Intel unveils GPU for servers with Tencent, Gamestream, and Ubitus

Intel showed off a long-awaited chip today, the first standalone graphics processing unit (GPU) for the server market. That’s a good thing for Intel, as we know it can make lots of money in datacenters. When it comes to graphics processing in the datacenter — needed for highly parallel processing tasks — Intel rivals Nvidia and Advanced Micro Devices have been making all the money.

To bring the Intel Server GPU to market, Intel is working with various software and services partners, including Gamestream, Tencent Games, and Ubitus. Tencent teamed up with Intel for volume-based server deployments of its new GameMatrix cloud gaming platform, which offers over 100 popular games. With Intel Xeon scalable processors — a complete software stack and H3C XG310 PCIe add-in card that contains four Intel Server GPUs — Tencent can now provide a high-density, low-latency, low-power Android gaming solution in an affordable way.

“This is our first discrete GPU for the datacenter,” Intel VP Jeff McVeigh said in an interview with GamesBeat. “This really brings forward the capabilities and leverages the Xe-LP product line. It’s being deployed in a PCI Express card from H3C and really is focused around some of the key workloads of Android cloud gaming and media streaming that provide a very substantial total cost of ownership advantage.”

Such GPUs become useful when Tencent serves games from the cloud to users on PCs, consoles, and smartphones. If a single GPU can serve lots of users, as is the case with this Intel Server GPU, then a company like Tencent can lower its cost per user. That was tough back in the old days — if anyone remembers OnLive — because a single GPU could only serve one user at a time.

Tencent’s Arena of Valor (a multiplayer online battle arena akin to League of Legends) is one of the most popular mobile games on the market today, and it demands performance. Intel Server GPUs can scale to handle many game streams at the same time. With the H3C XG310 PCIe card, Intel can stream 60 streams in 720P at 30 frames per second. When two cards are used, Intel can stream 120 simultaneous users, still at 30 FPS. Typically, the H3C XG310 card can support greater than 100 simultaneous users in a two-card system that includes Intel Xeon Scalable Processors and up to 160 simultaneous users, depending on the specific game title and server configuration.

“We’ve been working really closely with Tencent in the space,” Intel VP Lynn Comp said in an interview. “What’s really interesting about this space is the balance of GPU density that we can support and the bandwidth the CPU supports. It’s a well-balanced system. Getting to a high number of users per GPU is really important to the total cost of ownership.”

With its first discrete GPU for the datacenter, Intel is further expanding its suite of software for enhanced cloud gaming and media experiences. The company is targeting Android cloud gaming and high-density media transcode and encode applications for real-time over-the-top video streaming. The chip follows the launch of the standalone Intel Iris Xe Max graphics chip for PCs.

Intel announced that Acer, Asus, and Dell have signed on as its first customers.

OneAPI

Above: Intel Server GPUs sit on a H3C board.

Image Credit: Intel

Intel also announced that it will release its Intel OneAPI tools in December to help customers make use of its new hardware. Putting an emphasis on this software isn’t a bad idea for Intel. But I think Intel’s best shot at coming from behind and winning in graphics will be through the cold glint of hardware. I’d say becoming the fastest GPU in the West is the straightest path to a datacenter manager’s heart. But that’s just a guess.

The OneAPI initiative means Intel will have its application workloads, middleware, frameworks, languages, and libraries running across all of its chips, including CPUs, GPUs, accelerators, and field-programmable gate arrays, McVeigh said.

“With the OneAPI, what we’re really trying to accomplish here is broaden out the diversity of architectures to support the workloads and unified programming model across those architectures,” McVeigh said. “We’ve got a standards-based solution that we will make broadly available.”

The idea is that software is optimized to run across different Intel processing units, known as XPUs, and that the software runs better because the various devices were designed with it in mind. McVeigh said the XPU vision means Intel can offer a mix of architectures on the hardware side with a unified experience on the software side. This should make life easier for developers, he said.

The new Intel Server GPU uses the Xe-LP microarchitecture and is designed for high-density, low-latency Android cloud gaming and media streaming.

As we approach an era with billions of intelligent devices and an exponential growth of data, CPUs will need to shift to a mix of architectures across CPUs, GPUs, FPGAs, and other accelerators.

Intel said the launch of the Intel Server GPU is another step toward extending the company’s offering in the XPU era. This period also requires a comprehensive software stack. Developers will be able to access a common, open, and standards-based programming model across Intel XPUs with Intel’s OneAPI tools.

With OneAPI, developers can choose the best architecture for the specific problem they are trying to solve without having to rewrite software for the next architecture and platform.

Above: Intel’s GPUs and OneAPI software go together.

Image Credit: Intel

The Intel DevCloud, where developers can test code and workloads on a variety of Intel architectures, is also expanding to include new Intel Iris Xe graphics hardware. Intel Iris Xe MAX graphics is now available for public access, and Intel Xe-HP is available to select developers.

OneAPI has received industry support, including recent endorsements from Microsoft Azure and TensorFlow. A number of leading research organizations and universities also endorse OneAPI. In addition, the University of Illinois Beckman Institute for Advanced Science and Technology today announced the establishment of a new OneAPI center of excellence (CoE). The university is using the OneAPI programming model to extend the life sciences application nanoscale molecular dynamics (NAMD) to additional computing environments. NAMD, which simulates large biomolecular systems, is helping researchers tackle real-world challenges such as COVID-19.

Developers can leverage the common API in today’s Media SDK, which will be incorporated into the OneAPI Video Processing Library next year.