Interactive AI Interconnect Topology
Real-time multi-tier Dragonfly+ / Clos fabric packet simulation
Topology:
Radix-128 Clos
GPU Cluster Scale
16,384 GPUs
Per-Link Interconnect Bandwidth
800 Gbps
Target Switching Hop Latency
450 ns
Fabric Architecture Paradigm
Funding Allocation & Telemetry
$205M Series Funding deployment and training efficiency
Comm. Overhead
4.2%
All-Reduce idle penalty
Effective Throughput
842,050.0
TFLOPS sustained FP8
Bisection B/W
13.10
Petabits / sec total
Scaling Efficiency
95.8%
vs linear compute
$205.0M Strategic Capital Deployment
$205,000,000 Total
R&D (ASIC/Optical)
$123.0M (60%)
Foundry Scaling
$61.5M (30%)
Operations / GTM
$20.5M (10%)
Fabric Architecture Benchmark Comparison
| Architecture | Collective Latency | Loss Recovery | Simulated Overhead |
|---|---|---|---|
| Omni-Path (Cornelis) | Cut-Through / Hardware | Packet-level Dispersal | 4.2% |
| InfiniBand NDR | Credit-based Flow | Go-Back-N Retransmit | 5.8% |
| Ultra Ethernet (RoCEv2) | PFC / ECN Congestion | Selective ACK Retransmit | 8.9% |
State persists locally in real time. Download full cluster matrix and telemetry below.