Intel has started to offer a few reference designs to the board partners about the new Intel Xe-HPG. It seems that the company has a few of the custom designs ready. Raja Koduri, the senior Vice President and the General Manager of the graphics group, has gone through an interview by ASCII, a Japanese site.
We also know that Intel has chosen TSMC N6 because they have an insufficient supply of the capacity of the inhouse nodes.
Raja Koduri has given us the answer to several questions about the Intel Xe-HPG architecture, which will be launching in the Q1 of 2022. The series will be placed under the very new Arc Alchemist. He was questioned as to why Intel is taking in chips from TSMC instead of using the company’s own “7” node.
The senior Vice President of the company has responded by saying that the company’s capacity of mass p[roduction wasn’t a lot, and it was not capable of mass-producing the graphics processing units. Intel has no intention of competing with its CPU series, such as the Alder Lake.
Determination of the manufacturing capacity of the entire process was very important. This could be assumed at the very beginning of the design making. The advanced processes which Intel follows showcase that it does not have adequate capacity.
The company has not yet decided whether the Intel Xe-HPG will be a successor of Alchemist. This decision will be relying totally on the external foundry of the company. It seems that the company might soon be shifting to a few of the advanced node technologies with AMD RDNA3 and NVIDIA Lovelace series with the help of the TSMC N5 process.
After the new Intel Xe-HPG, it seems that the Xe-LP architecture will be compatible with the supersampling technology of the company.
Raja Koduri has also confirmed whatever Intel had announced earlier. The Xe-based architecture, which was already there earlier in the Xe-LP, will now be supporting the XeSS technology. This is the version that Intel has, which provides super-resolution on several AI platforms.
In the Xe-LP, we will see the DP4a instruction. This has the additional feature of backward compatibility with all the latest GPUs. XMX based versions were far more efficient as compared to that of the XeSS of the DP4a version. However, it belongs to the low-power capacity genre, so it will definitely be a good addition to the series along with the Intel Xe-HPG.
Since the supersampling technique states that the XeSS is backward compatible, we can say that the Iris Xe DG1 can also use it. It is also being said that it will be working based on the 11th generation GPUs, just like the ones of Tiger Lake.
We will not be seeing Ponte Vecchio’s Xe-Link in the Intel Xe-HPG series.
There was another very interesting piece of news that Raja had not confirmed. It was whether the Intel Xe-HPG series would come with the support of multiple GPUs like the AMD Crossfire and NVIDIA SLI.
Both these technologies do not have a big market as of now, mainly because of a couple of reasons. One of them is that the performance is very subpar, and the other factor is that multiple GPUs coming from the same vendor become a bit on the expensive side and are also difficult to get hold of. Intel has no intention to be a part of a multi-GPU market that is on its way downward.
What are the plans regarding customized cards and Arc series?
The most interesting part of the interview was the matter of whether there will be any launches of customized cards which will be based on the GPUs of the Arc Alchemist series.
As it was informed earlier, we will get to see Intel offering some reference designs to bring on board a few of its partners. This news was brought out earlier during Intel’s drone show. Moore’s Law is Dead also showcased a similar design and will be boarding the partners who wish to use it or might not use it as well. This can be implemented into their semi-custom versions.
Stay connected with MediaScrolls to get more updates on the Intel Xe-HPG series and to receive several other news from Intel.