8
Nearly four years after announcing its return to the consumer graphics card segment, Intel formalizes its first series of Arc GPUs. This first draft focuses on laptops with five interesting mobile models.
Led by Raja Koduri for four years, the team in charge of Intel’s graphics vision finally sees the end of the tunnel and delivers, as promised, a first series of consumer graphics cards. An event in the small world of tech gaming where Nvidia’s GeForces reign outrageously against AMD’s Radeons. A third actor therefore comes to interfere and, if he does not revolutionize the genre, brings in his luggage a series of Arc graphics cards that are resolutely interesting on paper.
Intel Arc A-Series: the Xe HPG architecture deciphered
As announced a few months ago, this first generation of dedicated graphics cards is called Arc Alchemist and is based on the Xe HPG architecture, designed for use gaming and creative. For this launch, Intel is focusing exclusively on the mobile sector, that is to say laptops. Two chips are currently produced, the ACM-G10 and ACM-G11, so as to offer integrators five graphics cards ranging from entry-level (Arc 3 A350M) to premium (Arc 7 A770M).
This launch is obviously an opportunity for Intel to tell us more about how its GPUs work. Like what AMD and Nvidia offer, Intel breaks down its chips into different blocks in order to deliver several variations of graphics cards. Arc GPUs thus integrate a more or less high number of blocks called Render Slice (RS): two on the entry-level ACM-G11 mobile chip and eight on the ACM-G10 for the mid-range and high-end.
These Render Slices are composed of several elements. First there are the Xe-Core calculation sub-blocks, four in number. Each of these Xe-Cores contains 16 vector calculation blocks (Vector Engines, XVE) each composed of 8 calculation units and 16 calculation units dedicated to AI (Matrix Engines, XMX). These XMX units have been specifically developed to accelerate AI-related processing and inference, similar to the GeForce RTX Tensor units. Moreover, it is the XMX units that will be used to execute the processing ofupscaling XeSS imaging — a technology that will compete with Nvidia’s DLSS.
Intel also offers concurrent execution, allowing calculations to be performed at the same time on these units. We had seen it on the GeForce RTX 30, this allows in particular to optimize performance on DLSS, which will result in improved performance under XeSS at Intel.
Each Xe-Core is accompanied by a processing unit dedicated to accelerating the effects of raytracing (Ray Tracing Unit, RTU). Intel has also decided to integrate these specific units into its entire product line, including the most entry-level versions. Asked about this, Intel told us that they were well aware that the cost of activating the raytracing on graphics performance is important, and that at first glance it may seem quite irrational to want to enable such effects on entry-level graphics cards — we saw this on the Radeon RX 6500s, for example, where similar activation makes no sense. The company is however rather confident, arguing that certain small effects can be envisaged and that it is also up to game developers to offer effects that are both relevant and not very greedy. Extensive program…
Intel has also developed a new hardware block dedicated to video, the Xe Media Engine. This block accelerates decoding up to 8K HDR at 60 Hz on 12 bits, and encoding up to 8K HDR on 10 bits. Four video codecs are supported with VP9, AVC (H.264, x264), HEVC (H.265, x265) and, above all, the promising AV1 which will eventually become the standard, especially in 8K (Netflix is slowly starting to deploy it, for example). The acceleration of AV1 encoding is a first in the field, Nvidia contenting itself with offering the decoding of this codec on its GeForce RTX 30. Most encoding software will be compatible at the launch of Arc, Intel setting the example from HandBrake, Adobe Premiere, DaVinci Resolve, Xsplit or even FFMPEG.
Finally, the block dedicated to the display takes the name Xe Display Engine. This can handle a maximum of four 4K HDR displays at 120Hz or two 8K HDR displays at 60Hz. Full HD and 1440p can be displayed at a maximum frequency of 360Hz. No HDMI 2.1, however: it have to make do with HDMI 2.0b and DisplayPort 1.4a – with future compatibility with DP 2.0 10G. Incidentally, Intel obviously mentions compatibility with the Adaptive Sync protocol, allowing the frequency of compatible monitors to fluctuate according to the image rate of the source (game, video, etc.).
An adaptive operating frequency
Intel also wanted to take a quick look at the operating frequency of its GPUs. Not surprisingly, this frequency is not fixed, but variable depending on the load applied to the GPU, and within the limit of the energy envelope allocated to it. On a rather light activity, the frequency will tend to be high while it will be less so on a more sustained activity. The firm tells us that the frequency given in the technical characteristics ultimately corresponds to a kind of average that we can expect.
In addition, laptop manufacturers will benefit from a certain latitude on the energy envelope of Intel GPUs. Depending on the chassis and its cooling capabilities, OEMs will be able to define a higher or lower energy envelope for a given chip. Two GPUs with the same name will therefore not behave in the same way in games, complicating the equation when buying a computer. Again, this is an approach that is reminiscent of Nvidia’s with its GPU Boost frequency and fairly wide energy envelopes on the mobile GeForce RTX 30s.
Arc A-Series: entry-level Arc 3 first
As we saw above, Intel currently produces two separate chips, the ACM-G10 and ACM-G11. The first is intended to power the mid- and high-end versions of Arc A-Series mobile graphics cards. The second is intended for entry-level versions. For now, five graphics cards are on the program: Arc 3 A350M, Arc 3 A370M, Arc 5 A550M, Arc 7 730M and Arc 7 770M. Only the first two will be integrated for the moment, the other three not being expected before the start of summer 2022.
These two small graphics cards are quite modest, simply carrying 4 GB of GDDR6 memory on a 64-bit bus and 6 or 8 Xe Cores (for 768 and 1024 processing units) and as many units dedicated to raytracing. Obviously, these are energy efficient chips with an envelope that can be adjusted between 25 and 35 W on the A350M and between 35 and 50 W on the A370M.
As much to admit that with such specifications, these two small graphics cards will have no vocation to titillate laptops gaming equipped with large GeForce RTX 30. During its presentation, Intel also refrained from any comparison with the competition, contenting itself with opposing its dedicated graphics cards to its integrated Iris Xe solutions. The tone is set: an Arc 3 A370M will target the game at 60 fps in Full HD on light games set to an average level of graphic effects. For competitive games, the promise is given of running at around 90 fps, always in Full HD and with medium details. To see how these cards will really behave against the GeForce RTX 3050 and other GeForce MX570 from Nvidia, for example.
We now look forward to getting our hands on the first laptops with Intel Arc graphics cards. The Samsung Galaxy Book2 Pro will be one of the first laptops to host an Arc 3. A dozen manufacturers will follow suit with references from Acer, Asus, Dell, Msi or even Lenovo and HP, to name a few. only a few.
The starting price is given at $899, which should translate to around €1000 in France. Hopefully the desktop models will arrive soon enough, probably in early summer.