Intel Corp. today expanded its data center chip portfolio with a new lineup of graphics processing units, the Flex Series, that’s optimized for use in servers.

The Flex Series includes two GPUs on launch. The first GPU includes 16 graphics processing cores and 12 gigabytes of onboard memory, while the second features twice as many cores and 16 gigabytes of memory. Both chips are based on Intel’s Xe HPG graphics card architecture, which is optimized primarily for running video games.

Intel says the two chips support a variety of workloads. 

Flex Series graphics cards can perform video transcoding, or the process of changing the file format in which video content is stored. Companies such as broadcasters perform transcoding regularly as part of their day-to-day technology operations. Intel says the Flex Series also lends itself to powering cloud-based video game streaming services, virtual desktop environments and AI applications.

According to Intel, a single Flex Series chip can simultaneously transcode up to 36 video streams with a resolution of 1920 pixels by 1080 pixels. Companies that require additional performance can use multiple GPUs side-by-side. According to Intel, 10 Flex Series graphics cards may be integrated into a single server to process 360 video streams at once.

The chipmaker says the GPUs provide better performance than comparable Nvidia Corp. silicon in some situations. During internal performance tests, Intel determined that a Flex Series chip can provide five times as much throughput as Nvidia’s A10 graphics card when performing video transcoding. Moreover, Intel says that the Flex Series uses as little as half the electricity.

The Flex Series’ performance is partly facilitated by a built-in hardware module optimized to run the AV1 video encoding software. Encoding, or the task of compressing video files, is often a part of companies’ transcoding workflows. Intel says that its new GPUs’ built-in AV1 module makes it possible to perform processing with 30% less bandwidth than AVC, a previous-generation encoding system.

The other major use case that Intel is targeting with the Flex Series on launch is video game streaming. In particular, the chipmaker says that Android game developers can use Flex Series chips to offset certain computing tasks from users’ devices. According to Intel, a single Flex Series graphics card can process up to 68 video game streams with a resolution of 1280 pixels by 720 pixels. 

“Today’s data center infrastructure is under intense pressure to compute, encode, decode, move, store and display visual information,” said Jeff McVeigh, the vice president and general manager of Intel’s Super Compute Group. “Intel ​Flex Series GPU is a breakthrough design that uniquely solves today’s computing demand while providing flexibility and scalability for the immersive experiences of tomorrow.”

Over the next few months, Intel expects a half-dozen hardware makers to launch data center systems equipped with Flex Series graphics cards. The initial set of systems will be optimized for transcoding and Android game streaming, according to the chipmaker. Further down the line, Intel hardware partners will introduce Flex Series servers optimized for desktop virtualization, Windows game streaming and artificial intelligence use cases.

The Xe HPG graphic card architecture that the Flex Series uses is one of four that Intel has developed as part of its effort to establish a bigger presence in the GPU market. For data centers, Intel also offers the Xe HPC architecture, which is designed to run AI workloads. Xe HPC chips contain more circuits optimized for matrix operations, a type of calculation that neural networks use to process data.

The Xe HPC architecture is the basis of Ponte Vecchio, Intel’s most advanced AI chip for the data center market. Ponte Vecchio includes 100 billion transistors organized into dozens of processing modules. According to the Intel, those modules can provide more than 45 teraflops of combined performance.

Image: Intel