Intel's new discrete GPUs will rely on the same node as both competitors
Intel should finally release a generation of its discrete “Battlemage” graphics cards with Xe2 HPG architecture later this year, but it will be competing with AMD’s Radeon RX 8000 and Nvidia’s GeForce RTX 5000 cards for most of its lifetime. It does look like Intel will have one advantage on its side, though – Battlemage will be on the same manufacturing node, and at least that way it won’t be disadvantaged right off the bat.
According to DigiTimes, Intel has again contracted TSMC to produce these GPUs. The previous Arc GPUs of the “Alchemist” generation already used TSMC’s 6nm node (N6) instead of Intel’s own fabs (graphics cards were ahead in this, but now even CPUs will go this way). The important thing here is that Battlemage opts for a new, more advanced production node. According to DigiTimes, it will utilise the N4 technology, a 4nm node that is an improved evolution of the 5nm N5 technology.
While Intel’s Lunar Lake and Arrow Lake processors will already use TSMC’s 3nm node (N3B), graphics cards will be a bit more conservative and stick with the more proven 4nm node. Intel will miss out on the opportunity to get ahead of AMD and Nvidia, but it’s possible that 3nm technology wasn’t guaranteed to be ready for powerful GPUs with relatively large chips back when the call had to be made.
With the 4nm node, Intel should effectively have almost the same starting position as Nvidia. Nvidia’s new Blackwell generation were initially believed to be on 3nm manufacturing node, but currently it seems the company will keep making them on a 4nm node as well, after all. However, it’s possible that the Nvidia noe will be a bit better than the N4 used by Intel, since Nvidia got its own custom 4N node from TSMC (and will utilise a further tweaked 4NP node for Blackwell). While this node is still derived from a 5nm node, like N4, it could have aggressive optimizations coupled with Nvidia’s specific GPU hardware design optimizations. This could result in better performance and efficiency (albeit at the cost of a more complex or more restrictive design rules) than what is achievable with the general-purpose N4 node.
However, in terms of the quality of the technology used, Intel could probably still enjoy a similar starting position, and in the end it will come down to the quality of the graphics architecture the individual vendors developed and to who can extract the potential out of the same production technology more skilfully.
A bit of a question is the new generation of AMD graphics cards. It’s not yet confirmed what manufacturing technology its Radeon RX 8000s will be made with. But they’re supposed to be cheaper graphics cards for the mainstream, and so aren’t likely to use expensive 3nm technology. So the 4nm TSMC node is also likely to be used for them, so Intel won’t be at a disadvantage in terms of the technology used here either, and so again the architecture will be the deciding factor.
3nm Battlemage GPU could exist in Lunar Lake
However, the production of Battlemage on the 4nm node only concerns the discrete version of these GPUs, which will be sold as second-generation Intel Arc graphics cards (branded B380, B750, B770 and so on). According to DigiTimes, the Core Ultra 200V “Lunar Lake” ultra-mobile processors coming out at the end of September will use TSMC’s 3nm node for both the CPU cores and the GPU cores. Apparently they could in fact be integrated in the main Compute tile with the GPU cores.
Read more: Intel Lunar Lake: The most efficient x86 processor detailed
Arrow Lake processors, which will be a separate branch of the same Core Ultra 200 generation intended for more powerful laptops and desktops, will probably have the graphics integrated separately on its own tile, or in the SoC tile. Presumably, therefore, their graphics will use an older process. However, according to earlier reports, the Battlemage architecture is not even used in Arrow Lake processors, they will reportedly use the Alchemist (Xe LPG) architecture like the already available Meteor Lake processors.
Sources: TechPowerUp, DigiTimes
English translation and edit by Jozef Dudáš
⠀
⠀