Intel Iris Xe MAX Graphics

The Intel Iris Xe MAX (DG1 Low Power - LP or iDG1LPDEV) is a dedicated PCIe 4.0 mobile entry-level graphics card with 96 execution units (EUs) based on the Gen 12 architecture. Compared to the integrated Intel Iris Xe in the Tiger Lake CPUs, the Xe MAX offers 4 GB dedicated LPDDR4x graphics memory (68 GB/s versus 56 GB/s for the MX350) and higher clock speeds of 1.65 GHz (vs. 1.35 GHz). Both support DP4A DLBoost instructions to speed up AI tasks and two media engines (including AV1 decoding in hardware). Currently the GPU is only available for Tiger Lake based laptops and offers some possibilities to use the iGPU and dGPU together (using the Deep Link software framework). The GPU uses hybrid graphics to turn off the Xe MAX when not in used (display ports are routed through the iGPU, but the Xe MAX features four dedicated display ports in comparison to the Nvidia GeForce MX350).

Furthermore, Intel supports to shift TDP between the two chips for better CPU performance or balancing graphics workloads.

The gaming performance is on average slightly better than a fast Intel Iris Xe iGPU with plenty of TDP and cooling headroom. There are even some games where the iGPU is faster thanks to Deep Link sharing power and cooling. Intel targets 1080p gaming with low to medium settings with the GPU and provided examples where the MAX is on par or better than a dedicated Nvidia GeForce MX350. 

According to Intel, the GPU is mostly aimed at content creators that can make use of the additional compute units, AI (DP4a) acceleration or media encoders.

As the Iris Xe Graphics, the Xe MAX supports Variable Rate Shading, Adaptive Sync, Async Compute, DirectX 12.1, OpenGL 4.6 and OpenCL 2.

As the Tiger Lake CPUs, the Iris Xe MAX dGPU is manufactured in the modern 10nm SuperFin process at Intel. Intel aims the Max-GPU for laptops with a combined power budget of 35 Watt and up. The power consumption of the GPU alone is specified similar to the 25W of the GeForce MX350 (we guess including the graphics memory).

Benchmarks