Intel Arc Battlemage 20-Core GPU Spotted At 2.85GHz With 12GB VRAM
With that in mind, we're pretty excited for today's leak, which is a result from the Geekbench Compute database for an "Intel Xe Graphics RI" featuring 160 compute units, a peak frequency of 2850 MHz, and 12GB of video RAM. We're not completely clear on what Geekbench is detecting as "160 compute units", but if it's the individual Xe Vector Engines then we're looking at a Battlemage GPU with 20 "Xe2-Cores", by Intel's nomenclature.
Assuming that both Geekbench's count of compute units and our speculation on what it means are accurate, that makes this a relatively modest GPU in terms of compute resources—broadly comparable to something like AMD's Radeon RX 7600. However, the 12GB of onboard memory means that this part has either a 96-bit or 192-bit memory interface, not the 128-bit interface of that GPU. Given the relatively modest proportions of the chip, we're banking on this being an entry-level Battlemage part.
That conclusion is supported by the meager score produced by the GPU. 97943 isn't a "broken" result, but it's not great considering the aforementioned Radeon RX 7600 can easily produce a score twice as high. Of course, Intel has not prioritized OpenCL, preferring to focus on DirectML and its own OneAPI compute libraries; the much larger Intel Arc A770 barely scores higher than this despite being really a quite large GPU with nearly four times as many Xe vector engines. So saying, we expect that this score won't correlate to gaming performance.
Lunar Lake's Xe2 integrated GPU puts up a credible challenge against not only AMD's "Phoenix" and "Hawk Point" processors—those being the Ryzen 7040 series, Ryzen 8040 series, and Ryzen Z1 Extreme—but also even against the company's "Strix Point" parts using a larger RDNA 3.5 GPU. It offers truly impressive performance considering the power budget of the Lunar Lake processors. We're expecting big things from Battlemage, but Intel hasn't said a peep about dates. Rumors have vaguely posited a launch early in 2025. Let's hope it's sooner than later so that Intel can avoid the debacle of the Arc 1 launch, which was forced to compete against next-gen GPUs with newer architectures and manufacturing processes.