Microprocessor giant Intel (NASDAQ:INTC) has talked about the breadth of solutions that it plans to bring to market to serve as "accelerators" for key emerging workloads, such as machine learning.
One thing Intel hasn't really talked about much, though, is graphics acceleration in the data center. Graphics specialist NVIDIA (NASDAQ:NVDA) has garnered a lot of attention from investors because of the explosive growth in its data center business -- it grew 186% year over year last quarter, continuing a multi-quarter winning streak. But Intel has been silent on the topic.
After all, graphics isn't exactly Intel's strong suit today. The integrated graphics technology on its consumer processors are adequate for basic tasks, but they're neither particularly fast nor efficient.
It would seem, though, based on a recent job listing posted on Intel's website that the company, indeed, embrace GPU-based computing over the long-term.
A graphics job inside Intel's data-center group
In scanning Intel's job boards, I came across a job listing titled "Graphics and Media Architecture Lead." I know Intel builds graphics processors to integrate into its consumer processors, so I nearly skipped over the listing, as I didn't initially think there could be good insights to be gleaned from it.
I'm glad I opened it.
The listing is for a position within the Intel Datacenter Pathfinding and Architecture Group, a group that Intel says is chartered to "deliver breakthrough innovations to fuel [data-center group] growth, technology leadership and a differentiated data center roadmap across compute, storage/memory, [and] network I/O."
The job description gets even more interesting, stating that "to meet the explosive demand of data centers, it will require Intel to deliver high performance, scalable, and power efficient graphic/media solutions with the necessary [hardware] and [software] components." It also says the company is "seeking a leader that will help shape the future of [Intel's] offerings."
After that, the listing says that, among other things, the individual hired to do this job will be tasked with working to build "world class solutions utilizing GPU evolution for [General Purpose GPU] compute in computer vision, and deep neural networks."
It also doesn't seem Intel is hiring for some pie-in-the-sky pseudo-academic research position, either. The listing clearly says it wants somebody who will "understand current and future customer application needs, and then delight those customers by delivering architectures that exceed expectations."
This appears to be a very clear sign that Intel is planning to take general-purpose computing on graphics processors much more seriously than it has in the past.
Major data-center customers obviously find value in using graphics processors to accelerate key workloads, as NVIDIA's data-center graphics processor business is growing at a crazy pace -- so Intel would be wise to build products that those customers would like to buy.
For Intel to build competitive solutions, though, it's going to need to invest heavily in improving its graphics architecture; Intel's current best graphics architecture is quite poor compared with NVIDIA's current best -- and that "best" is about to be replaced with something even better in short order.
Moreover, even if Intel were to build great graphics architectures, it would still face a significant hurdle in the marketplace. The clear majority of programs for general-purpose computing on graphics processing units are written specifically for NVIDIA's CUDA platform, which has evolved significantly over the past decade or so.