Intel debuts Arc discrete GPUs for laptops
Lower-end of the family to roll out first – and desktops will have to wait
Intel hopes to compete against Nvidia and AMD in the discrete GPU market with now-launched standalone graphics chips for laptops. These components are said to feature a slew of technologies designed to provide smoother gaming experiences and faster content-creation performance.
The chipmaker kicked off its Intel Arc A-Series on Wednesday with the lowest-end mobile GPU of three performance tiers, Intel Arc 3, which will arrive in laptops starting in April. But it comes with the same features as the more powerful Intel Arc 5 and Intel Arc 7 graphics that will start hitting the market in early summer.
For those eager to get their hands on Intel's discrete graphics cards for desktops, you will also have to wait, but the details Intel provided today about the new laptop GPU products will serve as a preview for what people should expect.
More than a dozen PC makers have lined up new thin-and-light laptop designs with Intel Arc 3 graphics, including Acer, Asus, Dell, HP, Lenovo and Samsung, and the notebooks will start at $899. Among the first to come out is the Samsung Galaxy Book2 Pro, which is among the Arc-powered devices verified to meet premium laptop standards in the Intel Evo program.
At its highest power, Intel Arc 3 laptop graphics can provide more than 60 frames per second at a 1080p resolution, according to the company. Intel said this is significantly higher than what the integrated Intel Iris Xe graphics in its 12th-generation Core CPUs are capable of, and it applies to several popular games at medium or high settings, including Hitman 3, Doom Eternal, Destiny 2 and Final Fantasy XIV.
When it comes to popular multiplayer games like Fortnite, GTA Online, Rocket League and Valorant, Intel Arc 3 can hit 90 frames per second or higher at 1080p on high or medium settings, Intel said.
If you're wondering how Intel Arc 3 compares to similar GPUs from Nvidia or AMD, the company did not provide such information, but with Arc-powered laptops arriving soon, we're hoping to see performance comparisons from reviews and users rather in the not too distant future.
As for content creation, Intel compared Intel Arc 3 to the aforementioned integrated graphics in 12th-gen Core CPUs and said it provides a 30 percent boost for the Handbrake video transcoding application and a 60 percent boost for Davinci Resolve handling a similar workload. It also provides an 80 percent boost for the AutoReframe function and a 140 percent boost for the Scene Edit Detect function in Adobe's Premiere Pro video editing software.
Intel has previously teased some features in the Intel Arc A-Series, formerly code-named Alchemist, including hardware-based ray tracing and XeSS, its AI-powered image upscaling technology. But the chipmaker divulged more information today, including the specs of Intel Arc 3, Intel Arc 5 and Intel Arc 7, plus new features meant to smooth motion and mitigate so-called "screen-tearing" in games.
All of the Intel Arc A-Series GPUs are based on Intel's Xe high-performance graphics microarchitecture, an offshoot of the Xe graphics architecture. Xe will also power Intel's upcoming Ponte Vecchio HPC accelerator that is going into the United States' Aurora supercomputer.
Two new features disclosed are Speed Sync and Smooth Sync, which are providing new ways to enable smoother experiences in games with any monitor. Intel is offering them as complementary to the standard Adaptive Sync feature, which matches the monitor's refresh rate with the frame rate of games to eradicate screen-tearing.
But while Adaptive Sync only works with variable refresh rate monitors, Speed Sync works with any monitor, and it's meant to fight against the screen-tearing that can happen in games running at high frame rates with VSync turned on in the graphical settings. It does this by virtualizing the frame buffer and ensuring the monitor is only displaying the latest frame.
Smooth Sync, on the other hand, fights against screen-tearing by blurring the fractured areas in the game's image, which essentially happens because the display is showing an old frame and a new frame at the same time. This is meant to benefit games running at lower frame rates.
- AMD to Intel: Take our GPU talent? Two can play that game
- AMD: Our latest, pricier mega-cache Epyc processors leapfrog Intel's
- Intel targets edgy 5G and AI with new chips, software
- Intel to spend €17bn on chip mega-factory in Germany
To help users adjust Intel Arc's graphical settings and keep on top of graphics driver updates, Intel is providing what it calls the Intel Arc Control application. The software comes with some handy features for gamers, like built-in streaming and automatic screen recording for games, and it also manages the downloading and installing of driver updates.
The software side
Intel is also providing a slew of software resources for developers who want to take advantage of Intel Arc GPUs. This includes Intel Graphics Performance Analayzers, the oneAPI Video Processing Library, the Intel Game Dev AI Toolkit and the Intel VTune Profiler.
As previously mentioned, all Intel Arc graphics chips come with hardware-based ray tracing, which works with DirectX 12 or VulcanRT software to deliver real-time simulation of light and reflections in games and simulation programs. They also all come with XeSS, which is Intel's answer to Nvidia's Deep Learning Super Sampling (DLSS) and AMD's FidelityFX Super Resolution upscaling technologies that are enabled in their respective GPUs.
