Remember that rumour Intel was scaling back on its plans for its next-gen Battlemage graphics architecture we reported last month? At the time we said something about it didn’t square up. But now there’s new, arguably more substantial, information that implies Intel hasn’t reduced its ambitions for Battlemage at all. Intel’s next Arc GPU could still be targeting performance up to Nvidia RTX 4080 levels.

Battlemage, of course, is the follow up to the Alchemist GPU architecture in Intel’s current Arc GPUs including the A770 and A750. It’s generally expected to appear next year. The precise timings aren’t clear, but obviously the sooner the better for both gamers and Intel itself.

Anyway, the new data dump as communicated by Tom’s Hardware and originally unearthed by Twitter user Harukaze5719 involves the test tools Intel supplies for partners preparing to produce graphics cards based on its upcoming GPUs. To cut a long story short, it’s an interposer that’s related to the chip package mounting on the graphics card PCB.

And here’s the thing. The interposer for the high performance Battlemage GPU is larger than the equivalent for Alchemist. Now, this doesn’t say anything directly about the die size of the highest performance Battlemage GPU. But it does strongly imply that it’s a pretty decent size.

harukaze5719 post

(Image credit: harukaze5719)

Moreover, the interposer for the lower performing of the two upcoming Battlemage dies is much smaller. Because the chip package is smaller and so will be the die size of a lower performing GPU.

The larger Battlemage interposer is only slightly bigger than big Alchemists. 2.5 per cent larger, apparently. Intel has also said that Battlemage will be built on TSMC’s N4 process, which is effectively the same as Nvidia has used on its RTX 40-series cards. And when you take all that information together, it implies very hopeful things about Battlemage.

That’s because the current big Alchemist GPU clocks in at 406mm² and 22 billion transistors. It’s not possible to say for sure how big Battlemage will be based on the interposer size. But the fact that it’s slightly bigger than Alchemists strongly implies that it will at least be a chip on a similar scale even if it isn’t actually larger.

Arc roadmap

Battlemage is real enough and has appeared on Intel’s official roadmaps for several years. (Image credit: Intel)

Alchemist is built on TSMC’s N6 process. So an N4 chip of a similar physical size will contain far more transistors. For context, Nvidia’s AD103 GPU as used in the RTX 4080 is 379mm² and contains 46 billion transistors.

The point is that if the big Battlemage GPU is physically about the same size as the big Alchemist GPU, it will contain roughly 50 billion transistors and be comparable with an RTX 4080 for sheer complexity.

That doesn’t guarantee it will match an RTX 4080 for actual gaming performance. After all, big Alchemist is roughly on par with the RTX 3070 Ti’s GA104 GPU for size and complexity but can’t match its performance.

But this information about Battlemage definitely does imply Intel has not scaled back its ambitions. If anything, on die size and complexity, it looks to be positioned a little higher versus Nvidia GPUs than Alchemist.

We’ll only know for sure when Battlemage finally arrives next year. But, Lordy, are we hoping that it can indeed get close to the RTX 4080 for performance. That would be such an enormous boost for the GPU market as a whole. 

It might also be particularly important if rumours that AMD, in turn, is scaling back its plans for its own next-gen RDNA 4 graphics architecture—if those rumours are true, which is impossible to say for now.

Sign up to get the best content of the week, and great gaming deals, as picked by the editors.

[ For more curated Computing news, check out the main news page here]

The post Intel’s next-gen Battlemage GPU could take the fight to Nvidia’s RTX 4080 after all first appeared on

New reasons to get excited everyday.

Get the latest tech news delivered right in your mailbox

You may also like

Notify of
Inline Feedbacks
View all comments

More in computing