Leadership Across the Compute Spectrum
The range of computing applications today is incredibly varied—and it’s growing more so, especially with the proliferation of data, edge computing, and artificial intelligence. However, different workloads require different types of compute.
Intel is uniquely positioned to deliver a diverse mix of scalar, vector, matrix, and spatial architectures deployed in CPU, GPU, accelerator, and FPGA sockets. This gives our customers the ability to use the most appropriate type of compute where it’s needed. Combined with scalable interconnect and a single software abstraction, Intel’s multiple architectures deliver leadership across the compute spectrum to power the data-centric world.
- Scalar architecture typically refers to the type of workloads that are optimal on a CPU, where one stream of instructions operates at a given rate typically driven by CPU clock cycles. From system boot and productivity applications to advanced workloads like cryptography and AI, scalar-based CPUs work across a wide range of topographies with consistent, predictable performance.
- Vector architecture is optimal for workloads, which can be decomposed into vectors of instructions or vectors of data elements. GPUs and VPUs deliver vector-based parallel processing to accelerate graphics rendering for gaming, rich media, analytics and deep learning training and inference. By scaling vector architectures from client, data center, and the edge, we can take parallel processing performance from gigaFLOPS to teraFLOPS, petaFLOPS, and exaFLOPS.
- Matrix architecture derives its name from a common operation typically performed for AI workloads (matrix multiplication). While other architectures can execute matrix multiply code, ASICs have traditionally achieved the highest performance implementing the type of operations typically needed for AI inferencing and training, including matrix multiplication.
- Spatial architecture is a special architecture usually associated with an FPGA. Here, the data flows through the chip, and the computing operation performed on the data element is based on the physical location of the data in the device. The specific data transformation algorithm that has been programmed into the FPGA.
Scalar Focused: Versatile, General Purpose
From system boot to productivity applications to advanced workloads like cryptography and AI, most computing needs can be covered by scalar-based central processing units, or CPUs. CPUs work across a wide range of topographies with consistent, predictable performance.
Intel delivers two world class microarchitectures for CPUs with the Efficient-core microarchitecture and the Performance-core microarchitecture. These microarchitectures are at the center of the various lines of CPUs for Intel from low TDP Mobile devices to powerful Xeon® based data centers. Our scalable range of CPUs gives customers the choice to balance performance, power efficiency, and cost.
Vector Focused: Highly Parallel Processing
Graphics processing units, or GPUs, deliver vector-based parallel processing to accelerate workloads such as real-time graphics rendering for gaming. Because they excel at parallel computing, GPUs are also a good option to accelerate deep learning and other compute intensive workloads.
Matrix Focused: Accelerators and New CPU Instructions
From the data center to edge devices, AI continues to permeate all aspects of the compute spectrum. To that end, we’ve developed purpose-built accelerators and added microarchitectural enhancements to our CPUs with new instructions to accelerate AI workloads.
An application-specific integrated circuit (ASIC) is a type of processor that are built from the ground up for a precise usage. In most cases, ASICs will deliver best-in-class performance for the matrix compute workloads it was designed to support.
Intel is extending platforms with purpose-built ASICs that offer dramatic leaps in performance for Matrix applications. These include Habana AI processors and the Ponte Vecchio High Performance Compute GPUs with new XMX (Xe Matrix Extensions) technology. Each XMX engine is built with deep systolic arrays, enabling Ponte Vecchio to have significant amounts of both vector and matrix capabilities in a single device.
In addition, Intel® Deep Learning Boost (Intel® DL Boost), available on 3rd Gen Intel® Xeon® Scalable processors and 10th Gen Intel® Core™ processors, adds architectural extensions to accelerate Vector Neural Network Instructions (VNNI). In order to dramatically increase the instructions per clock (IPC) for AI applications, we have introduced a new technology called Intel® AMX (Advanced Matrix Extensions). This technology will first be available as part of our next generation Sapphire Rapids architecture that significantly increases matrix type operations.
Spatial Focused: Reprogrammable FPGAs
Field programmable gate arrays, or FPGAs, are integrated circuits that can physically manipulate how their logic gates open and close. The circuitry inside an FPGA chip is not hard etched—it can be reprogrammed as needed.
Intel® FPGAs provide completely customizable hardware acceleration while retaining the flexibility to evolve with rapidly changing computing needs. As blank, modifiable canvases, their purpose and power can be easily adapted again and again.
At Intel, we’re planning for the architectures of the future with research and development in next-generation computing. Among these are quantum and neuromorphic architectures.
Six Pillars of Technology Innovation for the Next Era of Computing
Intel is innovating across six pillars of technology development to unleash the power of data for the industry and our customers.
Notices & Disclaimers2 3 4
Product and Performance Information
Based on internal Intel estimates.
All product plans, and roadmaps are subject to change without notice. Statements on this webpage that refer to future plans or expectations are forward-looking statements. These statements are based on current expectations and involve many risks and uncertainties that could cause actual results to differ materially from those expressed or implied in such statements. For more information on the factors that could cause actual results to differ materially, see our most recent earnings release and SEC filings at www.intc.com.
Code names are used by Intel to identify products, technologies, or services that are in development and not publicly available. These are not "commercial" names and not intended to function as trademarks.
All information provided here is subject to change without notice.