Today, Intel announced the Arc B580 and B570 ‘Battlemage’ GPUs, in what was perhaps one of the worst-kept secrets of the graphics card industry. While Intel won’t comment on future products, these are the first two of what we expect will eventually be a full range of discrete GPUs for the Battlemage family, designed for both desktop and mobile markets. The Arc B580 with 12GB of VRAM debuts at $248, while the B570 comes equipped with 10GB of VRAM and retails for $219.
Swipe to scroll horizontally
| Graphics Card | Launch Price | Xe-Cores | VRAM Capacity (GiB) | TBP | GPU Cores (Shaders) | XMX Cores | Ray Tracing Cores | Graphics Clock (MHz) | Max Boost Clock (MHz) | VRAM Speed (Gbps) | VRAM Bus Width (bits) | Render Output Units | Texture Mapping Units | Peak TFLOPS FP32 | Peak XMX FP16 TFLOPS (INT8 TOPS) | Memory Bandwidth (GBps) | Interface | TBP (watts) | Launch Date |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Arc B580 | $249 | 20 | 12 | 190W | 2560 | 160 | 20 | 2670 | 2850 | 19 | 192 | 80 | 160 | 14.6 | 117 (233) | 456 | PCIe 4.0 x8 | 190 | Dec 13, 2024 |
| Arc B570 | $219 | 18 | 10 | 150W | 2304 | 144 | 18 | 2500 | 2750 | 19 | 160 | 80? | 144 | 12.7 | 101 (203) | 380 | PCIe 4.0 x8 | 150 | Jan 16, 2025 |
We’ve known the Battlemage name officially for a long time, and in fact, we know the next two GPU families Intel plans to release in the coming years: Celestial and Druid. But this is the first time Intel has officially spilled the beans on specifications, pricing, features, and more. Most of the details line up with recent leaks, but Intel also gave some concrete performance expectations. Let’s start there, as that’s probably the most interesting aspect of the announcement.
Intel made two important comparisons with the Arc B580 announcement. First is how it stacks up against the existing Arc A750, and second is how it compares with the Nvidia RTX 4060. The testing was done at 1440p, as Intel says that’s the target resolution for its new GPU. Nvidia said the RTX 4060 targeted 1080p gaming, though we’d argue that was far more about saddling the GPU with only 8GB of VRAM than it was about the raw compute available.
Starting with its own A750, Intel shows a performance uplift of 24% on average across an extensive 47-game test suite. Twenty of the games have XeSS support, which is enabled in the testing, but since it’s Intel vs. Intel, that shouldn’t have a major influence on the results. The performance improvement ranges from 0% — League of Legends and DOTA 2 are almost completely CPU limited — to as much as 78% in Fortnite, with 31 games showing an 18–42 percent improvement.
Against the RTX 4060, Intel shows a 10% performance advantage across the same 47-game test suite, but this time without enabling XeSS or DLSS upscaling. We feel that’s a fair comparison, since upscaling algorithms differ in how they work and the resulting image quality. There’s a lot more variability with the RTX 4060 comparison as well, with a delta of -17% to as much as +43%. Six of the games show a minor to modest performance loss, while ten of the games show more than a 20% improvement.
It’s worth noting here that our own GPU benchmarks hierarchy puts the RTX 4060 14% ahead of the Arc A750 (at 1440p ultra testing), with the Arc A770 16GB 11% ahead of the A750. By that metric, Intel’s numbers match up with our own. That means it’s a reasonably safe bet that the B580 will offer performance roughly on par with the RX 7600 XT (23% faster than the A750), or alternatively 2019’s Nvidia RTX 2080 Super.
Earlier rumors suggested the B580 would match the RTX 4060 Ti, but our testing puts that GPU 41% ahead of the A750. Of course, the test suites do matter, and we’re presently in the process of revamping all of our GPU testing for this launch and into 2025. Quite a few of the newer games will likely show more of a performance hit on 8GB cards like the base 4060 Ti.
Perhaps just as important as the performance is the intended pricing. We’ve already spoiled that with the headline, but now, with performance data in tow, the $249 launch price looks even more impressive. That puts the Arc B580 into direct price competition with AMD’s RX 7600 8GB card, while Nvidia doesn’t have any current generation parts below the 4060 — you’d have to turn to the previous generation RTX 30-series, which doesn’t make much sense as a comparison point these days as everything besides the RTX 3050 isn’t really in stock anymore.
But again, we have to exercise at least some caution here. The RTX 4060 launched in mid-2023, about 18 months ago. It’s due for replacement in mid-2025. Whether that will happen or not remains to be seen, but beating up on what was arguably one of the more questionable 40-series GPUs isn’t exactly difficult. And that brings us to the specifications for Intel’s new Arc B580 and B570.
