Open for all architectures and vendors, SYCL* and oneAPI are helping usher in a new golden age for computer architecture with performance and portability benefits that break down barriers to innovation.
SYCL and oneAPI: Open Foundations for the Future of Accelerated Computing
James Reinders, Intel
The hardware accelerators of the future will be incredibly diverse. From accelerators developed for AI and security, to new frontiers such as blockchain, neuromorphic, and graph acceleration, the innovations of tomorrow will demand open foundations like SYCL and oneAPI.
These are critical in the drive toward openness because they are designed from the ground up to be open for all vendors and all architectures. This forms a far better foundation for software than proprietary, single-vendor solutions that stifle innovation and limit our choices.
That’s why today, instead of teaching SYCL and oneAPI, I am focusing on the significant role these programming models should play as the foundation for all software — and why that matters so much.
In 2018 two famous computer architects, John Hennessy and David Patterson, announced that we had entered “A New Golden Age for Computer Architecture.” This new age is bringing exciting innovations from hardware vendors across the world — not just Intel!
Soon, every compute device on the planet will have access to multiple acceleration resources.
The future will include GPUs from many different vendors, as well as domain-specific accelerators for security, data processing, deep learning, neuromorphic computing, blockchains, discovery, and much more.
That’s why the accelerator-related software choices we make today are so important. They will either empower us to harness this incredible future, or cause us to be locked out of it.
We’ve already seen impressive results with CUDA-based applications that were migrated to C++ with SYCL and oneAPI support. As opposed to proprietary programming models, SYCL and oneAPI keep applications open for the future, as well as open the door for higher performance and enhanced portability.
Have we figured all of this out and gotten it all perfect? Definitely not. But that’s the beauty of community-based development: we will advance together with a shared vision of staying open to supporting the entire industry.
We recognize that Intel is in a unique position to help make a multivendor, multiarchitecture future a reality. We are committed to helping ensure customers and developers have choices, and our support of SYCL and oneAPI as the foundational software reflects that commitment.
Increasingly, no computer will do all computation on a single architecture from a single vendor. Instead, devices will be made of multiple chiplets that come together to form a system. These advances result in nearly endless possibilities. In the coming years, Intel will design its own chiplets, as well as open its fabs to manufacture chiplets from other vendors, and combine them with other architectures. It is important to Intel, our customers, and all computer users that open ecosystems prevail.
I invite you to dig into this topic, so you can see for yourself how limited the future will be if we do not ensure it is truly open to all.
We can ask ourselves: How much do we value the ability to take advantage of future innovations, wherever they may arise?
The foundational choices we make today will impact us for decades.
We must compare being open against choices that limit our future by favoring one architecture or vendor over another — certainly, a perilous choice in this fast-moving world.
To learn more, use the QR code link here or listed below to find additional resources about how SYCL and oneAPI can play a foundational role for the future of C++ as well as other languages and libraries.
By working together, we can promote a diverse, competitive, and exciting future for computers, software programmers, and the world.