banner

Networking
Services

Get the best RDMA Networking purposely build for AI

InfiniBand Networking

Quantum InfiniBand is a high-performance interconnect solution primarily developed by NVIDIA for AI workloads. InfiniBand networking is used to support next-generation performance and scalability. We built our NVIDIA Quantum InfiniBand Cluster Networking with QM9700 and QM9790

Low Latency

Ultra-High Bandwidth and Low Latency

We use 400G none-blocking InfiniBand Networking to connect across all compute network, the latency is in the microsecond range

Computing

In-Network Computing

SHARP (Scalable Hierarchical Aggregation and Reduction Protocol): Offloads collective operations like reductions into the network fabric, greatly speeding up performance for distributed AI training

QoS

Congestion Control and QoS

We use Adaptive Routing and Explicit Congestion Notification ensure traffic flows smoothly even under heavy load

Tolerance

Reliability and Fault Tolerance

We design our Infiniband networking with Enhanced error detection and correction with support for failover and redundancy

Scalability

Scalability

With up to 15% higher effective throughput than alternatives, run your GPU clusters at their maximum potential and accelerate time to market

RoceV2 Networking

Create virtual, accelerated networks to manage your cloud resources on CanopyWave—powered by NVIDIA BlueField-3 DPUs. Securely and efficiently connect compute, storage, and everything else for GenAI

High Performance

High Performance

RDMA bypasses the CPU for data transfers, reducing latency and CPU overhead

Workloads

Excellent for AI Workloads

Used in distributed AI training systems where fast GPU-to-GPU communication across nodes is needed

Scalability

Scalability

Supports multi-tenant cloud environments

Advanced Congestion Control

Advanced Congestion Control

We build our RoceV2 Networking resilience to congestion and packet loss

Incredibly Fast Networking
Services For GenAI

Don’t let a shoddy network slow you down. Get the absolute most out of your GPUs