Swipe to scroll horizontally
| Graphics Card | Arc B580 | Arc B570 |
|---|---|---|
| Architecture | BMG-G21 | BMG-G21 |
| Process Technology | TSMC N5 | TSMC N5 |
| Transistors (Billion) | 19.6 | 19.6 |
| Die size (mm^2) | 272 | 272 |
| Xe-Cores | 20 | 18 |
| GPU Cores (Shaders) | 2560 | 2304 |
| XMX Cores | 160 | 144 |
| Ray Tracing Cores | 20 | 18 |
| Graphics Clock (MHz) | 2670 | 2500 |
| Max Boost Clock (MHz) | 2850 | 2750 |
| VRAM Speed (Gbps) | 19 | 19 |
| VRAM Capacity (GiB) | 12 | 10 |
| VRAM Bus Width (bits) | 192 | 160 |
| L2 (L1) Cache Size MiB | 18 (5) | 13.5 (4.5) |
| Render Output Units | 80 | 80? |
| Texture Mapping Units | 160 | 144 |
| Peak TFLOPS FP32 | 14.6 | 12.7 |
| Peak XMX FP16 TFLOPS (INT8 TOPS) | 117 (233) | 101 (203) |
| Memory Bandwidth (GBps) | 456 | 380 |
| Interface | PCIe 4.0 x8 | PCIe 4.0 x8 |
| TBP (watts) | 190 | 150 |
| Launch Date | Dec 13, 2024 | Jan 16, 2025 |
| Launch Price | $249 | $219 |
Battlemage will have higher GPU clocks, up to 2670 MHz on the B580 and 2500 MHz on the B570. Except, as with Alchemist, these are more approximate clocks and the real-world clocks are likely to be higher. Arc A770, for example, had a stated 2100 MHz graphics clock, a conservative estimate based on running a bunch of workloads. The maximum boost clock was 2400 MHz; in our testing, we saw average clocks of 2330–2370 MHz.
Each Xe-core has eight vector units and eight XMX units. That’s half the count of Alchemist, except each unit is twice as wide — 512 bits for the new vector engines and 2048 bits on the XMX units. In practical terms, that means the total number of vector or matrix instructions executed per Xe-core remains the same: 256 FP32 vector operations (using FMA, fused multiply add), and 2048 FP16 or 4096 INT8 operations on the XMX units.
Boost clocks multiplied by core counts give us the total theoretical compute, but here’s where things get interesting. If you compare the raw specs for the new Battlemage GPUs with the raw specs for the existing Alchemist GPUs, it looks like Intel took a step back in certain aspects of the design. For example, the A750 has 17.2 TFLOPS of FP32 compute, while the new B580 ‘only’ offers 14.6 TFLOPS. This is where architectural changes play a role.
One of the big changes is a switch from native SIMD32 execution units to SIMD16 units — SIMD being “single instruction multiple data” with the number being the concurrent pieces of data that are operated on. With SIMD32 units, Alchemist had to work on chunks of 32 values (typically from pixels), while SIMD16 only needed 16 values. Intel says this improves GPU utilization, as it’s easier to fill 16 execution slots than 32. The net result is that Battlemage should deliver much better performance per theoretical TFLOPS than Alchemist.
We saw this when we looked at Lunar Lake’s integrated graphics performance. The new Xe2 architecture in Lunar Lake has up to 4.0 TFLOPS of FP32 compute while the older Xe architecture in Meteor Lake has 4.6 TFLOPS, but Lunar Lake ultimately ended up being 42% faster at 720p and 32% faster at 1080p. Other factors are at play, including memory bandwidth, but given Intel’s claimed performance, it looks like Battlemage will deliver a decent generational performance uplift.
In terms of the memory subsystem, there are some noteworthy changes. The B580 will use a 192-bit interface with 12GB of GDDR6 memory, while the B570 cuts that down to a 160-bit interface with 10GB of GDDR6 memory. In either case, the memory runs at 19 Gbps effective clocks. That results in a slight reduction in total bandwidth relative to the A580 and A750 (both 512 GB/s), with the A770 at 560 GB/s. The good news is that these new budget / mainstream GPUs will both have more than 8GB of VRAM, which has become a limiting factor on quite a few newer games.
Battlemage will also be more power efficient, as the total graphics power is 190W on the B580 compared to 225W on the A750. That means Battlemage delivers 24% more performance while using 16% less power, a net potential improvement of nearly 50% (depending on real-world power use, which we haven’t measured yet). Intel also claims a 70% improvement in performance per Xe-core, based on the architectural upgrades, which we’ll discuss more in the architecture section.
Part of the power efficiency improvements come from the move to TSMC’s N5 node versus the N6 node used on Alchemist. N5 offered substantial density and power benefits, and that’s reflected in the total die size as well. The ACM-G10 GPU used in the A770 had 21.7 billion transistors in a 406 mm^2 die, and BMG-G21 had 19.6 billion transistors in a 272 mm^2 die. That’s an overall density of 72.1 MT/mm^2 for Battlemage compared to 53.4 MT/mm^2 on Alchemist.
Note, however, that Nvidia’s various Ada Lovelace GPUs have overall transistor densities of 109–125 MT/mm^2 on TSMC’s custom 4N process. AMD’s RDNA 3 GPUs also have densities of 140–152 MT/mm^2 on the main GCD that uses TSMC’s N5 node. That means Intel still isn’t matching its competition on transistor density or die size, though AMD muddies the waters with RDNA 3 by using GPU chiplets.
And finally, Intel will use a PCIe 4.0 x8 interface on B580 and B570. Given their budget-mainstream nature, that’s probably not going to be a major issue, and AMD and Nvidia have both opted for a narrower x8 interface on lower-tier parts. Presumably, it determined that there was no need for a wider x16 interface, and likewise, there wasn’t enough benefit to moving to PCIe 5.0 — which tends to have shorter trace lengths and higher power requirements.