Intel has had many new irons in the fire to serve its customers and position the company for financial growth. Two of those growth areas were discrete graphics and machine learning training. Its “six pillar strategy” outlined the six things the company would focus on, two of those being expanding the definition of compute and leaning even more into software. These were smart moves. A few weeks ago, Intel announced its gold release of the oneAPI toolkit and its first discrete graphics card for the data center, the Intel Server GPU.
Gold Release of Intel oneAPI toolkits
The first announcement from Intel was the gold release of its oneAPI toolkits. oneAPI is Intel’s developer software project, which will provide developers with a unified environment that allows them to build applications across different compute architectures in a single tool. Intel can enable an experience like oneAPI primarily because of its learnings while developing and operating OpenVINO. The gold release of the toolkits includes a set of tools and libraries for building specific applications. Also, oneAPI allows developers to reuse code across different hardware like CPUs, GPUs, FPGAs, or ML accelerators. The toolkits will use new hardware capabilities and instructions like Intel AVX-512 and Intel DL Boost on CPUs. By December, the toolkits will be available for free within the Intel DevCloud. There will also be commercial versions of the toolkits available. Intel’s technical consulting engineers will support those versions.
The oneAPI project has gained the support and endorsements from industry giants like Microsoft Azure and Tensorflow. The software project is going well to date, but there is a lot more work to be done to continue building out the tools and libraries of oneAPI.
Intel’s oneAPI, or as I like to call it, “magic API,” is the company’s most ambitious software project currently, and it also has tremendous competitive advantage potential if Intel can pull it off. With oneAPI, a developer can use a single API instead of writing on four or five different ones. That allows the developer to choose the best architecture for their problem and use it for the development duration. It also makes it a lot easier for customers and developers to go ‘all-in” on Intel compute. oneAPI doesn’t remove the need or ability for developers to drill into underlying code and make optimizations to get the absolute best performance and efficiency- they can, but oneAPI enables an abstraction layer that makes it easier for a developer to arbitrate between different compute units.
The gold release of oneAPI toolkits is the natural next step for Intel right now, especially as the number of intelligent devices and the amount of data continues to grow. To develop applications for these devices requires a mixture of architectures that span CPUs, GPUs, FPGAs, AI, and ML accelerators. I believe these toolkit announcements show the company is very committed to pulling a unified programming model and giving developers what they need to build on oneAPI. I talked about Intel’s oneAPI software project more extensively in a previous writeup, and you can access that here.
Intel Server GPU announcement
Intel has been on a roll with its graphics announcements as of late. With the Intel Iris Xe MAX available in OEM systems, Intel is keeping the ball rolling with its next release. After announcing the new toolkits for oneAPI, Intel also announced its first discrete Server GPU, the Intel Server GPU. The intended use of the Intel Server GPU isn’t HPC or AI but instead will power Android cloud gaming and high-density media transcoding and encoding.
Intel is positioning the Server GPU as a low cost per stream and energy-efficient solution. Customers should be able to power more gamers per GPU, which reduces the total cost of ownership. Intel built its Intel Server GPU on the Xe – LP microarchitecture, a low-power, energy-efficient architecture. The new Intel Server GPU can pair with Intel’s Xeon Scalable processors and Intel’s open-sourced and licensed software to complete the hardware and software combination.
Intel didn’t release much in terms of technical specification except that the GPU itself comes with a 128-bit pipeline and 8GB of dedicated low-power DDR4 memory. It’s worth mentioning that the Intel Server GPU is also a fanless design and all cooling from the server. Intel also announced a cloud gaming solution with H3C called the H3C XG310 PCIe card, and it utilizes four Intel Server GPUs. Each of the Intel Server GPUs can support up to 20 Android games per GPU and 100 simultaneous users within a 2-card system. There is also the ability to scale up to 160 concurrent users depending on the server configuration. Intel now has every piece of hardware and software it needs to provide customers with a great cloud gaming experience with this launch.
I think the Intel Server GPU announcement is another step in the right direction for Intel’s graphics business. It’s not the end, but the beginning. Since Raja Koduri joined the company in 2017 as a Chief Architect, Intel focused on developing a new graphics architecture and discrete and integrated solutions. We saw a sneak peek of the new Xe graphics architecture with the demos of DG1 at CES 2020. I wrote about that presentation and demo, and you can access that here. We now formally know the DG1 as the Intel Iris Xe MAX, which is available now in partner devices with Acer, Asus, and Dell. I wrote a full piece on that announcement, and you can access that here. I give all that context to say this Intel looks to be committed to winning in graphics. Intel GPUs are moving into integrated and discrete notebooks, and the server is next on that path.
The Intel Server GPU is an introduction to what is to come from Intel in the GPU space. The card is a great way to test the waters and show that Intel’s Xe graphics architecture works for cloud gaming, transcoding, and encoding workloads. The GPU is an excellent way for Intel to dip its feet in the Server GPU space without diving headfirst. HPC and AI-focused cards are on the horizon. In terms of the oneAPI toolkit launch, the toolkits’ gold release is the right step for building a unified and mature development environment. There are still plenty of tools and libraries to add to oneAPI to get it where it needs to be, but Intel is progressing nicely, especially when considering the beta version was released about a year ago next month.
Note: Moor Insights & Strategy writers and editors may have contributed to this article.