Despite a rocky start, Intel’s Arc GPUs have earned their place among the best graphics cards you can buy. And now, after a long wait, we finally know what the next generation has in store.
Code-named Battlemage, Intel’s next-gen graphics cards bring a slew of improvements at a competitive price point. Here’s everything we know about Intel Arc Battlemage, from specs to prices and availability.
Intel Battlemage specs
Arc B580 | Arc B570 | |
Xe-cores | 20 | 18 |
Shaders | 2,560 | 2,304 |
XMX cores | 160 | 144 |
RT cores | 20 | 18 |
Maximum frequency | 2,850MHz | 2,750MHz |
VRAM | 12GB GDDR6 192-bit | 10GB GDDR6 160-bit |
Bandwidth | 456Gbps | 380Gbps |
TBP | 190W | 150W |
As of right now, Intel has announced two GPUs, and they’re fairly close in terms of specifications. The current top graphics card is the Arc B580, followed by the Arc B570. There have been plenty of rumors about other possible releases, but Intel’s made no mention of them at this time. Let’s dig into what we have.
On the surface, the numbers aren’t in favor of the B580, as it actually comes with fewer Xe cores than its predecessor, the A580 — 20 versus 24. The same goes for the number of ray tracing cores. However, as you’ll find out later, Intel still claims to deliver sizable performance improvements.
There’s been a huge uplift in maximum clock speeds, with the Arc A580 reaching a whopping 2,850MHz maximum frequency, and the B570 trailing closely behind at 2,570MHz. The GPUs also received a much-needed boost to VRAM, now offering 12GB and 10GB, respectively — although Intel is sticking to GDDR6 memory whereas Nvidia’s RTX 50-series is likely to upgrade to GDDR7.
Despite the larger memory capacity, the bus width is actually smaller than on the Arc A580, with 192-bit on the B580 and an unusual 160-bit on the B570. Both have memory clocked at 19Gbps, resulting in total bandwidth of 456GB/s for the B580 and 380GB/s for the B570. That’s a drop compared to the A580, which offered 512GB/s.
Although some of the specs are better and some seem worse, Intel’s boost to VRAM marks a huge change for the B580, and allows the company to advertise it as a 1440p graphics card. In fact, Intel is mostly comparing the B580 not to its predecessor, but to the Arc A750 — which, while not quite the flagship of that lineup, was a comparatively higher-tier GPU.
Following Intel’s lead, comparing the B580 to the A750 reveals that the new cards are also more power-efficient, with a total board power (TBP) of 190 watts for the B580 and 150W for the B570. Meanwhile, the A750 sports 225W. However, there’s still a gen-on-gen power consumption hike if you compare it to the A580, which has a TBP of 185W — but it’s a negligible difference.
The B580 and the B570 both use the PCIe 4.0 x8 interface, which is not unexpected given that these are budget GPUs.
This generation marks a departure from Intel’s first gen, when the company opened the lineup with the flagship Arc A770 and its cutdown sibling, the Arc A750. This leaves many wondering: What else does Intel have in store for Battlemage? Buckle up, because we’re moving away from facts and toward leaks and rumors because Intel itself has yet to speak about a possible B770 or any other Battlemage graphics card.
Originally, leakers like RedGamingTech claimed the flagship card would come with 64 Xe cores — double what’s available on the A770 we have now, and with an entirely new architecture in tow. At the beginning of the year, RedGamingTech revised that rumor, saying that the flagship GPU would come with 56 Xe cores. That’s still a massive bump over the A770. The latest speculation is a little bit different, however.
Frequent Twitter leaker Harukaze5719 found some shipping manifests that gave us insight into the rumored specs of Intel Battlemage. According to the manifests, Battlemage will come in two flavors: X2 and X3. The X2 card is said to be the flagship, and the specs are drastically cut down compared to what RedGamingTech speculated on earlier this year. Given that we haven’t seen any mentions of X3 in relation to the B580, take this with some skepticism.
BMG X2
BMG X3https://t.co/O5v3Tl1Wno pic.twitter.com/lr8uD3bCCz— 포시포시 (@harukaze5719) June 29, 2024
The X2 card is said to come with 32 Xe2 cores, which translates to a (presumed) 4,096 stream processors (SPs) and 512 execution units (EUs). That’s exactly the same as Intel Arc A770, but the Battlemage architecture is also said to provide some improvements regardless of the number of EUs. The second card was said to sport 28 Xe2 cores (3,584 SPs and 448 EUs).
According to speculation, the so-called Battlemage GPU with 56 Xe2 cores was canceled. Intel appears to be adopting a similar strategy to AMD’s RDNA 4 in this generation by sticking to the budget part of the market. However, just a few months ago, the rumors were vastly different. Apart from the flagship card with much better specs, RedGamingTech claimed that the card could reach clock speeds of up to 3GHz and would come with a massive 112MB of L2 cache.
In addition to the model with 56 Xe Cores, the leakers had said that a model with 40 Xe cores was in the works. This one seems a bit more grounded in reality, packing the aforementioned cores, a 192-bit memory bus, and none of the Adamantine cache. RedGamingTech speculated that this design might claim the flagship slot, citing the poor margins that could come from the 56 Xe core model.
