CPU Device

The CPU plugin is a part of the Intel® Distribution of OpenVINO™ toolkit. It is developed to achieve high performance inference of neural networks on Intel® x86-64 and Arm® CPUs. The newer 11th generation and later Intel® CPUs provide even further performance boost, especially with INT8 models. For an in-depth description of CPU plugin, see:

• CPU plugin developer documentation.

• OpenVINO Runtime CPU plugin source files.

Note

The scope of the CPU plugin features and optimizations on Arm® may differ from Intel® x86-64. If the limitation is not mentioned explicitly, the feature is supported for all CPU architectures. CPU inference on ARM64 is not supported for Windows.

Device Name

The CPU device name is used for the CPU plugin. Even though there can be more than one physical socket on a platform, only one device of this kind is listed by OpenVINO. On multi-socket platforms, load balancing and memory usage distribution between NUMA nodes are handled automatically. In order to use CPU for inference, the device name should be passed to the ov::Core::compile_model() method:

Python

    import openvino as ov

    core = ov.Core()
    compiled_model = core.compile_model(model, "CPU")

C++

        ov::Core core;
        auto model = core.read_model("model.xml");
        auto compiled_model = core.compile_model(model, "CPU");

Supported Model Precision

CPU plugin supports the following data types as inference precision of internal primitives:

• Floating-point data types:

  • FP32 (Intel® x86-64, Arm®)

  • BF16 (Intel® x86-64)

  • FP16 (Intel® x86-64, Arm®)

  • :ref:`MXFP4 <mxfp4_support>` (Intel® x86-64)

• Integer data types:

  • INT32 (Intel® x86-64, Arm®)

• Quantized data types:

  • uINT8 (Intel® x86-64)

  • INT8 (Intel® x86-64)

  • uINT1 (Intel® x86-64)

Hello Query Device C++ Sample can be used to print out supported data types for all detected devices.

Quantized Data Types Specifics

Selected precision of each primitive depends on the operation precision in IR, quantization primitives, and available hardware capabilities. The u1/u8/i8 data types are used for quantized operations only, i.e., those are not selected automatically for non-quantized operations.

For more details on how to get a quantized model see the low-precision optimization guide.

Note

Arm® platforms execute quantized models in simulation mode: the whole model (including quantization operations) is executed in floating-point precision.

Floating Point Data Types Specifics

CPU plugin supports the following floating-point data types as inference precision of internal primitives:

• f32 (Intel® x86-64, Arm®)

• bf16 (Intel® x86-64)

• f16 (Intel® x86-64, Arm®)

The default floating-point precision of a CPU primitive is f32. To support the f16 OpenVINO IR on platforms that do not natively support float16, the plugin internally converts all the f16 values to f32, and all calculations are performed using the native precision of f32. On platforms that natively support half-precision calculations (bfloat16 or float16), the half-precision type (bf16 or f16) is automatically used instead of f32 to achieve better performance (see the Execution Mode Hint). Thus, no special steps are required to run a model with bf16 or f16 inference precision.

Important

The bf16 floating-point precision appears to have some limitations that impact the inference accuracy in LLM models. For more details, refer to this article.

Using the half-precision provides the following performance benefits:

• bfloat16 and float16 data types enable Intel® Advanced Matrix Extension (AMX) on 4+ generation Intel® Xeon® Scalable Processors, resulting in significantly faster computations on the corresponding hardware compared to AVX512 or AVX2 instructions in many deep learning operation implementations.

• float16 data type enables the armv8.2-a+fp16 extension on ARM64 CPUs, which significantly improves performance due to the doubled vector capacity.

• Memory footprint is reduced since most weight and activation tensors are stored in half-precision.

