On March 30, Intel officially released the Intel Iris ARC mobile discrete graphics, code name Alchemist (Alchemist), Intel's first consumer-side unique display product launch has been 24 years, after which Intel's unique display product development has stagnated and focused on core display development.
After years of technology accumulation, Intel previously launched DG1 graphics for the server market, and this year officially launched Intel Iris ARC exclusive display products for consumers, and the first batch of A series products launched for mobile terminals includes three models of Sharp 3/5/7.
Among them, Intel Iris 3 is mainly for the mainstream game market, Ruixuan 5 is mainly for the performance game market, and Ruixuan 7 is mainly for audiophile hardcore games. Intel introduced the A350M and A370 M products of the mobile A series. The new Intel Iris Graphics support XeSS hypersampling, full AV1 hardware acceleration, Smooth Sync jitter filtering, deep Link technology, and provides a full range of scenarios such as gaming, creative design, and power control.
The first intel iris ARC is the Samsung Galaxy Book2 Pro thin and light product, which has obtained the stringent certification of Intel Evo and has been officially listed in overseas markets.
In the future, with the advantage of Intel's share of the processor market, there will be a large number of Notebook products equipped with Intel Iris ARC. Products certified by Intel Evo will also be further improved in battery life and display capabilities.
At present, Acer, Asus, Dell, Haier, HP, Lenovo, MSI, Samsung, Intel NUC and other brands or products have the intention of launching a sharp unique notebook, through the sharp unique display, Intel can also integrate its own products in the future and launch first-party Intel notebooks.
The supporting Intel Iris Control Panel is also launched simultaneously with the launch of Intel Iris Display products, which integrates functions such as automatic driver update, performance monitoring, performance tuning, live stream management, game highlight moment generation, event promotion, etc., and can be used without mandatory login.
Next, we will take a detailed look at the underlying architecture and technical highlights of the new Intel Iris ARC discrete graphics.
Underlying architecture
Intel Iris ARC unique display products are based on the Intel Xe HPG architecture, the core uses the built-in XMX Xe core, including Xe media engine, Xe display engine, and Xe graphics pipeline three core functions.
With Xe HPG microarchitecture, Intel Iris Graphics has a lot of flexibility during development, and render slices are the basic modules of Xe HPG microarchitecture, each Xe HPG render slice contains 4 Xe cores, 4 optical tracing units, 4 samplers, geometry engine, raster engine, HiZ engine, and 2 pixel backend components.
Each Xe core contains the XMX matrix engine, XVE moderate engine, light chasing unit, sampler, etc., which constitute a complete Xe core, but also the basic computing unit of Xe HPG microarchitecture, which is different from the previous eu concept of execution unit, through the rendering slice composed of 4 Xe cores, in different combinations to form different SoCs to form different product forms.
Intel Iris Graphics forms different product lines by overlaying rendering slices, with a minimum of 2 and a maximum of 8, and a variety of products through different forms of combination. For light chasing and DX12 Ultimate, the Xe HPG microarchitecture is also well supported.
Back on the Xe cores, each Xe core provides 16 x 256-bit XVE vector engines, 16 x 1024-bit XMX matrix engines, and 192 KB of shared L1 cache. The XVE moderate engine is used to perform traditional image processing calculations, while the XMX matrix engine is mainly used for AI acceleration.
The XVE vector engine can perform 16 FP32 operations, 32 FP16 operations, and 64 INT8 operations per clock cycle, with dedicated FP floating-point execution interfaces and shared INT/EM execution interfaces. The XMX matrix engine can perform 128 FP16/BF16 operations, 256 INT8 operations, and 512 INT4/INT2 operations per clock cycle.
The XMX hash rate gain is huge compared to traditional MAC or advanced DP4a, and we know that MAC is the basic SIMD vector instruction used in graphics, performing a total of 8 parallel multiplications and 8 parallel additions per clock cycle. The DP4a, on the other hand, is optimized for AI computations that don't require 32-bit precision, performing a total of 32 parallel multiplications per clock cycle, 32 cumulative operations, or a total of 64 operations per cycle, which is 4x better performance than standard SIMD MAC.
The XMX Matrix Engine takes it to the next level by pipelining multiplication plus 4 depths. As with DP4a, each operand is divided into 4 blocks, which are independently multiplied and added up—64 operations per stage—(shown by the purple tiles). With 4 stages, each clock produces 256 operations – a 16x performance increase over a traditional 32-bit SIMD MAC.
