Intel GPU Muscles Up for Media Processing and the Visual Cloud

August 25, 2022 by Jake Hertz

Optimized for media streaming, Intel’s new GPU family hopes to meet the video processing needs of data centers and the cloud.

One of the single largest contributors to global Internet traffic is video, with Cisco reporting that almost 80% of all Internet traffic comes from either business or consumer video. That number is only expected to increase in today’s world of increasingly remote work and constant content generation.

With that in mind, in order to build out the computing infrastructure necessary to support the future of Internet usage, many are placing heavy emphasis on video processing. This week, Intel unveiled technology along those lines by announcing its new family of data center GPUs designed specifically for media streaming.

In this article, we’ll look at the new family, some of the underlying technology, and the potential impact on the field.


Flex Series GPUs for Data Centers

The devices in the new family, called the Flex Series, are described by Intel as data center GPUs that were meant to be flexible and robust while being optimized for applications such as media streaming and cloud gaming. Intel calls this the “intelligent visual cloud.”


A Flex Series GPU card.

A Flex Series GPU card. Image used courtesy of Intel


To do this, the Flex Series is what Intel says is the industry’s first data center GPU to be integrated with a hardware-based AV1 encoder, enabling Flex Series to offer 5x the media transcode throughput performance and 2x the decode throughput performance as compared to competitive solutions. 

Another unique feature of the Flex Series is that it’s built off of an open architecture. The architecture was designed to support a standards-based software stack, enabling organizations to mitigate concerns of design complexity and cost when bringing Flex-based solutions to the market.

As a corollary of this open architecture, Intel claims that its Flex Series will reduce the need for data centers to use disparate solutions and allow for easier management of proprietary and heterogeneous systems.

According to Intel, the GPU will come in two SKUs, the 140 series for maximum density and the 170 series for maximum peak performance. The 140 series will feature 2 GPUs per Card, 16 total X Cores, 4 fixed function media blocks, a peak performance of 8 TFLOPS at FP32, and a TDP of 75 W. The 170 series, on the other hand, will feature 1 GPU per card, 32 X cores, 2 fixed function media blocks, 16 TFLOPs at FP32, and a TDP of 150 W. More information can be found in the Flex Series solution brief.

Xe-HPG Microarchitecture for Gaming and Media

From an EE standpoint, an interesting feature of the new Flex Series is its use of Intel’s proprietary Xe-HPG architecture. According to Intel, their Xe-HPG microarchitecture was designed specifically for performance and scalability in use cases such as gaming and media creation.

The foundational compute building block of the Xe-HPG architecture is Intel’s Xe-cores, each of which is configured with a set of 16 256-bit vector engines for graphics acceleration, as well as 16 1024-bit matrix engines for AI acceleration.


Xe-core as part of the Xe-HPG microarchitecture.

Xe-core as part of the Xe-HPG microarchitecture. Image used courtesy of Intel


The microarchitecture consists of a number of different “render slices,” which are individual sets of Xe-cores and ray tracing units, each interconnected through a high bandwidth memory fabric, sharing a large L2 cache. In this way, Intel has designed the Xe-HPG architecture, and the Flex Series GPUs, to be flexible and scalable, simply adding more slices to a GPU infrastructure if need be.


Boost for Data Centers and Cloud Computing

As video, media, and gaming continue to take up a significantly larger share of the global Internet traffic, we need to develop hardware that can accelerate these workloads. This is exactly what Intel aimed to do with the Flex Series GPUs, and with it, they hope to be able to improve the performance and efficiency of data centers and cloud computing everywhere.