This article is authorized by Supernova Reprinted, other media reprinted by the Super Net agreed.
If we pay attention to the links between news, in fact, will find that in today's video card player community, when it comes to the use of multi-graphics, both NVIDIA SLi, or AMD CrossFire, in fact, are becoming more and more niche, Getting farther and farther away from the player's eyes and ears.An example of a concentrated performance is that in the white paper after NVIDIA introduced the Pascal architecture last year, the future will no longer be open to ordinary users for more than three-way SLI Game support is only available in the driver to focus on Two-Way SLI optimization, in order to tap higher performance support requires Enthusiast Key.
And even being more effective in the minds and using the more convenient AMD CrossFire is even more modest after the release of the Radeon RX Vega 64/56 graphics card, which seems a bit spontaneous, so we usually SLI, CrossFire, in the end is what technology? What is the reason for the gradual cooling? This is the lesson to be explained today.
Brothers are better than single-handed, and it's hard to see such propaganda now
What is SLI?
'Multi-GPU' literally means more than two graphics cards or two GPUs in parallel at the same time. Among the companies that have mastered this technology are the SLIs NVIDIA has put forward, the CrossFire two proposed by ATi / AMD If you want to follow the course of history, we must first talk about the SLI technology, its full name is 'ScalableLink Interface', most of our readers know this technology is because of NVIDIA, but the technology was originally Not developed or even represented by NVIDIA, perhaps the older users still remember 3dfx, the famous graphics card company that launched the Voodoo series graphics card in 1998. They took the lead in launching Scal-Line Interleave in 1998, On the Voodoo 2 graphics card, and as most readers know, NVIDIA was one of the last few survivors of the already fiercely competitive video card market and later acquired 3dfx, then purchased Scal -LineInterleave ', and officially released its magic changes in 2004, but it should be noted that this and the future we are familiar with SLI technology there is a very big difference, To say that only the greatest common denominator is that the first letter of the word.
This Voodoo 2 video card comes from Wikipedia
Before the advent of Voodoo 2 graphics card, there is no technology in the world that can connect parallel computing of two independent graphics cards, not to mention the goal of doubling the performance doubling of all the people. At that time, 3dfx passed the 2000 VSA100 graphics chip , And even double-core, or even four-core grand occasion, which for some graphics industry in the current Cambodian life eruption of the scene, after all, Radeon Pro Duo, Titan Z think it has been like a long time ago Products, not to mention Quad-GPU grand occasion.
Are both SLI technologies the same?
Need to pay particular attention to, 3dfx, NVIDIA introduced SLI technology is very different, the first is the connection mode, 3dfx introduced Scal-Line Interleave in two graphics card which need to be connected through a dedicated cable, and NVIDIA's Scalable Link Interface requires what we commonly refer to as a bridge, followed by the data transfer method, the 3dfx SLI technology is analog transmission, the digital signal is converted to analog signals and then the operation of the synthesis of the frame , And NVIDIA SLI technology is the use of digital signal transmission, the two graphics calculated by the composite screen was synthesized into analog signals.
In addition to the above characteristics, the most central issue in the Multi-GPU technology: screen processing task allocation, 3dfx, NVIDIA's idea is completely different. The former 3dfx idea is to use the screen by block by block Two parts, two graphics cards or GPU render different parts of the screen instantaneous frame, rendering after the synthesis, in other words the fusion of the previous frame and the latter frame, and NVIDIA's ideas are very different, the approach is Divide the single frame into several parts of the screen. In the case of Two-Way SLI, the screen is divided into two parts. If three, four or even the highest eight, then each one is responsible for rendering a few points.
More importantly, in the latter part of the development, NVIDIA to SLI technology which resource allocation problems to be improved, in other words, to achieve 3: 2, 4: 3, 5: 3 single screen split effect, so do The purpose is to ensure that the main and secondary cards in the SLI system need a consistent working time at work because the main card is not only to render the part they need to render but also to assign, command and control the tasks first, so if The main card to complete the rendering longer than the secondary card, then there is a secondary card waiting for the main card to allocate the next frame of 'free time', this problem is similar to we often pay attention to avoid the CPU performance is weak, unable to play GPU Performance issues.
The SLI technique proposed by 3dfx is to divide the picture into pieces on a block-by-block basis
This is the NVI SLI technology
What is CrossFire?
Speaking NVIDIA's technology come back to talk about AMD's CrossFire, if SLI technology seems unintelligible, then 'CrossFire' sounds very cool, mighty, and catalyst, adrenaline, volcanic island as domineering side Lu (as long as not understood as 'Cross Fire' is good.) CrossFire give you the impression perhaps better than the NVIDIA SLI, such as free bridges, exchange of fire more efficient and so on.
The name CrossFire was first released in the summer of 2005. The published technical details at that time were completely different from those of 3dfx. NVIDIA did the trick in rendering using odd-numbered frames and even-numbered frames. You are responsible for 1st, 3rd, 5th, I'm in charge of 2nd, 4th, 6th, the first graphics card to support this technology is Readon X800, X850, but still need to use the cable to connect the main graphics card and vice graphics card.
Here you may find it strange: how do I need to differentiate between the main card and secondary card? CrossFire technology is actually the case, the history is to develop the third generation of CrossFire technology, which is to ATi RV570 core baked before it began to end Instead of requiring cables, the history of primary and secondary cards needs to be subdivided into a medium of data movement that appears to be close to NVIDIA's bridge, albeit a mere image.
