The truth is that the chip giant already gave us enough details about its new graphics architecture during the presentation of Lunar Lake, the SoC generation that has an integrated Intel Arc Battlemage GPU, but we were waiting to know all the keys to this generation. applied to graphics cards general consumption.

This wait is over and on schedule. Intel told us yesterday all secrets by Arc Battlemage in an online event and today we can finally share with you all the keys to this new graphics generation based on the Xe2 architecture. The first thing we need to be clear about is that this new architecture is an evolution, not a revolution.

There is a very big difference in both, which I will explain to you now. Intel Xe2 architecture maintains the same foundation which we saw in the Xe architecture, which means we have the same partitioning into Xe blocks (in this case second generation) with vectorization units, geometry and rasterization.

No revolution is happening because there is no drastic change at the architectural levelbut there have been important developments with the Xe2 architecture, as Intel has introduced notable changes that are not limited to performance improvements and that affect the central pillars of the graphics core.

Key changes in Intel Arc Battlemage

• Improves GPU utilization and workload distribution.
• Increases performance on the Xe2 core.
• Improvements in the blending subsystem (color output) and rendering preloading.
• New ray tracing cores with 3 through pipes, 18 box intersection units and two triangle intersection units.
• Improved micro and macro analysis of all graphics acceleration features.
• Up to three times higher performance when generating vertices and mesh shaders.
• ALU with native SIMD16 support and improved XMX tools.
• Native support for indirect execution.
• Support for out-of-order sampling.
• Improvements to template caching, compression, and primitives.

The Xe2 core consists of 8 512-bit vector engines8 x 2048-bit XMX matrices, ray tracing acceleration core, thread classification block, sampler and 256KB of shared L1 cache.

As we can see, this architecture covers all fronts, because yes hardware specialized for AI acceleration as well as ray tracing. It also maintains thread management optimization with classification blocks, something that works in a similar way to the SER technology present in the GeForce RTX 40.

Matrix units can be worked with INT2, INT4, INT8, FP16, BF16 and TF32. Native SIMD16 support prevents Intel Arc Battlemage from having to resort to emulation like Arc Alchemist did, and other optimizations have been introduced to improve overall performance and FP64 (double precision) performance.

The ray tracing acceleration core was another big benefit of this new architecture. They increased by 50% the number of passing pipes and the calculation of frame penetration units, and folded the number of units intersecting the triangle and the BVH cache.

A view of the BMG-G21 graphics core

It is the first graphics core that the chip giant will use in its new generation Intel Arc Battlemage. It consists of:

• 5 render blocks.
• 20 Xe2 cores.
• 160 nut units.
• 20 cores to accelerate ray tracing.
• 20 texture swatches.
• 10 pixel backends.
• 18 MB L2 cache.
• 192-bit bus with GDDR6 memory support.
• Two multi-format X encoders.

Intel promises up to 70% more performance

According to Intel, the Xe2 core has up to 70% more performance than the first generation Xe core and offers 50% more power per watt consumed. Keep in mind that when we rasterize, we’re talking about raw energy data, so without taking into account technologies like ray tracing or AI.

As expected, AI played a big role in Intel’s Arc Battlemage presentation. The chip giant not only confirmed improvements to Intel XeSS Super Resolution, its intelligent scaling technology, but also announced Intel XeSS Frame Generation release.

The frame generation is coming to Intel XeSS2

With Intel XeSS Super Resolution it is possible improve game performance by up to 80% in 1440p, it depends on each specific game and its dependence on the processor, and the improvement can be up to 170% if we activate ray tracing.

It is a very important technology, but there comes a point where it is no longer able to offer a higher level of performance because of the reduction of the base resolution we run into a major bottleneck caused by the CPU. This is something we already saw at the time with NVIDIA DLSS and AMD FSR, two ecosystems that eventually made the leap to frame generation to overcome this limitation.

The Santa Clara giant took exactly the same route with Intel XeSS2, which it now includes AI accelerated image generation. This technology is accompanied by another focused on reducing latencyessential addition to counteract the latency spikes that occur when using frame generation in games.

According to data provided by Intel, thanks to its latency reduction technology, we can Reduce response time by up to 45%. This technology is implemented at the driver level, so we can activate it with a simple click.

Intel XeSS2 frame generation uses specialized hardware and uses information from the previous frame to generate a new frame. Motion vectors and image depth are analyzed and all information obtained is moved to reprojection enginesconsisting of an optical flow unit, another motion vector unit and a unit that is responsible for mixing and fusion to create the final image.

Thanks to the frame generation present in Intel XeSS2, this is possible multiply the frame rate per second in games up to 3.9 times. As this is a hardware-accelerated and AI-powered technology, its resulting quality should be higher than what AMD offers with FSR 3.1. Its big rival will be NVIDIA’s frame generation, which also uses AI and dedicated hardware.

And speaking of drivers, Intel also confirmed it driver level enhancements including support and performance as well as configuration, management and overclocking tools.

Intel Arc B570 and B580 specifications

These are the first two graphics cards that Intel will introduce as part of its new generation Arc Battlemage. Both will go to cover middle band, but will do so at two different price and performance levels.

Intel Arc B570

• 5 render blocks.
• 18 Xe2 cores.
• 144 matrix units.
• 18 cores to accelerate ray tracing.
• 2500 MHz GPU.
• 160-bit bus.
• 10 GB of GDDR6 memory.
• Bandwidth 380 GB/s.
• PCIe Gen4 x8 interface.
• Two multi-format X encoders.
• 150W TBP.
• Three DisplayPort 2.1 outputs and one HDMI 2.1.
• 203 TOPs in INT8.
• You need an additional 8-pin power connector.

Performance, price and launch

This graphics card will compete in the lower mid-range market. It is designed for move games to 1080p resolutionand will be available from January 16, 2025 219 dollars. It will compete in the Radeon RX 6600 series.

In Spain it should be around 229 euros, approximately.

Intel Arc B580

• 5 render blocks.
• 20 Xe2 cores.
• 160 nut units.
• 20 cores to accelerate ray tracing.
• 2670 MHz GPU.
• 192-bit bus.
• 12 GB of GDDR6 memory.
• Bandwidth 456 GB/s.
• PCIe Gen4 x8 interface.
• Two multi-format X encoders.
• 190W TBP.
• Three DisplayPort 2.1 outputs and one HDMI 2.1.
• 233 TOPs in INT8.
• You need an additional 8-pin power connector.

Performance, price and launch

With this model, Intel is targeting the middle class. It is designed for play in 1440p resolution with guarantees and promises to be up to 10% more powerful than the NVIDIA GeForce RTX 4060 in games where graphics memory is not a limiting factor.

It will hit the market on December 13 with a price tag 249 dollars. It should be in Spain between 279 and 299 euros.