Intel Launches Iris Xe Max Discrete GPU With New ‘Deep Link’ Feature

With its first Xe discrete graphics product, Intel is introducing a new software framework called Deep Link that aggregates CPU and GPUs to accelerate performance. ‘I think all these companies are going to try to leverage the CPU-GPU advantages as much as possible,’ one Intel partner says.

After three years of teasing its return to the discrete graphics market, Intel has revealed its first product in the category, the Intel Iris Xe Max graphics, which will pair with the chipmaker’s new Tiger Lake processors for ultra-thin laptops to focus on content creation.

The Santa Clara, Calif.-based company announced the launch of its first discrete GPU on Saturday, which a spokesperson said was tied to the availability of systems in multiple markets, including China. The first laptops to use Iris Xe Max graphics — the Acer Swift 3x, Asus VivoBook Flip TP470 and Dell Inspiron 15 7000 2-in-1 — are now available, according to Intel.

[Related: Intel: Rocket Lake Will Mix Ice Lake CPU Architecture, Tiger Lake Graphics]

In the announcement, Intel reiterated plans to release later this year a discrete GPU for servers based on the same Xe-LP microarchitecture as Iris Xe Max. The company said it also plans to bring low-power, Xe-LP discrete GPUs to value desktops in the first half of 2021. It also reaffirmed plans to release discrete GPUs for high-performance servers and gaming PCs in 2021.

Intel said the new laptops with Intel discrete GPUs will take advantage of both the Iris Xe integrated graphics in its 11th-generation Core processors, also known as Tiger Lake, and Iris Xe Max discrete graphics using a new technology that the company calls Intel Deep Link.

“We set out to redefine the role of discrete graphics in thin-and-light laptops and address a growing segment of creators who want more portability,” Intel graphics executive Roger Chandler said in a statement. “Iris Xe MAX graphics and Intel Deep Link technology serve as examples of the kind of platform-level innovation that Intel plans to bring to market in the future as we execute to our scalable Xe roadmap.”

Deep Link may sound similar to Nvidia’s NVLink interconnect technology in how it links two GPUs — Iris Xe integrated graphics and Iris Xe Max discrete graphics — except instead of an interconnect, it’s a common software framework that “aggregates” the two GPUs as well as the CPU to dynamically share power as well as enable faster AI-based content creation and encoding.

Deep Link capabilities are made possible by Intel Media SDK, Intel OpenVINO and Intel vTune software libraries, and they will need to be enabled by application developers. Applications that have embraced Deep Link so far include HandBrake, Open Broadcaster Software and Topaz Gigapixel AI, according to Intel. Deep Link support for Blender, Magix, CyberLink and Fluendo is coming soon, the company added.

With Deep Link’s Dynamic Power Share feature, Intel said it can divert all power and thermal resources to the CPU when Iris Xe Max is idle, which it said can result in 20 percent better CPU performance for video encoding versus an 11th-gen Core i7 laptop with Nvidia’s GeForce MX350 discrete graphics.

Intel also said Dynamic Power Share can enable a laptop with a 30-watt, 11th-generation Core i7 and Iris Xe Max to deliver 40 percent faster performance for Adobe Lightroom’s enhance details feature over a laptop with a 20-watt, 10th-gen Core i7 and GeForce MX350 discrete GPU.

When it comes to enhanced AI features, Intel said the 11-gen Core i7-Iris Xe Max pairing can perform the photo upscale workload in Topaz Labs’ Gigapixel AI application roughly seven times faster than the 10th-gen Core i7-GeForce MX350 configuration. In a video demonstration, an Intel representative pointed out how Deep Link enables utilization of both the Iris Xe integrated graphics and Iris Xe Max discrete graphics while the GeForce MX350 system doesn’t make use of Intel’s integrated graphics.

Intel said Deep Link’s Hyper Encode feature is projected to provide nearly double the single-stream encode performance over Nvidia’s GeForce RTX 2080 Super with NVEnc fixed function encode engines.

While Intel focused on performance for content creation workloads, the company showed that laptops with Iris Xe Max discrete graphics can either outperform or provide comparable performance for 1080p gaming to laptops with the GeForce MX350 discrete graphics. The two games where Iris Xe Max had the biggest lead was Metro Exodus, with over 40 frames per second, and GRID 2019, with over 45.

However, Intel said, there are some situations where the Iris Xe integrated graphics of an 11th-gen Core processor can faster gaming performance than the Iris Xe Max discrete graphics. The company said this is typically due to latency issues related to traversing the PCIe bus, among other things.

“With a common software framework, we can identify whether games will perform better on the processor‘s Iris Xe graphics or the Iris Xe MAX graphics and match them to the IP that will deliver the best experience,” an Intel spokesperson said. “With our continued investment in integrated graphics, we see this as a capability that will continue to evolve as we match the best experience to the right IP.”

Iris Xe Max, which is based on Intel’s 10-nanometer SuperFin technology, comes with 96 execution units, a 1.65GHz frequency, 4GB in memory, two multi-format codec engines as well as support for LPDDR4x memory, PCIe 4.0 and Intel Deep Learning Boost with the DP4A instructions. Features include variable rate shading, adaptive sync and asynchronous compute.

Dan Young, owner and CEO of Xidax, a Salt Lake City, Utah-based Intel system builder partner, said Intel’s new Deep Link technology could be a major advantage for the company as it seeks to gain a foothold in the discrete graphics market and compete against Nvidia and AMD.

“I think Intel’s going to be a driving force in GPUs,” he said.

Between Intel’s Deep Link and AMD’s new Smart Access Memory technology, which givers users improved performance when they use AMD CPUs and GPUs, Young said he expects stronger integration between CPUs and GPUs to become a larger trend. That could give both Intel and AMD a better fighting chance against GPU juggernaut Nvidia — at least for now.

“I think all these companies are going to try to leverage the CPU-GPU advantages as much as possible,” he said.