The best application of XMX's boost is the XeSS supersampling anti-aliasing technology, which provides higher performance in games than traditional high-resolution rendering, and generates beautiful high-resolution images from low-resolution renderings through neural network-assisted motion vectors, somewhat similar to NVIDIA DLSS.
XeSS supersampling anti-aliasing technology will officially arrive this summer, and the first games to support XeSS include 14 games such as Tomb Raider: Shadow, Super Touring Car Racing: Legends, Ghost Line: Tokyo, Death Stranding, Blood Hunt, CHORVS, Arcadegeddon, Killer 3, and more.
With the Xe media engine, Sharp Graphics supports codecs in a variety of mainstream formats, including H.265/HEVC, H.264/MPEG-4/AVC, VP9, and AV1.
Hardware codec acceleration for AV1 support Intel Iris Graphics is the first GPU provider to offer these formats with extremely low processor utilization. Due to av1's excellent efficiency, AV1 will also become the mainstream video format in the future, it is more efficient than H.264 and HEVC, can achieve better picture quality with lower bandwidth and smaller file size, and AV1 has no royalties.
Intel Iris graphics on AV1 hard decoding capabilities compared to the traditional soft decoding in the encoding speed of 50 times, the current FFMPEG, Handbrake, Adobe Premiere Pro, Davinci Resolve, XSplit have integrated Ruixuan AV1 hard decoding support.
The Xe display engine is mainly built for the current stage and future display technology, and at this stage, Intel Iris Graphics supports HDMI 2.0b, DP 1.4a, and DP 2.0 10G will also be supported. With Intel Iris Graphics, players can enjoy the highest output of 2 8K@60 HDRs or 4 4K@120 HDRs.
In gaming scenarios, Intel provides several synchronization technologies to help players have a better experience, including VESA standard Adaptive Sync ripstop technology supported by Intel Iris Graphics. Speed Sync, a new technology that provides acceleration for the current frame of the game, speed Sync achieves a low-latency-free tear-free effect by turning off V-Sync and rendering the frame as a whole.
Smooth Sync is another screen optimization technology introduced by Intel that blurs the boundaries of two torn frames to reduce visual distortion to make the picture look more coherent and smooth.
Performance
The new Intel Iris Monochrome display product contains a total of 2 different SoC designs, codenamed ACM-G10 and ACM-G11, of which ACM-G10 contains a total of 32 Xe cores and optical tracking units, 16MB L2 cache and 256-bit GDDR6 interface, 16 PCIe 4.0 interfaces, and ACM-G11 contains 8 Xe cores and optical tracing units, 4MB L2 cache, 96-bit DDR6 interface, and 8 PCIe interfaces. Both chips contain 2 Xe multifunction codec engines and 4 image output engines.
Regarding the frequency problem, we know that different frequencies require elegance and power consumption are not the same, in fact, according to the daily use of the scene, the notebook often presents a dynamic distribution of frequency power consumption in different load scenarios. Based on this distribution, Intel Iris graphics cards in the allocation parameters are, often set a representative load, and then according to the frequency of this load, the frequency of the graphics card is defined. Different platforms have different TDP, and under the more relaxed TDP restrictions, the distribution range of clock frequencies will also increase as a whole.
Therefore, Intel divided the first five graphics products of the A series according to this, of which the first Gli-Hyun 3 A370M contains 8 Xe cores and optical chase units, the main frequency is 1550MHz, 8GB GDDR6 64-bit video memory, and the TGP is between 35-50W; the Sharp 5A550M contains 16 Xe cores and optical tracking units, the main frequency is 900MHz, 8GB GDDR6 128-bit video memory, TGP is between 60-80W; the Ruixuan 7A770M contains 32 Xe cores and optical chase unit, 1650MHz, 16GB GDDR6 256 bit video memory, TGP between 120-150W. Ruixuan 3 products have been officially launched, and Ruixuan 5/7 will be officially launched this summer.
In terms of game performance, the first batch of listed Ruixuan A370M graphics cards are mainly for medium and high-quality games, focusing on large-scale games with scenes under 1080P frames. Compared to the 96EU Xe core display, it has a frame rate of more than 60 frames.