Short for Xe Super Sampling, XeSS relies on hardware-based AI acceleration within the Intel Arc chip to upscale images using deep learning to improve overall image-quality, which has the net effect of improving overall performance in games.
How XeSS works is that a lower-resolution raster image from a game is fed into a neural network, and AI inferencing is applied to the image, using training data to generate something that looks much higher quality. But the difference between a regular high-quality image and a high-quality image generated by XeSS is that the latter requires much less processing, which, in turn, improves a game's frame rate.
Intel currently has 14 games lined up, including Death Stranding and Hitman 3, that will enable XeSS in early summer. The chip biz said it is working on enabling more titles over the next several months.
Another previously teased technology is Intel Deep Link, which enables Intel Arc GPUs to work with Intel CPUs and integrated graphics to provide performance improvements across various applications. The technology actually debuted in what was Intel's first discrete GPU from 2020, Intel Iris Xe Max, but the mobile GPU was essentially a test run for Intel's new graphics ambitions and had a very limited footprint in the laptop market, so this will be new for most users.
The features enabled by Deep Link include Dynamic Power Share, which allows the computer to figure whether to send more power to the Intel Arc GPU or Intel CPU to optimize performance. The company said this can improve performance by up to 30 percent for content creation applications and other demanding kinds of software.
Deep Link's Hyper Encode combines the processing power of media engines within the Intel Arc GPU and Intel CPU, which the company said can speed up video encoding by up to 60 percent compared to the integrated Iris Xe graphics found in its 11th-generation Core CPUs. Hyper Compute, on the other hand, combines the compute and AI engines of Intel's discrete and integrated graphics with an Intel CPU to speed up various workloads.
Specs and architecture details
Now onto the specs [PDF] you've been waiting to hear about: Intel Arc 3 actually consists of two laptop graphics chips, the A350M and the A370M, the latter of which enables 60 frames per second across several games, so something to be aware of. They both feature the same 4GB of GDDR6 memory and 64-bit memory bus width, but the two chips differ elsewhere to optimize performance for laptops with different thermal requirements.
This is where the differences are between the two Intel Arc 3 chips, and don't worry, we'll drill down into what some of these things do: The A350M comes with six Xe cores, six ray tracing units, a 1,150MHz graphics clock speed and a power range of 25 to 35 watts. The A370M, on the other hand, comes with eight Xe cores, eight ray tracing units, a 1,550 graphics clock speed and power range of 35 to 50 watts.
The mid-range Intel Arc 5 consists of one laptop graphics chip, the A500M, which comes with 16 Xe cores, 16 ray tracing units, a 900MHz graphics clock speed, 8GB of GDDR6 memory, a 128-bit memory bus width and a power range of 60 to 80 watts.
The high-range Intel Arc 7 consists of two chips. The A730M comes with 24 Xe cores, 24 ray tracing units, a 1,100MHz graphics clock speed, 12GB of GDDR6 memory, a 192-bit memory bus width and a power range of 80 to 120 watts. The more powerful A770M comes with 32 Xe cores, 32 ray tracing units, a 1,650MHz graphics clock speed, 16GB of GDDR6 memory, a 256-bit memory bus width and a power range of 120 to 150 watts.
All Intel Arc GPUs come with multiple Xe Media Engines, which, for the first time in a GPU, comes with hardware acceleration for AV1 encoding and decoding, which Intel said is up to 50 percent more efficient than the more common H.264 codec. In addition, they come with a Xe Display Engine, which can support up to one 1080p HDR display running at a 360Hz refresh rate or four 4K HDR displays running at 120Hz.
As indicated in the specs, every Intel Arc GPU comes with multiple Xe processing cores, which serve as the basic building block of every chip. These Xe cores are accompanied by ray tracing units, which, as you guessed, makes real-time ray tracing possible. Every GPU also comes with thread sorting units, render backends and geometry and rasterization blocks, which are essential to graphics processing.
Inside every Xe core are multiple components, but the most noteworthy one is the 1024-bit XMX Matrix Engine, of which there are 16 alongside the same amount of 256-bit Xe Vector Engines. What makes the Matrix Engine special is that it can speed up the GPUs' AI inferencing capabilities by up to 16 times to 256 operations per clock for INT8 math, compared to the GPU's vector capabilities. It can also acceleration operations for FP16, BF16, INT4 and INT2 math formats.
This AI acceleration is pivotal to how Intel's XeSS upscaling feature works, and Tom Peterson, an Intel fellow working on Intel Arc graphics, said XeSS is only the first algorithm that will take advantage of the chip's many Matrix Engines, so we're rather interested in where Intel takes this next.
"That technique is just the first of what we expect will be many different algorithms all using [the Matrix Engines]," he said in a briefing with journalists.
"I don't know if this is the real term, but I call it neural graphics, which is the combination of AI and traditional raster to get better images. This is sort of the future of rendering as you know, and it's going to become a much more important, significant part going forward." ®