Intel Releases Xe Max Graphics, Details Power Sharing and Deep Link

Intel is launching its Iris Xe Max graphics in thin-and-light laptops, starting today with launches in China. They will first appear in the Acer Swift 3x, Dell Inspiron 15 7000 2-in-1 and Asus VivoBook Flip TP470. These laptops will come to the United States in the coming weeks. 

Formerly codenamed DG1, the chip has 96 EUs, a frequency up to 1.65GHz and, yes, LPDDR4x memory for power savings. Rumors of the DG1 test vehicle using 3GB of GDDR6 were, apparently, incorrect. Instead, Xe Max uses a 128-bit interface with the lower power memory to ultimately end up with more bandwidth than competing solutions like Nvidia's GeForce MX350.

The company is aiming the Xe Max laptops at content creators. Intel is using its ownership over both the Tiger Lake CPU and the discrete GPU in a feature called Deep Link, combining the processing with a software framework to boost performance.

This includes dynamic power sharing between the CPU (with the integrated Iris Xe graphics) and the discrete Iris Xe Max GPU. In a Handbrake encode, Intel claimed the 28W CPU in the Acer Swift 3x easily bested an Intel reference system combining a Core i7-1185G7 with an Nvidia GeForce MX350. It also outperformed when enhancing details in Adobe Lightroom against a Lenovo Slim 7 with a 10th Gen Core i7 and MX350.The power sharing sounds somewhat similar to AMD's SmartShift tech. 

Additionally, Deep Link lets the processors share Xe graphics, DP4a instructions, and Xe media encoders. In a demo of Topaz Gigapixel AI, Intel showed the Xe Max and Tiger Lake upscaling images with OpenVINO and DLBoost DP4a far faster than Ice Lake with an MX350 using Tensorflow. The combined Iris Xe and Iris Xe Max, in the demo, finished faster than the competitive laptop completed a single image.

The combined media encoding should allow for up to 1.78x faster encoding than using an RTX 2080. Going up against a Gigabyte Aero 15 with a Core i9-109080HK and an RTX 2080 Super Max-Q in Handbrake, the Acer Swift 3x transcoded ten 1-minute clips from 4K/60fps AVC to 1080p / 60 fps HEVC faster. Intel says this is thanks to Intel's QuickSync with the integrated and discrete graphics working together simultaneously, compared to Nvidia's NVENC. 

Intel is also working on ways to use the four media encoders present in an Iris Xe Max laptop (two in the CPU's integrated graphics, two in the dedicated GPU) to accelerate single stream encoding. Intel hopes to have that available in the first half of 2021, and says the performance should potentially double that of a single RTX 2080 using NVENC.

Deep Link will support HandBrake, OBS, Topaz Gigabpixel, XSplit and more today, with Blender, CyberLink anad others getting support in the near future.

Intel isn't positioning Iris Xe Max as a gaming chip, but is claiming you can get solid 1080p performance versus an MX350.  In a chart, Intel showed that if offers more performance in games like Shadow of the Tomb Raider, The Witcher 3, Middle Earth: Shadow of War, Hitman 2, Metro Exodus and Grid 2019, though the MX350 won out in Borderlands 3 and Gears Tactics. 

Interestingly, while the Iris Xe Max beat the Iris Xe in Metro Exodus (in fact, the integrated graphics also surpassed the MX350), both the integrated graphics and MX350 were faster in DOTA 2. This seems like an abnormality, in that the higher clocked GPU without shared RAM should generally deliver higher frame rates. Intel said it will test games and have a list in the Intel Graphics Command Center to ensure the best-performing GPU is used.

This is not a high-end part, and unlike the media encoding and content creation workloads, Intel isn't doing multi-GPU in any form right now. The laptops will switch between integrated and dedicated Xe Graphics as deemed best.

Of course, we'll have to get our hands on these laptops ourselves to see just how well Intel Xe Max performs in our testing.

Intel told Tom's Hardware that DG1 will also come to desktop, as a discrete card for some entry and mid-level prebuilts from OEMs. That will happen in 2021, but these definitely aren't enthusiast class chips — we'll need Intel's Xe HPG for that, which will arrive some time later in 2021.