIA's NVLink Fusion Rack Network
CEO Sid Sheth says the deal cuts XPU-to-XPU latency threefold and multiplies packet rates tenfold over standard Ethernet.
- NVLink Fusion slots Raptor XPUs into NVIDIA's liquid-cooled MGX rack architecture, beside its own Vera CPU.
- Sixth-generation NVLink moves 3 TB/s of all-to-all bandwidth per XPU inside the rack.
- ConnectX-9 SuperNICs, BlueField-4 DPUs and Spectrum-X Ethernet carry the traffic between racks.
Why it matters: d-Matrix's chip now depends on NVIDIA's rack, CPU and network standards to reach any customer.