Intel is developing its own range of architecture-based gaming video cards Vehicle HPG, expected to debut this year. Recently, the boss of the graphics division Raja Koduri hinted that the project at an advanced stage, publishing the image of a test of the 3DMark suite rendered by a new gaming GPU.
Waiting for official information, they have circulated in the past few hours rumors on presumed specifications of the future range of cards Intel designed to compete with the proposals of Nvidia and AMD. Obviously take them with the benefit of the doubt, but it would appear that Intel is working on three GPU variants for a total of at least six models, listed here by the most powerful at scaling:
- Xe-HPG 512EU: GPU DG2-512EU, 4096 shader, 16/8GB di VRAM GDDR6, bus a 256 bit
- Xe-HPG 384EU: GPU DG2-384EU, 3072 shader, 12/6GB di VRAM GDDR6, bus a 192 bit
- Xe-HPG 256EU: GPU DG2-384EU, 2048 shader, 8/4GB di VRAMGDDR6, bus a 128 bit
- Xe-HPG 192EU: GPU DG2-384EU, 1536 shader, 4GB di VRAMGDDR6, bus a 128 bit
- Xe-HPG 128EU: GPU DG2-128EU, 1024 shader, 4GB di VRAMGDDR6, bus a 64 bit
- Xe-HPG 96EU: GPU DG2-128EU, 768 shader, 4GB di VRAMGDDR6, bus a 64 bit
Recall that Intel so far has only spoken broadly about the Xe HGP project, emphasizing that it will represent the union of some aspects of the other architectures – Xe LP, Xe HP and Xe HPC – on which he has worked or which he is completing. Gaming architecture should take graphics efficiency from Xe-LP, scalability from Xe-HP and compute efficiency from Xe-HPC, without missing the ability to handle hardware ray tracing and GDDR6 memory.
As for production, Intel could rely on TSMC (6 nanometers?) Or Samsung, still unclear, but if the plans have not changed compared to what was declared at Architecture Day 2020, I will not exploit their plants. Intel has already introduced SG1 in the server field, Iris Xe in the desktop field and Iris Xe Max as a dedicated GPU on notebooks. All three of these projects are based on the Xe LP architecture and have no ambitions in the gaming sector.