For more details about the bfloat16 format, see the BFLOAT16 – Hardware Numerics Definition white paper. To check if the CPU device can support the half-precision data type, use the query device properties interface to query ov::device::capabilities property, which should contain FP16 or BF16 in the list of CPU capabilities:

Python

core = ov.Core()
cpu_optimization_capabilities = core.get_property("CPU", device.capabilities)

C++

ov::Core core;
auto cpuOptimizationCapabilities = core.get_property("CPU", ov::device::capabilities);

Inference Precision Hint

If the model has been converted to half-precision (bf16 or f16), the ov::hint::inference_precision is set to ov::element::f16 or ov::element::bf16 and can be checked via the ov::CompiledModel::get_property call. The code below demonstrates how to get the element type:

Python

core = ov.Core()
compiled_model = core.compile_model(model, "CPU")
inference_precision = core.get_property("CPU", hints.inference_precision)

C++

ov::Core core;
auto network = core.read_model("sample.xml");
auto exec_network = core.compile_model(network, "CPU");
auto inference_precision = exec_network.get_property(ov::hint::inference_precision);

To infer the model in f32 precision instead of half-precision (bf16 or f16) on targets with native half-precision support, set the ov::hint::inference_precision to ov::element::f32.

Python

core = ov.Core()
core.set_property("CPU", {hints.inference_precision: ov.Type.f32})

C++

ov::Core core;
core.set_property("CPU", ov::hint::inference_precision(ov::element::f32));

The Bfloat16 software simulation mode is available on CPUs with Intel® AVX-512 instruction set that do not support the native avx512_bf16 instruction. This mode is used for development purposes and it does not guarantee good performance. To enable the simulation, the ov::hint::inference_precision has to be explicitly set to ov::element::bf16.

Note

If ov::hint::inference_precision is set to ov::element::bf16 on a CPU without native bfloat16 support or bfloat16 simulation mode, an exception is thrown.

Note

Due to the reduced mantissa size of half-precision data types (bfloat16 or float16), the resulting half-precision inference accuracy may differ from the f32 inference, especially for models that were not trained using half-precision data types. If half-precision inference accuracy is not acceptable, it is recommended to switch to the f32 precision. Also, the performance/accuracy balance can be managed using the ov::hint::execution_mode hint, see the Execution Mode Hint.

Execution Mode Hint

In case ov::hint::inference_precision is not explicitly set, one can use ov::hint::execution_mode hint to direct the run-time optimizations toward either better accuracy or better performance. If ov::hint::execution_mode is set to ov::hint::ExecutionMode::PERFORMANCE (default behavior) and the platform natively supports half-precision calculations (bfloat16 or float16) then bf16 or f16 type is automatically used instead of f32 to achieve better performance. If the accuracy in this mode is not good enough, then set ov::hint::execution_mode to ov::hint::ExecutionMode::ACCURACY to enforce the plugin to use the f32 precision in floating point calculations.

For more details and code examples, see the Precision Control.

Supported Features

Automatic Device Selection

If a system includes OpenVINO-supported devices other than the CPU (e.g. an integrated GPU), then any supported model can be executed on all the devices simultaneously. This can be achieved by specifying AUTO:CPU,GPU.0 as a target device, and adding the CUMULATIVE_THROUGHPUT parameter.

Python

    core = ov.Core()
    compiled_model = core.compile_model(model, "AUTO:CPU,GPU.0", {hints.performance_mode: hints.PerformanceMode.CUMULATIVE_THROUGHPUT})

C++

        ov::Core core;
        auto model = core.read_model("model.xml");
        auto compiled_model = core.compile_model(model, "AUTO:CPU,GPU.0", ov::hint::performance_mode(ov::hint::PerformanceMode::CUMULATIVE_THROUGHPUT));

For more details, see the Automatic Device Selection.

Multi-stream Execution

If either ov::num_streams(n_streams) with n_streams > 1 or ov::hint::performance_mode(ov::hint::PerformanceMode::THROUGHPUT) property is set for CPU plugin, then multiple streams are created for the model. In case of CPU plugin, each stream has its own host thread, which means that incoming infer requests can be processed simultaneously. Each stream is pinned to its own group of physical cores with respect to NUMA nodes physical memory usage to minimize overhead on data transfer between NUMA nodes.

For more details, see the optimization guide and thread scheduling introduction.

Note

When it comes to latency, be aware that running only one stream on multi-socket platform may introduce additional overheads on data transfer between NUMA nodes. In that case it is better to use the ov::hint::PerformanceMode::LATENCY performance hint. For more details see the performance hints overview.

Dynamic Shapes

CPU provides full functional support for models with dynamic shapes in terms of the opset coverage.

Note

The CPU plugin does not support tensors with dynamically changing rank. In case of an attempt to infer a model with such tensors, an exception will be thrown.