In the game scenes such as Fortnite and GTA V that require high frame rates, the frame rate of the Ruixuan A370M has exceeded 90 frames in high quality, which has reached a smooth level.
In the creative production scenario, compared with the integrated graphics card of the 12th generation Core, the performance on the platform equipped with the A370M discrete graphics card has also been significantly improved. In terms of video encoding and decoding, davinci Resolve, for example, can improve the performance of 4K H.264 to H.265 by up to 60%. In terms of AI-related functions, such as the two application scenarios in Adobe Promiere Pro, there is a doubling of performance improvements.
The improvement in the creative scene depends not only on the graphics card itself, but also on the huge improvement brought by Intel's new Deep Link technology. Let's take a look at how Deep Link works.
Intel Deep Link technology
Intel Deep Link technology is different from the previous pure dynamic power sharing, Intel Iris Graphics and Intel 12th generation Core processors in addition to the dynamic sharing of power consumption, but also the introduction of super coding and super computing power capabilities.
Dynamic power sharing maximizes cpu or GPU performance within the limits of system power consumption. Intel has been exploring this technology for a long time. Back in 2016, in the Kobe-Lake G era, we had the first version of Dynamic Power Sharing, dynamically distributing power between CPU die and GPU die.
This feature is also being further applied between today's ADL and A-series discrete graphics cards, where when running a load, if the CPU needs more power, the power is distributed more to the CPU, and vice versa for the GPU, with the ultimate goal of making the load perform better.
The second technology is supercoding, which is originally designed to improve the efficiency of codecs for end users. In the previous codec process, the coding work was usually placed on the codec of a graphics card, and the coding efficiency became the performance bottleneck of the entire process; in fact, the current Intel notebook system, such as the system equipped with the 12th generation Core processor and the Sharp A series discrete graphics, the integrated graphics and discrete graphics have hardware coding capabilities. Therefore, the super coding technology is to use the codec engine of two graphics cards at the same time to greatly improve the codec efficiency.
This collaboration is achieved through OneVPL's API interface. OneVPL is a cross-platform, open framework through which applications can identify and invoke multiple multimedia engines on the platform, making full use of video processing capabilities. When super encoding begins to work, groups of decoded raw frames are handed over to oneVPL via specific API functions, which are then assigned to different multimedia engines in groups and copied to the corresponding memory for caching. Regardless of how many frames each group has, the corresponding cluster or solo multimedia engine will begin to encode according to the set format. OneVPL will complete the subsequent packaging work, and stitch the encoded frames into a final video to output. This parallel processing is a significant increase in coding efficiency over a single graphics card.
There is also a logic similar to supercoding in terms of hash rate improvement, that is, to involve the entire system as much as possible, and the right module does the right thing, and the same logic is true of the technology of super computing power.
Notebooks powered by Intel Iris Discrete Graphics can benefit from the computing power of discrete graphics, but the compute engine is also available in the integrated graphics of Intel CPUs. The computing power is maximized by rationally distributing the load to different calculation engines. This uses the MLS framework in OpenVino to maximize the computing power.
MLS intelligently distributes load to different hashrate modules, helping MLS make decisions about applications or load characteristics such as latency sensitivity, throughput, performance requirements, and power consumption, and assigning load to discrete graphics, integrated graphics, or CPUs.
Through several key technologies of Deep Link, the combination of Intel Core Notebook + Intel Sharp Graphics brings a significant performance boost in the creative scene. The various modules of the system work more closely together, so that the performance of each module is fully released. Based on this philosophy, Deep Link combines modules on intel platforms to further improve overall efficiency.
summary
Intel in the dormant for many years, finally opened the road to the single display, the first batch of listed unique display products mainly for mobile terminals, with Intel's strong share in the processor field, the future Intel Iris unique display products will also become a powerful force after the N card, A card, the graphics card market will enter the "Three Kingdoms era". At a time when the price of the graphics card market is high, Intel's entry is a good thing for consumers, and more choices mean that the price war between products will start.
For the industry, Intel's i+i solution is not only conducive to Intel's overall control of the product, but also allows Intel to provide further consistency between developers and partners.
Intel's going to the sea will undoubtedly stir up the huge market of independent graphics cards, and we will wait and see how such a "three countries" situation will develop in the future.