Intel Reveals Plans to Launch Xe-HPG Gaming GPUs in 2021

Hardware ray tracing, GDDR6 memory.

IntelXE442432
Last Updated:

Intel has officially announced today plans to launch its very first discrete GPU aimed at the gaming market in 2021- the Xe-HPG. In statement confirming the news, Intel explains.

“And, our enthusiast and gamer friends, we heard your requests for Xe for enthusiast gaming. We added a fourth micro-architecture to the Xe family: Xe-HPG optimized for gaming, with many new graphics features including ray tracing support. We expect to ship this micro-architecture in 2021 and I can’t wait to get my hands on this GPU!”

 

Additional details have surfaced courtesy of a new report from the ever-prolific folks over at VideoCardz, who claim that Intel intends to launch the Xe-HPG GPU with an eye on taking on AMD’s Big Navi and NVIDIA’s Ampere RTX 3000-series GPUs. Xe-HPG has reportedly been in development since 2018.

According to VideoCardz, underpinning Intel’s GPU will a new micro-architecture dubbed Xe-HPG optimized for enthusiast gaming and falling roughly between the Xe-LP and Xe-HP architectures. The Xe-HPG will allegedly consist of one tile.

The report includes specifications for Intel Xe-HP cards, which help shed light on what to expect from the HPG. HP will feature three variants, defined by their tile numbers: a 1-tile, 512 execution unit, 10.6 TFLOPs variant; a 2-tile, 1024 execution unit, 21.2 TFLOP model; and a 4-tile, 2048 execution unit, 42.3 TFLOP option. Based on what we know about the one tile Xe HP, the HPG could conceivably feature an identical count of 512 execution for 4096 cores, or potentially more if Intel decides to rejig its EU count. We’ll have to wait for official specifications to get a firm grasp on this aspect of the GPU.

The Xe-HPG will also offer gaming-oriented features, namely hardware-accelerated ray tracing and GDDR6 memory, reportedly to up the performance per dollar offering. Ray tracing capabilities could put it on par with NVIDIA’s existing solution found on the Turing family of GPUs, and AMD’s take on hardware ray tracing expected to arrive with its RDNA2 cards.

VideoCardz explains that Intel is still evaluating the cards at its labs. Intel confirmed they are based on the 10nm process node, and the company says it is dishing out manufacturing duties to an external foundry. Samsung may be involved; its 10nm process has been pumping out silicon for some time now, but this remains entirely speculative at this point. We can’t discount TSMC being in the mix either.

Intel’s Raja Koduri, offered the following to journalists gathered for the announcement:

“At this point in time, we won’t be disclosing which foundries, or which process node from the foundry. We’ll disclose those details closer to the launch.”