Intel is building a stand-alone graphics product based on the Xe architecture, the first prototype card DG1 we’ve seen, but the positioning is more advanced, mainly for 1080P online games. Recently, the foreigner from Inel Iris Plus’s open source programming documentation found traces of DG2, which is based on Xe_HP. HP is an acronym for High Performance, a far cry from the Xe_LP core of the DG1.
The driver document also seems to imply the configuration information for the EU unit of DG2 (equivalent to the NVIDIA SM stream processor cluster), as follows:
– iDG2HP512 ” “Intel(R) UHD Graphics, Gen12 HP DG2” “gfx-driver-ci-master-2624”
– iDG2HP256 ” “Intel(R) UHD Graphics, Gen12 HP DG2” “gfx-driver-ci-master-2624”
– iDG2HP128 ” “Intel(R) UHD Graphics, Gen12 HP DG2” “gfx-driver-ci-master-2624”
Under simple translation, DG2 has designed three EU sizes, 128, 256 and 512. Previously, DG1 was configured with 96 EU, single-precision floating point 2 to 3TFlops. Combined with this is the core of high energy, 512 EU DG2 graphics card touch 10 to 15 TFLOPS is not difficult.
It was previously reported that there were eight stream processors in each set of Eu Xe_HP, so the 512 EU could be converted to 4096 stream processors. In addition, Xe_HP support mcM, up to 4 cores, that is, 16384 stream processors.
This explains that the Xe architecture is supported by both game graphics and data center acceleration cards with greater performance density.
Such strength, and based on intel 7nm EUV technology, consumer-grade areas called board NVIDA Ampere, AMD RDNA2/CDNA should not be a problem.