Intel’s Open OneAPI and Cloud: Powering Heterogeneous Computing

Intel recently expanded its commitment to fully open system, software, and hardware standards, inviting Linux founder Linus Torvalds to discuss open ecosystems and open‑source.

Torvalds recalled creating Linux on a 386‑based PC in 1991 and emphasized that developing software for x86 machines remains crucial in 2022.

Open Commitment

Intel CEO Pat Gelsinger highlighted the industry’s potential when openness, choice, and trust are embraced.

Torvalds noted that while early Linux was compiled for 386 CPUs, today’s workloads shift AI and data analysis to GPUs, FPGAs, and ASICs, which outperform CPUs for such tasks.

At the two‑day Intel event, Gelsinger promoted the concept of “open accelerated computing,” allowing developers to write code without worrying about specific chips.

Intel announced the 2023 version of the oneAPI programming toolkit, which automatically targets CPUs, GPUs, and other accelerators, relieving developers of hardware‑specific concerns.

CTO Greg Lavender described a single universal programming language that runs everywhere, likening it to a breakthrough similar to Java.

Intel employs roughly 17,000 software engineers and is building software services on its chips.

Earlier this year Intel acquired Codeplay, known for its SYCL parallel programming model, which enables standard C++ code to run across CPUs, GPUs, and other processors.

The oneAPI compiler, based on Codeplay’s SYCL, generates standard C++ source for Intel and other vendors’ GPUs, CPUs, and FPGAs.

OneAPI 2023

The 2023 release includes 42 tools and supports the fourth‑generation Sapphire Rapids CPUs.

Intel plans to launch a new Sapphire Rapids chip early next year, delayed due to verification issues.

OneAPI can automatically migrate CUDA code to SYCL C++ via the SYCLomatic tool, described as “a washing machine that cleans proprietary code into an open form.”

Intel has opened OneAPI governance, allowing contributors to vote on specifications.

Joe Curley noted that OneAPI is still early‑stage but community‑driven, with Intel remaining the governing body.

OneAPI specifications comprise the Khronos SYCL language, standardized library bindings, and developer tools.

Intel Developer Cloud

Intel announced the Intel Developer Cloud, offering cloud‑based access to upcoming Intel chips such as Sapphire Rapids and Gaudi 2 AI processors, enabling developers to collaborate and have software ready when hardware ships.

The cloud provides the latest oneAPI and SYCL layers, facilitating deployment of standard C++ applications in heterogeneous environments.

It also helps developers create cloud‑native applications that can run on Google Cloud, AWS, or Azure using future Intel chip instances.

Differentiated IP and Multi‑ISA Support

Intel aims to mix ARM, x86, and RISC‑V cores in a single package, leveraging SYCL so code runs regardless of the underlying architecture.

Gary Martz highlighted Intel’s differentiated IP, software services, multi‑ISA strategy, and a unified API.

While parallelism is not a first‑class citizen in standard C++, Intel believes SYCL offers advanced acceleration capabilities.

Project Amber

Intel introduced Project Amber, a confidential computing service that verifies data integrity between devices using cryptographic codes before allowing data into a secure execution region.

The technology is crucial for AI and machine‑learning workloads that ingest data from multiple sensors, ensuring the data’s authenticity.

Amber’s new feature, the TDX instruction, creates an encoded secure VM layer in upcoming Sapphire Rapids chips, removing the hypervisor from the trust boundary during VM entry and exit.

TDX enables a consistent proof‑of‑concept across hybrid‑cloud environments and multiple cloud providers.

OneAPI remains compatible with languages and models such as Python, OpenMP, and Fortran.

Other Announcements

Intel unveiled Getti, a computer‑vision AI suite that helps developers quickly build AI models, demonstrated by a partnership with the Royal Brompton Hospital to detect rare respiratory diseases without AI expertise.

Project Amber’s confidential computing service also supports secure data handling across distributed cloud nodes.