What is maximum throughput per connection?
As data-heavy AI and high-performance computing (HPC) workloads have grown, OCI has pushed these limits to the edge of current hardware capabilities.
For businesses connecting their on-premises data centers to OCI, the throughput limits are set by the physical port you rent:
Standard High-Speed: 10 Gbps per circuit.
Enterprise Scale: 100 Gbps per circuit.
Next-Gen AI/Big Data: 400 Gbps per circuit (available in select global regions).
Aggregation: If a single 400 Gbps connection isn't enough, you can use Link Aggregation (LAG) to group multiple physical connections together, effectively reaching multi-terabit speeds for a single virtual path.
For standard "rented" virtual servers, the network bandwidth is typically tied to the number of OCPUs (cores) you allocate.
The "Flexible" Scale: On most modern Intel, AMD, and ARM shapes, OCI provides approximately 1 Gbps of bandwidth per OCPU.
The Cap: This usually scales up to a maximum of 40GBps to 50 Gbps per VM, depending on the underlying hardware generation (e.g., E5 or E6 shapes).
If your application requires the absolute maximum throughput possible without virtualization overhead, Bare Metal is the only choice.
Standard Bare Metal: These servers typically come with dual 25 Gbps or 50 Gbps NICs, offering a total of 100 Gbps of throughput.
HPC & AI Clusters: For specialized workloads (like training an LLM), OCI uses a specialized RDMA (Remote Direct Memory Access) network over Converged Ethernet (RoCE). These nodes can reach 1,600 Gbps (1.6 Tbps) of throughput per node using multiple 200Gbps or 400Gbps network interfaces.
| Connection Type | Max Throughput | Best For... |
| Standard VM (Small) | 1 Gbps – 10 Gbps | Web servers, small apps |
| Standard VM (Large) | 40 Gbps – 50 Gbps | High-traffic APIs, App servers |
| FastConnect Dedicated | 400 Gbps | Hybrid cloud, Data migration |
| Bare Metal Standard | 100 Gbps | Large Databases, File Servers |
| HPC / GPU Cluster | 1,600 Gbps | AI Training, Fluid Dynamics |
It is important to remember that these high speeds apply to internal OCI traffic or dedicated FastConnect lines. If your server is talking to the public internet, your throughput will be limited by:
The Internet Gateway: While highly scalable, the public internet rarely supports sustained gigabit speeds to a single end-user.
The Region: Traffic moving between OCI regions (e.g., London to Tokyo) will have significantly higher latency and lower effective throughput than traffic moving within the same data center hall.
If you find that a single network interface is saturated, OCI allows you to attach multiple VNICs (Virtual Network Interface Cards) to a single instance. By distributing traffic across multiple VNICs, you can sometimes bypass OS-level single-stream bottlenecks and utilize the full physical bandwidth available to the hardware.