Intel UHD Graphics 630 vs Intel UHD Graphics G1 (Ice Lake 32 EU)

The Intel UHD Graphics G1 (Ice Lake with 32 EUs) is an integrated graphics card in the Ice Lake SoCs (10th generation core 10xxG1) for laptops. It is the slowest GPU version in the Ice Lake U series and offers only 32 of the 64 EUs (Execution Units / Shader Blocks). It offers no dedicated graphics memory (no eDRAM cache like the Intel Iris Graphics 655 predecessor of the Coffee Lake SoCs). The clock rate depends on the processor model. At launch it looks like all models have a 300 MHz base clock and only differ in the boost clock (1050 - 1100 MHz). Another difference is the TDP as it can be configured from 12 - 25 Watt in the 15 Watt U-models and maybe even higher in the 25 Watt models that will come later.

The performance should above the old Intel UHD Graphics 620 (Gen. 9.5) that was e.g. found in Whiskey Lake processors (thanks to the higher core count and improved architecture). Most likely the G1 graphics wont reach the performance of the old Iris Plus 640 that offered more cores and 64 MB eDRAM. Therefore, only a few new AAA games can be played fluently with the UHD Graphics G1.

A special new feature of the Gen11 graphics card is the new Variable Rate Shading (VRS) support. With it game designers can decide where to spend shading time and e.g. shade object in the background or behind fog with less resolution (up to using only one source for a 4x4 block). With this technique early results show up to 1.3x performance in Unreal Engine POC and 1.2x speedup in Civ 6. Up to now VRS is only supported by the new Nvidia Turing architecture (GTX 1650 and up).

Another improved hardware piece is the integrated video de- and encoder that was improved significantly according to Intel. They did not specify any more details, but the previous generation was able to decode VP9 and H.265/HEVC in Main10 profile with 10 bit color depth using the dedicated hardware.

The Ice Lake SoCs and therefore the integrated GPU are manufactured in the modern 10nm process at Intel that should be comparable to the 7nm process of TSMC.