Some runtime optimizations work better if the model shapes are known in advance. Therefore, if the input data shape is not changed between inference calls, it is recommended to use a model with static shapes or reshape the existing model with the static input shape to get the best performance.

Python

import openvino as ov

core = ov.Core()
model.reshape([10, 20, 30, 40])

C++

        ov::Core core;
        auto model = core.read_model("model.xml");
        ov::Shape static_shape = {10, 20, 30, 40};

        model->reshape(static_shape);

For more details, see the dynamic shapes guide.

Preprocessing Acceleration

CPU plugin supports a full set of the preprocessing operations, providing high performance implementations for them. For more details, see preprocessing API guide.

The CPU plugin support for handling tensor precision conversion is limited to the following ov::element types:

• bf16

• f16

• f32

• f64

• i8

• i16

• i32

• i64

• u8

• u16

• u32

• u64

• boolean

Model Caching

CPU supports Import/Export network capability. If model caching is enabled via the common OpenVINO™ ov::cache_dir property, the plugin automatically creates a cached blob inside the specified directory during model compilation. This cached blob contains partial representation of the network, having performed common runtime optimizations and low precision transformations. The next time the model is compiled, the cached representation will be loaded to the plugin instead of the initial OpenVINO IR, so the aforementioned transformation steps will be skipped. These transformations take a significant amount of time during model compilation, so caching this representation reduces time spent for subsequent compilations of the model, thereby reducing first inference latency (FIL).

For more details, see the model caching overview.

Extensibility

CPU plugin supports fallback on ov::Op reference implementation if the plugin does not have its own implementation for such operation. That means that OpenVINO™ Extensibility Mechanism can be used for the plugin extension as well. Enabling fallback on a custom operation implementation is possible by overriding the ov::Op::evaluate method in the derived operation class (see custom OpenVINO™ operations for details).

Stateful Models

The CPU plugin supports stateful models without any limitations.

For details, see stateful models guide.

Supported Properties

The plugin supports the following properties:

Read-write Properties

All parameters must be set before calling ov::Core::compile_model() in order to take effect or passed as additional argument to ov::Core::compile_model()

• ov::enable_profiling

• ov::hint::inference_precision

• ov::hint::performance_mode

• ov::hint::execution_mode

• ov::hint::num_request

• ov::hint::scheduling_core_type

• ov::hint::enable_hyper_threading

• ov::hint::enable_cpu_pinning

• ov::num_streams

• ov::affinity

• ov::inference_num_threads

• ov::cache_dir

• ov::intel_cpu::denormals_optimization

• ov::intel_cpu::sparse_weights_decompression_rate

Read-only properties

• ov::supported_properties

• ov::available_devices

• ov::range_for_async_infer_requests

• ov::range_for_streams

• ov::device::full_name

• ov::device::capabilities

Note

ov::affinity is replaced by ov::hint::enable_cpu_pinning. As such, it is deprecated in the 2024.0 release and will be removed in the 2025 release.

External Dependencies

For some performance-critical DL operations, the CPU plugin uses third-party libraries:

• oneDNN (Intel® x86-64, Arm®)

• Compute Library (Arm®)

Optimization guide

Multi-Threading Optimization

CPU inference will infer an input or multiple inputs in parallel on multiple logical processors.

For more details, see the thread scheduling introduction.

Denormals Optimization

Denormal numbers (denormals) are non-zero, finite float numbers that are very close to zero, i.e. the numbers in (0, 1.17549e-38) and (0, -1.17549e-38). In such cases, normalized-number encoding format does not have a capability to encode the number and underflow will happen. The computation involving such numbers is extremely slow on much hardware.

As a denormal number is extremely close to zero, treating a denormal directly as zero is a straightforward and simple method to optimize computation of denormals. This optimization does not comply with IEEE 754 standard. If it causes unacceptable accuracy degradation, the denormals_optimization property is introduced to control this behavior. If there are denormal numbers in use cases, and no or acceptable accuracy drop is seen, set the property to True to improve performance, otherwise set it to False. If it is not set explicitly by the property and the application does not perform any denormals optimization as well, the optimization is disabled by default. After enabling the denormals_optimization property, OpenVINO will provide a cross operation system/ compiler and safe optimization on all platform when applicable.

