At the end of March 2022, after a fairly brief conference, Intel once again launched its own exclusive product line "ARC" on a large scale after more than 20 years. Released two core-scale discrete graphics products across the desktop and notebook markets in one go, totaling at least eight sub-models.
Obviously, in today's era of "knowing everything" in the price of graphics cards, Intel's entry has naturally attracted considerable attention. Not to mention that according to some of the statements circulating in the network before, the unique performance of Rui Xuan is likely to have the mid-range level of other competitors from the beginning, so it is obviously very exciting.
However, for the purpose of prudence, we Sanyi Life did not write relevant analysis content at the first time, but patiently waited for more detailed technical information and running data leakage before starting the preparation of relevant manuscripts. Because of this, we hope that this content is not just for everyone to "see fresh", but can convey some more meaningful information.
Intel's first TSMC foundry GPU, energy efficiency does have advantages
In order to compare with everyone, we sanyi life first made the following table by collecting known information from public channels.
It can be seen that compared with the similarly positioned AMD Navi22 and NVIDIA GA104 cores, the "largest" DG2-512EU core in Intel's ARC product line has a significantly higher number of transistors, but the core area is not much larger. Obviously, the most important role here is the TSMC 6nm process used.
In fact, in the case that the AMD RX7000 series and NVIDIA's RX40 series have not yet been released, the high-end model of the Intel ARC family (that is, the DG2-512EU core) is indeed the most advanced product in the current high-end graphics card. And calculating its transistor density (53.4 MM/mm2) also reveals that it is significantly higher than both Navi22 and GA104.
And from the currently exposed main frequency and power consumption data, more advanced processes have indeed brought significant energy efficiency advantages to ARC. Under the premise of a larger core scale, a larger number of computing units, and higher specifications of the video memory controller, the mobile version of arc display with a low main frequency has essentially become the flagship mobile graphics card with the lowest nominal power consumption level at present, which has to be said to be quite powerful.
Graphics units are stacked, but gaming performance isn't necessarily particularly good
With the most advanced process in the industry at present, Intel naturally stacks their high-end ARC unique display with good stream processor specifications. Up to 4096 arithmetic logic units (ALUs), 256 texture units, 128 raster units, plus a "full" 256-bit memory bit width, and up to 512GB/s memory bandwidth.
At first glance, on some key graphics metrics (such as pixel fill rate, texture fill rate, and memory bandwidth), Intel's flagship unique display has actually surpassed AMD and NVIDIA's counterpoint products. So does this mean that Intel ARC has the ability to shake the position of A/N in the high-end graphics card market this time?
Unfortunately, not necessarily. In fact, if we carefully analyze the complete specifications and performance indicators of Intel ARC this time, we will find that its design ideas are more or less problematic in this era.
Now the complex model of the game is not necessarily built in advance, but depends on the graphics card to calculate it on time
First of all, we should know that for the 3D masterpieces after the DX11 era, many of the fine, gorgeous modeling, character hair and skin texture are no longer pre-designed faces and textures, but rely on the results of the graphics card to be calculated immediately.
This means that if you're playing an old game from ten years ago with extremely gorgeous graphics, Intel ARC's design idea may be very correct. Because of its extremely high texture and raster processing capabilities and very large memory bandwidth, it means that it can quickly render complex models that have been designed in advance.
But if you're playing a new game for 2021 and 2022, ARC's performance may not be as good, because its single-precision floating-point performance is actually equivalent to the level of competitors' mid-to-high-end products, and it has not yet touched the edge of the "flagship". However, the reality is that the era of piling high-precision models and 4K textures directly in the game installation package has long passed, and the real-time modeling based on "computational graphics", that is, the shadow of time, is the mainstream of today's large-scale 3D games.
Because of this, in the eyes of our three easy life, although some of the graphical indicators look "luxurious", Intel's ARC actually has a very strange feeling from the design. To put it simply, it's almost like "remaking" 2012 graphics cards with 2022 technology.
Lack of adaptation and cost constraints, ARC will take time to grow
Some friends may know that the current "boss" of Intel's graphics department is the famous Raja, who is also the chief designer of AMD Vega architecture. From this point of view, the reason why ARC is what it is not necessarily a design mistake, but more likely to be for some specific purpose.
So what is this "purposefulness"? On the one hand, although Intel has indeed "risen" again in the CPU market today, in the field of uniqueness, they can be regarded as "new players" no matter how much they say. This means that for Intel, they will inevitably have to face a considerable period of time, more than the A/N competitors lack of game optimization time.
However, it is well known that whether it is AMD's FSR, NVIDIA DLSS and other game screen optimization technologies, or computational graphics technologies such as ray tracing, mesh rendering, and surface subdivision, their effects are generally "optimized". That is to say, only if the game manufacturer takes the initiative to adapt, the computing power of the corresponding graphics card can be converted into the actual screen and frame rate advantage in the game.
So what if the game isn't optimized accordingly? In the case that "computing power" is useless, strengthening the "hard power" of graphic rendering, relying on higher texture and raster performance in hardware design, and improving the rendering ability at high resolution through greater video memory bandwidth, the game frame rate under "hard" 4K or even 8K resolution naturally becomes a relatively feasible stopgap measure.
In fact, if you look at AMD's graphics cards over the past few years, you will find that before the RDNA architecture, flagship products such as R9 Fury X and Vega64 were almost all similar design ideas.
NVIDIA's H100, the industry's strongest compute card, but its 3D performance is not even comparable to the 1650
Not only that, if you look at the graphics card product lines of the three major brands today, you will find that both AMD and NVIDIA have now realized the complete "separation" of professional computing cards and civilian game cards. Their professional card (e.g., CDNA, H100) architecture can completely abandon 3D rendering performance and focus on half-precision (for AI training) and double precision (for scientific computing, modeling) computing speeds.
However, Intel obviously does not have the courage to engage in "professional cards" and "game cards" two sets of architectures, so you will see that in the current original Intel game version of ARC, semi-precision, especially double precision performance is very high. Not only that, Intel even specially designed a model that allows arc solo display and Xe set display in the CPU to "cross fire" and work together. Interestingly, however, this mode is currently only available for video rendering (computation), not for game screen acceleration.
To put it bluntly, Intel, as a long-lost new player in the exclusive display market, had to put the game performance under the premise of lack of adaptation and the "versatility" that can be converted to professional cards at any time in a higher position when designing the first product. Because of this, we can say that the current ARC graphics card is obviously not a pure "game to GPU" from the design idea, so it is expected that it can come up to challenge the dominance of A/N in the field of high-end game graphics cards, in fact, it is still a bit unrealistic.
【The picture of this article comes from the network】