As things stand now, we have vastly different accounts from different sources, so it’s impossible to tell where Intel will go next.
Intel Battlemage pricing and release date
After months of speculation from leakers and scarce updates from Intel, the company announced Arc Battlemage on December 3, 2024. As a result, it managed to beat both Nvidia and AMD in launching next-gen graphics cards.
Both cards have different release dates and pricing. The Arc B580 is available from December 13 with a starting price of $250. Meanwhile, the B570 comes in cheaper at $220, but won’t become available until January 16.
Intel offers its own Limited Edition Arc B580, but there will also be partner designs from brands like ASRock, Acer, and Sparkle.
Although more GPUs are expected to follow, their release dates are currently unknown. A recent report tells us that Intel might wait as long as several months to launch the top-shelf Battlemage GPU, with Benchlife estimating that the card will arrive in the third quarter of 2025.
Intel Battlemage architecture
Intel calls the architecture behind Battlemage Xe2, following up on the Xe architecture we saw with Alchemist graphics cards. Here’s what we know about it so far.
First of all, Intel is once again using manufacturer TSMC for Battlemage, but it’s now employing the TSMC N5 process node as opposed to the N6 used in Alchemist. The new node comes with a boost to both density and power efficiency. While the BMG GPU sports fewer transistors than the ACM-G10 (21.7 billion versus 19.6 billion), the die size is much smaller, making for much greater density in Battlemage.
Intel has changed the number of vector and XMX units per Xe-core, cutting them in half. However, they’re now wider, reaching 2,048 bits for the XMX units and 512 bits for the vector units. As a result, the number of instructions executed on each core is unchanged from Alchemist. Tom’s Hardware reports that we’re getting 256 FP32 vector operations (using fused multiply add, or FMA), 2,048 FP16, and lastly, 4,096 INT8 operations on the XMX cores.
The publication also notes that the A750 has more teraflops of FP32 compute than the B580 — 17.2 TFLOPS compared to 14.6 TFLOPS — but that’s more of a specification than an issue with the Battlemage architecture at large, which still seems promising based on performance figures and which we’ll discuss in the next section.
With the Xe2 architecture, Intel targets higher GPU utilization, reducing software overhead and improving work distribution. Intel’s done a lot of work since Alchemist first launched, and Battlemage is here to hopefully reap the rewards. To that end, the company has talked about the various software improvements it has made since the launch of Alchemist — over 50 updates, in fact.
On the whole, Intel added support for Execute Indirect, switched to SIMD16 arithmetic logic units (ALUs), boosted the ray tracing units to add three traversal pipelines, and improved vertex and mesh shading performance per render. There have also been updates to the caching hierarchy for the VRAM, increasing L1 cache size. Battlemage also adds support for INT2 and TF32. Intel recorded a deep dive on the Xe2 architecture, which is worth checking out if you’re interested in learning more about it.
The current flagship is the BMG-G21 GPU, and it comes with five render slices at four Xe-cores each for a total of 20 Xe-cores. However, many leakers expect that we’ll see the BMG-G10 GPU one day, which could offer 32 Xe-cores, although it’s possible that Intel might have something more powerful up its sleeve.
Aside from the architectural changes, Intel also announced an update to its upscaling technology dubbed XeSS, which will now also feature frame generation (FG). Unlike Nvidia’s DLSS 3, XeSS FG is not limited to Battlemage and will run on any GPU with XMX AI cores, which includes both Alchemist and Battlemage. Intel also added XeSS Low Latency (XeLL), which will be enabled by default when FG is running, as the latter increases latency.
Intel Battlemage performance
Although we will need to test Battlemage ourselves to find out its true performance, Intel has shared a number of highly promising benchmarks of the B580. The card is matched up against the A750. Let’s check it out.
For starters, on average, Intel claims that the B580 will be up to 24% faster than the A750, and this is at 1440p on ultra settings. Some games don’t see much of an improvement, but a lot of it comes down to how much a particular title enjoys the extra 4GB of VRAM on the B580 compared to the A750.
As far as competition, Intel compares the B580 to Nvidia’s $300 RTX 4060, which is an interesting choice. Nvidia won’t have a direct upgrade to the RTX 4060 for a few months (if the leaks are correct), so Intel may get some breathing room to let its GPU shine, and it does a good job in the above benchmark. On average, without DLSS and XeSS, the B580 is said to be 10% faster than Nvidia’s RTX 4060.
Intel also boasts huge improvements in its proprietary XeSS technology, sometimes doubling performance compared to when the tech is not in use.
One final metric that Intel looked at was performance per dollar. At $250 and $220, the B580 and the B570 are the cheapest current-gen GPUs on the market. Seeing as Intel claims to deliver a performance upgrade over both the Nvidia RTX 4060 and the AMD RX 7600, the performance-per-dollar metric looks fantastic, with a 24% win for Intel in rasterization (versus Nvidia) and a 20% improvement in ray tracing.
These benchmarks are just the beginning. Once we’ve had the chance to test the card for ourselves, we will be able to see how it matches up against its last-gen counterparts and its AMD and Nvidia equivalents.