新川サーバー | エンタープライズ サーバー ハードウェア サプライヤー
Your GPU Cluster Is Sitting Idle. The Problem Is Not the GPUs.
You have eight XE9680 servers, each with 8x H100 GPUs. That is 64 H100s. Your training job should finish in 12 時間. Monitoring shows GPU utilization at 45%. The bottleneck is not compute — it is the 100GbE fabric between the nodes. Every gradient synchronization step stalls waiting for the network. That fabric is where the NDR200 InfiniBand adapter earns its place.
The Mellanox ConnectX-7 is a single-port 400Gb/s InfiniBand NDR200 (or 400GbE Ethernet) adapter built for the inter-node communication that multi-GPU AI training demands. With RDMA, GPUDirect, and in-network computing acceleration, it moves data between GPU memory and the network fabric without touching the host CPU or system memory.
技術仕様
| パラメータ | 仕様 |
|---|---|
| モデル | Mellanox ConnectX-7 (MCX755106AS-HEAT or equivalent) |
| Port Configuration | 1x QSFP112 (400ギガビット/秒) or 2x QSFP56 (200ギガビット/秒), configurable |
| InfiniBand | NDR200 (400 Gb/s per port), backward-compatible with HDR100/EDR/FDR |
| Ethernet | 400運ぶ / 200運ぶ / 100運ぶ / 50運ぶ / 25運ぶ / 10運ぶ (RoCE v2) |
| Host Interface | PCIe 5.0 x16 (512 Gb/s theoretical) |
| RDMA | InfiniBand RDMA, RoCE v2, GPUDirect RDMA, GPUDirect Storage |
| Message Rate | まで 215 million messages per second |
| Latency (IB) | < 600 nanoseconds (port-to-port) |
| In-Network Computing | SHARP (Scalable Hierarchical Aggregation and Reduction Protocol), MPI Allreduce offload |
| 暗号化 | Inline TLS/DTLS, IPsec, MACsec (hardware-accelerated, line rate) |
| CPU Offload | RDMA, NVMe-oF, SR-IOV, VirtIO, vDPA |
| NVIDIA Networking | NVIDIA Magnum IO, NCCL, UFM (Unified Fabric Manager) compatible |
| フォームファクタ | Low-profile PCIe, FHHL bracket included |
| 力 | ~25W typical, ~38W max |
| Physical | 167 バツ 69 んん (standard LP card) |
| 温度 | 0°C to 55°C |
| OSサポート | RHEL 8.6+/9.x, Ubuntu 20.04/22.04/24.04, SLES 15 SP4+, ヴイエムウェア ESXi 8.0+, Windowsサーバー 2022/2025 |
| 保証 | 3-year (メーカー) |
NDR200 InfiniBand vs 400GbE Ethernet — Which Fabric for AI?
| Decision Factor | NDR200 InfiniBand | 400GbE RoCE v2 |
|---|---|---|
| Tail latency (P99.9) | Consistent, predictable. In-network congestion control. | Varies with DCQCN/ECN tuning. Requires expertise. |
| GPU Direct RDMA | Fully supported and optimized (NVIDIA Magnum IO stack) | Supported but requires careful buffer registration. |
| SHARP (in-network allreduce) | サポートされています. Reduces gradient sync time by up to 30%. | 利用不可. Allreduce runs on host CPUs. |
| Switch ecosystem | NVIDIA Quantum-2 NDR switches. Managed fabric (UFM). | Any 400GbE switch (Arista, シスコ, NVIDIA Spectrum-4). |
| Learning curve | Higher. Requires InfiniBand subnet manager and fabric management. | Lower. Familiar Ethernet tooling, IP-based routing. |
| Best for | Dedicated AI training clusters, HPC, GPU-intensive workloads. | Converged data center fabrics, mixed AI+enterprise workloads. |
| Cost (per port, adapter + スイッチ) | プレミアム. Higher per-port cost. | Lower. Ethernet ecosystem scale economics. |
Compatible Server Platforms
PCIe 5.0 x16 slot required for full bandwidth. Compatible with Dell PowerEdge R770, R7715, XE9680, xFusion G5500 V7, 2488H V7, 5885H V7, and any server with a PCIe 5.0 x16 electrical slot and adequate chassis airflow. Single-port QSFP112 configuration requires a QSFP112 passive copper DAC or active optical cable to the switch.
新川経由のソース
We supply ConnectX-7 adapters pre-installed in Dell and xFusion servers, or as standalone upgrade kits. Factory-direct pricing, 3-年保証, そして世界的な発送.
