Intel Confirms Arc Battlemage Xe2 GPUs In The Labs With 30% Software Team Readying It Up, Hardware Team Moved To Next-Gen Celestial Xe3 GPUs

Intel has confirmed that its engineering teams have moved to the next-generation Arc Battlemage Xe2 and Celestial Xe3 GPUs with more good news on the way.

During an interview with PCWorld's Adam Patrick Murray and Brad Chacos, Intel's Tom Petersen once again confirmed that the next-gen Arc GPUs are coming soon and they will have more good news to share in 2024.

Tom pointed out that the next-gen Arc Battlemage Xe2 GPUs are on schedule for a 2024 release as we had previously reported and 30% of the software team is already working to ready up initial support for the discrete and integrated lineup that will ship with future Arc graphics cards & iGPUs. The first silicon is also confirmed to be running in the labs and while there isn't much that can be talked about Battlemage right now, we are hinting at a release before the next CES 2025 so we can see a target release window within 2H 2024.

More importantly, Tom also mentioned that the hardware team is working on the next big thing which is him referring to Intel's Arc Celestial Xe3 GPU family. These GPUs are expected to arrive in 2025 with the Panther Lake CPUs as an iGPU and also in discrete GPU flavors for mobile & desktop products.

It's coming, I am excited about it, and all our engineers you know how they are constantly doing their engineering things. I'd say about 30% of our engineers are working on Battlemage, mostly on the software side because our hardware team is on the next thing (Celestial), so think about it as the Battlemage has already  has its first silicon in the labs which is very exciting and there's more good news coming which I can't talk about right now.

We hope we are going see it before it (next CES '25).

Tom Petersen (Intel Fellow) - via PCWorld at CES 2024

Besides the next-gen Arc GPU stuff, Tom Petersen also laid out some interesting use cases for the NPU architecture that is widely being adopted by CPU manufacturers to power the AI PC segment. We haven't seen a lot of use cases for the NPU besides some casual PC uses such as facial & voice recognition but Tom proposes that this NPU can be leveraged for more advanced tasks in gaming such as character manipulation where AI can help create lifelike human animations.

It does kind of sound like this might be an early hint at Intel's own AI model similar to NVIDIA's ACE (Avatar Cloud Engine) which leverages AI to create lifelike and believable human-AI interactions and animations. It is great to see that Intel has emphasized AI for gaming too so we can see more experiences in the future.

We are very early days in the NPU. The NPU is mostly about running low-power, always-on kind of applications. Maybe running facial-recognition, gesture recognition so for a handheld, I mean I don't know what they are doing yet but I can easily imagine lots of col use cases. Different ways of human-machine interaction would be a great one.

But medium term is game-enabled AI. So imagine that it's AI for character manipulation or character animation enhancement. I think the greater implementation of AI that we haven't seen yet is animation augmentation. There's a lot of shitty animation in games today, no judgment, but image that you put in AI and you put in relatively crappy animations and you get lifelike human animation, that's free! You are welcome EPIC.

Tom Petersen (Intel Fellow) - via PCWorld at CES 2024

Finally, talking about the disaggregated future, Intel lays down the advantages of using a chiplet based design. The major benefit is cost but there's also the benefit of creating chips that haven't been made before or weren't in the realm of possibility due to monolithic design constraints.

Intel believes that there's still a continuum that hasn't been explored and they don't know if they will do something like that but considering that others are adopting more advanced chiplet and multi-tile architecture, it is something that the blue team has its eyes on.

To me, it's like disaggregation has two or three main benefits, one is just economics. You can build SOCs most cost-effectively because you can charge different IPs to specific technologies and you can get more chips with your latest technologies and you can use your older process technologies for pieces that are less important or less performant. That's one benefit. Of course, you have to get the packaging cost right and this is an emerging technology that we are continuing to invest in. That is number one.

But number two is that you can build things which you couldn't build before. There's obviouslly a continum that has not been fully explored here and I can't exactly talk about what wer are doing here or if we are ever going to do here but that's what something that others are doing too.

Tom Petersen (Intel Fellow) - via PCWorld at CES 2024

Based on previous reports, the Battlemage GPUs have not only fixed some of the architectural choices that felt a bit unnecessary for Alchemist but in the first silicon is shaping out to be very well. The first BMG-G10 GPUs (Xe2-HPG) were also reportedly running in labs so if all goes well, we can expect more information from Intel in the coming months.