BIOS Settings for Bare Metal Instances

You can disable cores to use fewer cores than the full size of the shape. The instance itself is billed for the full shape, regardless of whether all cores are enabled.

The following options are available:

• Use 25% of available cores
• Use 50% of available cores
• Use 75% of available cores
• Use 100% of available cores

The system rounds up the number of cores across processors and provisions an instance with a whole number of cores.

Used to optimize performance for workloads that are highly tuned for performance and have significant memory access. Non-uniform memory access (NUMA) configures how the memory is interleaved between cores and memory channels in the CPU.

NUMA is a computer memory design used in multi-core CPUs. With NUMA, the time it takes to access the memory depends on the physical location of the memory relative to the CPU. CPUs have memory channels that are connected to the memory modules (called DIMMs). The NUMA setting configures how the processor cores access the memory channels, and thereby the memory, on the CPU.

With the default NUMA setting, memory is interleaved across all channels of the CPU. Because of the location of the memory channels relative to the cores, this results in different access times for different memory locations. For most workloads, the difference does not have an effect: the difference is typically in nanoseconds, and is negligible compared to the software running on the CPU.

For high-performance computing (HPC) applications that are memory sensitive and highly tuned for performance, you can get predictable performance by configuring the NUMA settings. For example, you can choose a NUMA setting that uses only the memory that's closer to the core, resulting in higher memory bandwidth and lower memory latency.

The NUMA settings that are available depend on the processor type.

Lets you configure whether a single core (OCPU) allows multiple independent hardware execution threads. Simultaneous multithreading (SMT) is also called symmetric multithreading or Intel Hyper-Threading.

Intel and AMD processors have two hardware execution threads per core. SMT permits multiple independent threads of execution per core. In many cases, multithreading lets the instance better use the resources and increase the efficiency of the CPU.

When you disable multithreading, only one thread can run on each core. This can provide higher or more predictable performance for some workloads, such as high-performance computing (HPC) workloads with many floating-point operations. Disabling multithreading can also provide better performance for some older versions of Windows that have issues with larger core counts.

Access control service lets the platform enforce PCIe device isolation, required for VFIO device pass-through. You can enable or disable access control service.

Virtualization instructions include Secure Virtual Machine for AMD shapes or VT-x for Intel shapes. You can enable or disable virtualization instructions.

Lets you control whether I/O memory access goes through the input-output memory management unit (IOMMU). You can enable or disable the IOMMU.

When the IOMMU is enabled, the IOMMU can isolate user space applications from untrusted code running on the physical host. For bare metal shapes with remote direct memory access (RDMA) networks, by default, I/O memory access that travels over the RDMA network bypasses the IOMMU, going directly to the cluster network interface card (NIC) for higher performance.

You can customize the BIOS settings when you create a bare metal instance. The settings cannot be changed after you create the instance.