If you are not familiar with this generation of segmentation, then the fourth generation of CrossFire technology you must be very familiar with, and even overruled, because the fourth generation CrossFire technology from crater architecture R9 200 series graphics card, which is very close to us. In this generation of core architecture (GCN1.1), AMD began using XDMA bridgeless firefighting and support Frame Pacing (frame synchronization) technology.
In principle, CrossFire technology is generally considered more efficient than SLI technology, on the one hand the principle of resource allocation is similar, on the other hand because CrossFire technology in the multi-frame segmentation - combination of a variety of modes, for example, we compare Familiar Alternate Frame Rendering and SuperTiling modes are high-performance, require less resources, and have no obvious bottlenecks (though AFR itself does not need to be very high Threshold, so as a common multi-card rendering, in fact, SLI, CrossFire have applications).
Interactive frame rendering principle
Checkerboard isolated render
Even four or five years ago, discuss SLI, CrossFire technology users should be significantly more than today's multi-GPU enthusiasm for you, you may have seen the formation of Radeon HD 7950, R9 290 firefighting users have also seen try GTX 780 SLi players, but today you can see how many Radeon RX 570 CrossFire, GTX 1080 Sli users? One thing to pay special attention to is that Windows 10 has been officially deployed for more than two years, and in the formal push before Windows 10 Know this is very deep for the bottom of the system optimization, how deep?
And even can bring memory overlay and AMD, NVIDIA graphics hybrid crossfire super-dream, but the reality is that we all know that either CrossFire, or SLI, are their own GPU to use their own memory, and set up Multi GPU driver-driven support, Naturally it is impossible to realize the combination of the RX Vega 64 and the GTX 1080 out of thin air, but perhaps because there has been no clear progress so far, there are fewer users willing to try to build a Doka today.
As for why there are fewer and fewer users willing to try Multi-GPU right now, I think the main reasons are as follows:
# 1: single-core graphics performance has become increasingly powerful conquest of the game is no longer urgently need two or even three cards
This reason is actually similar to the former users obsessed with the CPU overclocking, because the performance itself for the demand there is a big gap, so you need to squeeze more performance overclocking, but in recent years the development of view, GTX 680, GTX 780 Ti, GTX 980 Ti, GTX 1080 Ti, Titan V performance has become more powerful, but the game itself, 1080p is still the mainstream resolution, the most reasonable anti-aliasing solution is still MSAA, the ambient light shading even Optimized to get better and better, so to buy a more high-end graphics card has been able to conquer the game, and even can spare no effort to challenge the 4K game, high-performance universal is bound to bring the pursuit of the ultimate demand is indeed, do not you see ' In fact, the underlying reason behind the fire's law is that the mid to high-end graphics cards are already very sweet. In fact, a GTX 1060 6GB graphics card is enough for many games.
# 2: The cost behind building a Multi GPU is still high, even higher
If you just want simple play single-core graphics, your processor can be more mainstream, Core i5-7600K, even Pentium G4560 are not bad, motherboards, chassis, power have a lot of choice, can be small, you can fine. But if you want to build a dual card, the first memory is not superimposed, then you need at least 4GB of video memory, then you need two GTX 1050 Ti, or RX 570, doubling the number, the budget is enough to buy the GTX 1070, Processors and motherboards need to be hardened because they require higher processing power to avoid bottlenecks and more PCI-E channels to ensure bandwidth and require more power to ensure output, larger chassis, and better wind Road to avoid the heat gathered, the noise will be greater, all this is the explicit and invisible costs, not to mention the current HB Bridge is not cheap.
# 3: Set up Multi GPU For the game optimization is a test of the field, it is likely less effective
Even if you set up a platform that has both performance and cost advantages, your dream may not be realized if your goal is to achieve higher gaming performance only, as the optimization of multi-card games and the completeness of single-card Are two concepts, if it is a single card, then as long as the graphics card architecture will not be too old, the quality is no problem, there is no system and driver failure, then the game is generally not a problem, but if the formation of dual-card games, then out The chance of error is not low, for example, black, no response, Huaping, Caton, etc., specific reference to the factory drive update log.
And if I have experienced the issue, it is impressive to use GTX 1080 Two-Way Sli, play "Wizard 3: Hunting," there will be a splash, Huaping problem, and finally by changing the full screen to the window In addition, I also tried RX 480 CrossFire, experiencing intensive non-responsive exit when playing "Sniper Elite 4", and finally how to solve it ... In fact ... ... did not solve it.
Even without errors, the efficiency of the formation of multi-card is the need for your attention, even if you update the driver, multi-card support files are timely, but the overall performance of the two cards for a single card is not uncommon 150%, and even many are below 150%, and computing is more efficient AMD CrossFire. There are a few examples of efficiencies exceeding 180%, and the cost behind this level of efficiency far exceeds 200%.
# 4: The pursuit of the ultimate, hands-on ability of the user, after all, only a handful
This sentence seems to be a commonplace among many issues, but the fact is that there are only a handful of users who have the will and ability to form a multi-card platform without considering other issues in real time. Most users, in fact, want only simple Single, easy to set up a platform to play the game, the lower the cost of natural maintenance, the better, and ultimately will not be a variety of problems, and bold attempts can not be separated from the natural problems, but also to summarize the law, Risking the risk of failure, this experience is not something everyone likes.
Moreover, the user is inertia, if you try to experience a single card did not encounter any problems, or even experience is good, there is no reason to try two or even three cards of the card, or encountered problems to ask who (certainly is a small super Gela)?
In short, if the old Multi GPU technology can be quite valuable after the use of valuable methods, it has now been very challenging for many reasons, its own meaning has become increasingly blurred, real-time Is a hardcore player, I believe also reluctant to set up multi-graphics, which is why there will be 'can not single double' reason for this.
