What is the hardware architecture of OCI DenseIO shapes?
Oracle Cloud Infrastructure (OCI) DenseIO shapes are specifically engineered for workloads that need to crunch massive amounts of data with the lowest possible latency. While "Standard" shapes rely on network-attached Block Storage, DenseIO shapes bring the storage directly onto the server's motherboard using local NVMe drives.
Here is the architectural breakdown of the DenseIO hardware stack.
In a standard cloud instance, your data travels over the network to a storage array. In a DenseIO instance, the storage is physically inside the chassis.
Direct PCIe Connection: The NVMe SSDs are connected directly to the CPU via the PCIe bus. This bypasses the network entirely, reducing latency from milliseconds to microseconds.
High IOPS: Because there is no "network hop," a single Bare Metal DenseIO instance can achieve upwards of 3.4 million IOPS (Input/Output Operations Per Second).
Storage Density: Modern shapes (like the E5 series) provide massive local footprints—often exceeding 50 TB of raw NVMe capacity on a single node.
DenseIO shapes aren't just about storage; they are balanced with high-tier compute to prevent bottlenecks. As of 2026, the primary architectures are:
The Processor: Currently utilizes 4th Gen (Genoa) and 5th Gen AMD EPYC processors.
Memory: Massive DDR5 memory footprint (up to 2.3 TB on bare metal).
Best for: Massive NoSQL databases (Cassandra, MongoDB) and Hadoop clusters where CPU core count is critical for parallel processing.
The Processor: Utilizes 4th Gen Intel Xeon Scalable (Sapphire Rapids) or newer.
Acceleration: Includes Intel Data Streaming Accelerator (DSA) to further speed up the movement of data between the local NVMe drives and memory.
Best for: Traditional relational databases (Oracle Database, Microsoft SQL Server) and applications optimized for Intel-specific instruction sets.
You can consume DenseIO in two different ways, each with a distinct physical path:
Bare Metal (BM.DenseIO): You have direct, raw access to the physical NVMe controllers. There is no virtualization layer. This is the highest possible performance tier.
Virtual Machine (VM.DenseIO): OCI uses PCI Passthrough or SR-IOV (Single Root I/O Virtualization). This allows the VM to "see" the local physical drive with near-zero overhead, even though it's running in a virtualized environment.
Even though the data storage is local, the Off-box Network Virtualization (SmartNIC) still plays a crucial role.
Management: The SmartNIC handles all the VCN (Virtual Cloud Network) traffic, ensuring that the heavy I/O happening on the local disks doesn't interfere with the server's ability to communicate with the rest of the cloud.
Data Protection: Because the storage is local, it is ephemeral. If the physical server fails, the data on those NVMe drives is lost. OCI’s architecture encourages using high-speed networking to replicate data across multiple instances (using tools like Data Guard or HDFS replication).
| Component | Bare Metal Spec (BM.DenseIO.E5.128) |
| CPU | 128 Cores (AMD EPYC 9J14) |
| RAM | 2,304 GB DDR5 |
| Local Storage | ~54 TB NVMe SSD (Local) |
| Network | 100 Gbps (with Cluster Networking support) |
| Target Workload | Big Data, OLTP, Data Warehousing |