There are cases when the application in which OpenVINO is used also performs this low-level denormals optimization. If it is optimized by setting the FTZ(Flush-To-Zero) and DAZ(Denormals-As-Zero) flags in MXCSR register at the beginning of the thread where OpenVINO is called, OpenVINO will inherit this setting in the same thread and sub-thread, so there is no need to set the denormals_optimization property. In such cases, you are responsible for the effectiveness and safety of the settings.

Note

The denormals_optimization property must be set before calling compile_model().

To enable denormals optimization in the application, the denormals_optimization property must be set to True:

Python

import openvino.properties.intel_cpu as intel_cpu

core = ov.Core()
core.set_property("CPU", intel_cpu.denormals_optimization(True))
compiled_model = core.compile_model(model=model, device_name=device_name)

C++

        ov::Core core;                                                    // Step 1: create ov::Core object
        core.set_property(ov::intel_cpu::denormals_optimization(true));   // Step 1b: Enable denormals optimization
        auto model = core.read_model(modelPath);                          // Step 2: Read Model
        //...                                                             // Step 3: Prepare inputs/outputs
        //...                                                             // Step 4: Set device configuration
        auto compiled = core.compile_model(model, device, config);        // Step 5: LoadNetwork

Sparse weights decompression (Intel® x86-64)

Sparse weights are weights where most of the elements are zero. The ratio of the number of zero elements to the number of all elements is called sparse rate. Thus, we assume that sparse weights are weights with a high sparse rate. In case of sparse weights, we can store only non-zero values in memory using special storage structures, which allows us to use memory more efficiently. In turn, this can give us better performance in the high memory bound workloads (e.g., throughput scenario).

Sparse weights decompression feature allows to pack weights for Matrix Multiplication operations directly in the CPU plugin at the model compilation stage and store non-zero values in a special packed format. Then, during the execution of the model, the weights are unpacked and used in the computational kernel. Since the weights are loaded from DDR/L3 cache in the packed format this significantly decreases memory consumption and as a consequence improve inference performance.

To use this feature, the user is provided with property sparse_weights_decompression_rate, which can take values from the interval [0, 1]. sparse_weights_decompression_rate defines sparse rate threshold: only operations with higher sparse rate will be executed using sparse weights decompression feature. The default value is 1, which means the option is disabled.

Note

Sparse weights decompression feature is disabled by default since overall speed-up highly depends on particular workload and for some cases the feature may introduce performance degradations.

Code examples of how to use sparse_weights_decompression_rate:

Python

import openvino.properties.intel_cpu as intel_cpu

core = ov.Core()
core.set_property("CPU", intel_cpu.sparse_weights_decompression_rate(0.8))
compiled_model = core.compile_model(model=model, device_name=device_name)

C++

        ov::Core core;                                                              // Step 1: create ov::Core object
        core.set_property(ov::intel_cpu::sparse_weights_decompression_rate(0.8f));   // Step 1b: Enable sparse weights decompression feature
        auto model = core.read_model(modelPath);                                    // Step 2: Read Model
        //...                                                                       // Step 3: Prepare inputs/outputs
        //...                                                                       // Step 4: Set device configuration
        auto compiled = core.compile_model(model, device, config);                  // Step 5: LoadNetwork

Note

The sparse_weights_decompression_rate property must be set before calling compile_model().

Information about the layers in which the sparse weights decompression feature was applied can be obtained from perf counters log. The “exec type” field will contain the implementation type with the “sparse” particle (“brgemm_avx512_amx_sparse_I8” in the example below):

MatMul_1800         EXECUTED         layerType: FullyConnected         execType: brgemm_avx512_amx_sparse_I8 realTime (ms): 0.050000  cpuTime (ms): 0.050000

Limitations

Currently, the sparse weights decompression feature is supported with the following limitations:

1. Model should be quantized to int8 precision.

2. Feature is only supported for Matrix Multiplication operations.

3. HW target must have Intel AMX extension support (e.g., Intel® 4th Generation Xeon® processors (code name Sapphire Rapids)).

4. The number of input and output channels of the weights must be a multiple of 64.

Additional Resources

• Inference Devices and Modes

• Optimization guide

• CPU plugin developer documentation