Intel took a step closer to realizing its XPU vision with the announcement of its oneAPI, first teased at Super Computing 2019, and the launch of its first data center-focused GPU today.

Intel's oneAPI seeks to provide an open, standards-based programming model that allows developers to choose which accelerator to run their software. Using this framework, Intel hopes to make it easier for developers to take advantage of the efficiencies of each of the various accelerator types, whether they be CPUs, GPUs, or FPGAs. Intel calls this vision XPU.

"All of those architectures are not supported with the same languages, libraries, and tools," said Jeffrey McVeigh, VP and GM of visual computing products at Intel. This "makes it very difficult to be able to reuse code and be productive across those architectures."

Intel made good on its XPU promise today, announcing the gold release of oneAPI would begin shipping in December.

oneAPI seeks to address one of the biggest challenges faced by developers today. That is, a piece of software developed to run on a CPU has to be retooled to take advantage of GPU or FPGA acceleration.

"The CPU today is the most broadly used and adopted architecture with software," McVeigh said. " As new architectures come to life, wouldn't it be great for that software that's already existing on the CPU to be able to be brought to those architectures seamlessly and to provide full performance?"

This is what oneAPI attempts to do by acting as a low-level abstraction layer for the software stack that enables developers to write once and run on anything.

To achieve oneAPI's mission, Intel is using a common language it calls Data Parallel C++.

According to McVeigh, this common language is built on an evolution of OpenCL with specific enhancements to single-source support and advancements to the most recent ISO C++ codebase.

This is combined with optimized libraries for specific workloads, including those for deep learning, mathematics, threading, and video processing. Along with a hardware abstraction layer and a compatibility tool, these features enable developers to move from one accelerator type to another without needing new code development, retooling, or proprietary binaries.

The compatibility tool is particularly important, explains McVeigh, as it enables developers to take source code implemented for use on Nvidia's CUDA platform and translate it to Data Parallel C++.

This software stack is being made available through a series of workload-specific toolkits which run the gamut from high-performance computing, IoT, rendering, and artificial intelligence.

All of these environments will be available free of charge for developers. Additionally, Intel will provide a supported version for customers that want additional assistance implementing oneAPI.

Alongside oneAPI, Intel unveiled its HC3 XG310 data center GPU, which is being aimed at video and game streaming applications.

The XG310 features four of Intel's Xe-LP-based GPUs, each boasting 128-bit pipelines and eight gigabytes of memory in a PCIe form factor.

The announcement marks Intel's return to the GPU market and the company's first data center GPU. However, compared to Nvdia's latest generation of GPUs, Intel is taking a different tact. Rather than focusing on sheer performance, the chipmaker aims to provide customers greater GPU density in order to serve more concurrent video streams for a given watt.

"Intel Xeon Scalable processors and Intel Server GPUs offer a high-density, low-latency, low-power, low-[total cost of ownership] solution," said Allen Fang, VP and GM of Tencent's XianYou Cloud Gaming platform, in a statement.

The HC3 XG310 datacenter GPU is now shipping.