Since ArrayFire was founded in 2007, there has been an explosion in software and its importance to our lives. Computers, connected to sensors and real-world outcomes, do really cool things that touch nearly every aspect of our lives.
I believe these are exciting times for technical computing and for HPC, as evidenced by the things showcased this week at SC 2020.
While ArrayFire focuses purely on software, our hardware partners turn our imaginative lines of code into real-world applications. AMD, NVIDIA, and Intel have each evolved tremendously since we started ArrayFire.
Over a decade ago, NVIDIA and its CEO-founder Jensen saw the opportunity to teach the world a new heterogeneous model of computing that overwhelmingly convinces scientists, engineers, and analysts to adopt GPU computing. We were an integral part of the early CUDA days, and we remember the energy at NVIDIA as GPU computing came to life.
What a gamble GPU computing was at the time, and what a tremendously beneficial outcome it has produced for applications of all types!
Since 2007, NVIDIA has transformed its hardware company to an enterprise focused on software and applications. CUDA is doing great things and is an integral part of getting AI done. ArrayFire customers of all sorts are finding big success with NVIDIA GPUs for their applications.
AMD has improved its GPUs so much that Oak Ridge is tapping the company to supply both CPUs and GPUs for its upcoming Frontier supercomputer, which will be about seven times more powerful than the fastest supercomputer currently in the world. A significant contribution by AMD has been its role in promoting open standards for heterogeneous computing.
Meanwhile, Intel has seen its market share eaten by the rapid transition to heterogeneous computing, along with internally slipping on product roadmaps. From 2007 until recently, Intel seemed to be a sleeping giant slow to adapt to heterogeneous computing. It was an instance of the classic innovator’s dilemma, as Intel struggled to cannibalize the importance of the CPU to keep up with the shift to heterogeneous computing.
The thing that stood out to me the most this week at 2020 is the clear message that Intel is committed to heterogeneous computing. Its messages on the XPU were excellent. Its focus on open standards with oneAPI represents a sincere opportunity to regain and grow market and mind share.
As Intel focuses on end-user applications and productivity for developers, it will be able to foster great growth for oneAPI. This requires heavy investment. Early adopters must see enough significant value, both in hardware and in ease of programming, to take the time to look into oneAPI.
The emergence of oneAPI is nicely tied to the development of SYCL, a high-level programming model for OpenCL. Having the weight of Intel behind SYCL is great news for open standards. We will look at SYCL in more detail in another blog post coming very soon.
I’ve met with many of the folks working on oneAPI at Intel. The energy among those folks is the best energy I have felt in a long time from a hardware partner. My sense is that Intel will be a dynamic force in the coming years as it works both from a place of opportunity as well as a place of being startled from its sleepy state.
At ArrayFire, we continue to give you the best heterogeneous computing experience by incorporating the best tools from AMD, NVIDIA, and Intel. We’re biased, but ArrayFire really is a wonderful way to express your algorithms neatly, get maximal performance with a productive API, and stay immune to ever-changing hardware.
All of us working in this space are going to experience a roaring 20s in AI and technical computing. Exciting times are ahead!