Leveraging the RDNA 2 graphics architecture, the Ryzen 6000 mobile processors were the first to adopt iGPUs packing fixed function Ray Acceleration (RA) units. Intel is looking to follow suit, adding the same capability to its 14th Gen Meteor Lake processors. The Gen12 Xe graphics architecture which was introduced with the 11th Gen Tiger Lake lineup will be replaced by a modified implementation of Xe-HPG. The latter is what powers the 1st Gen Arc Alchemist discrete GPU family.
However, it’s worth mentioning that the ray-tracing implementation on these next-gen iGPUs may be different from Alchemist. Furthermore, the XMX units will be fused off, making matrix multiplication unfeasible. Simpler AI workloads can still be carried out using mixed precision compute. Therefore it’s believed that the Meteor Lake GPUs will feature vector engines/Xe-Cores different from Alchemist.
High precision (FP64) DP calculations are carried out the same way as in Gen9 and 11, using the EM units. Over the past couple of generations, EM has been separated from the primary FP/INT pipelines into a single dual-wide configuration. This will generally result in slower FP64 compute performance while balancing INT/FO workloads.
Source: Coelacanth’s Dream
