At the disposal of the resource VideoCardz turned out to be fresh information regarding GPUs Xe from Intel. As it turned out, the “blues” have been developing another graphic architecture since 2018, which has the symbol Xe-HPG. According to the source, chips based on it will be used in Intel gaming graphics cards. They will be featured in the mid-range and high-end desktop segments.
In architecture Xe-HPG Intel has embodied all the advantages and benefits that the X architecture hase-LP (graphics performance), Xe-HP (scalability) and Xe-HPC (computing power). The developer himself has not yet officially confirmed this information, however VideoCardz indicates that X-based gaming graphics cardse-HPG will use GDDR6 memory. As for solutions based on architecture Xe-HP, Intel will use more expensive HBM memory in them.
Two years ago, NVIDIA introduced hardware-accelerated real-time ray tracing with the announcement of GeForce RTX 20-series graphics cards. AMD is set to introduce similar technology this year in GPUs based on the new RDNA2 architecture. According to VideoCardzIntel Confirmed Support hardware acceleration ray tracing graphics chips based on architecture Xe-HPG.
Intel itself will not be involved in mass production of GPUs. It will be assigned to a third party contractor. The question of who exactly is still open. Chips based on architecture Xe-HPG undergo laboratory tests. Intel is reportedly set to release discrete gaming graphics in 2021.
The resource recalled that Intel is working on the SG1 graphics accelerator based on the X architecturee-LP. This video adapter is intended for use as part of server equipment. It is indicated that its deliveries will begin this year.
Also, information has appeared confirming some specifications of the GPU series based on the X architecturee-HP. It will consist of three GPUs: one, two and four GPU clusters. An engineering prototype based on these chips is codenamed XeHP HD Graphics Neo. The highest performing one is labeled Arctic Sound.
According to the source, the single-cluster chip received 512 Execution Units (EU), the two-cluster version – 1024 EU, and the most productive and large four-cluster chip (pictured above) will have 2048 execution units.
The resource also reported performance data for each of the chips. For a single-cluster GPU, it is 10.6 teraflops, for a two-cluster GPU – 21.2 teraflops, and for a four-cluster GPU – 42.3 teraflops. Almost linear scalability.
If you notice an error, select it with the mouse and press CTRL + ENTER.