Intel has debuted its first discrete graphics processing unit (GPU) for the data center, Intel Server GPU, based on the Xe-LP microarchitecture and designed specifically for high-density, low-latency Android cloud gaming and media streaming.
“Today is a key moment in our ambitious oneAPI and XPU journey. With the gold release of our oneAPI toolkits, we have extended the developer experience from familiar CPU programming libraries and tools to include our vector-matrix-spatial architectures. We are also launching our first data center GPU based on Xe-LP microarchitecture focused on the fast-growing cloud gaming and media streaming segments,” said Raja Koduri, Intel senior vice president, chief architect and general manager of Architecture, Graphics and Software.
As the world moves into an era of billions of intelligent devices and an exponential growth of data, CPUs alone require a shift in focus to a mix of architectures across CPUs, GPUs, FPGAs and other accelerators. Intel collectively describes this as ‘XPU’ vision. The launch of the Intel Server GPU is said to be another step to extend the company’s offering in the XPU era.
With this, developers will be able to access a common, open and standards-based programming model across Intel XPUs with its oneAPI toolkits. “They deliver on the performance potential of the underlying hardware and lower software development and maintenance costs while reducing risks associated with deploying accelerated computing relative to proprietary, vendor-specific solutions,” said the company in a statement.
Intel Server GPU is based on Xe-LP microarchitecture, its graphics architecture, offering a low-power, discrete system-on-chip design, with a 128-bit pipeline and 8GB of onboard low-power DDR4 memory. “By using Intel Server GPUs alongside Intel Xeon Scalable processors, service providers can scale graphics capacity separately from the server count, supporting larger numbers of streams and subscribers per system, while still achieving low TCO,” claims the company.
Packaging together four Intel Server GPUs in H3C’s three-quarter-length, full-height x16 PCIe Gen 3.0 add-in card is claimed to support more than 100 simultaneous Android cloud gaming users in a typical two-card system and up to 160 simultaneous users, depending on the specific game title and server configuration.
Intel is working with various software and services partners, including Gamestrea
m, Tencent Games and Ubitus, to bring the Intel Server GPU to market. “Intel is an important collaborator on our Android Cloud Gaming solution. Intel Xeon Scalable processors and Intel Server GPUs offer a high-density, low-latency, low-power, low-TCO solution. We are able to generate over 100 game instances per two-card server for our most popular games, King of Glory and Arena of Valor,” said Allen Fang, vice general manager of Tencent XianYou Cloud Gaming Platform.
The Xe-LP-based Intel Server GPU is now shipping. Along with the recently introduced Iris Xe MAX graphics, the GPU aims to improve the visual computing experience for users around the world.
Announced at SuperComputing 2019, the oneAPI industry initiative is said to be a vision for a unified and simplified cross-architecture programming model that delivers performance without proprietary lock-in while enabling the integration of legacy code. With oneAPI, developers can choose the the architecture for specific problem that they are trying to solve without needing to rewrite software for the next architecture and platform.
Intel announced that the gold release of oneAPI toolkits will be shipping in December this year. It will be available for free, locally and in the Intel DevCloud, along with commercial versions that include worldwide support from the company’s technical consulting engineers. Intel will transition Parallel Studio XE and System Studio tool suites to its oneAPI products.
In addition, the Intel DevCloud, where developers can test code and workloads on a variety of architectures, is expanding to include new Iris Xe graphics hardware. Iris Xe MAX graphics is available for public access; and Xe-HP to select developers.
Intel Graphics Software: What’s new
One of Intel’s core strategies scaling GPUs from entry-level graphics to high performance computing (HPC) is the implementation of one code base. Toward this vision, Intel’s software stack now claims to supports multiple generations of graphics, including recently-launched 11th Gen Intel Core processors with integrated Iris Xe graphics and Iris Xe MAX discrete graphics. Expanding the code base to support data center products where Linux is more prevalent is next step in its Xe architecture strategy.
The company has created Project Flipfast to improve the Linux gaming experience. The Flipfast stack allows end users to run a graphical application in a virtual machine (VM) while retaining native GPU performance and full host integration with zero-copy sharing between the VM and host. It claims to enhance gaming performance, and the technology directly applies to data center game streaming applications.
Intel said that Implicit SPMD Program Compiler (ISPC) will run on top of oneAPI Level Zero, which provides low-level, direct-to-metal interfaces tailored to devices in the oneAPI platform and is the overall hardware abstraction layer. ISPC, powered by oneAPI, is a variant of the C programming language, which enables single program, multiple data programming and is used to accelerate Intel OSPRay ray tracing engine on Intel CPUs.