I have a couple of free software that can make all the graphics card and graphics computing related devices in your hands 2 or even 4 times more performant, believe it or not?
2 weeks ago, if you asked me this question, I would rather believe that my boss is Qin Shi Huang, but now it's two weeks later, yes, there really is such software, she is free, open source, sexy, she doesn't care how complicated your past is, she is gentle and compatible with every graphics card in the world, and she doesn't mind if the soul of the cards belongs to AMD or Nvidia.
So the question is: how does such an awesome software do it, where do I download it, how do I use it, is there any risk? BlaBla...... Let's explain them one by one.
Everything has to start with game development, as we all know that most of the games that eat performance are 3D games, and 3D games are developed whether it is buildings, characters, plants, animals...... If time and cost permit, a real 3D model will be made in the game world as much as possible, so that players will feel more intuitive, three-dimensional and real in the process of playing.
So the question is, the vast majority of players play games through 2D flat panel displays, so how to present 3D models to 2D displays?
The answer is: quantum mechanics
Ah kidding, the answer is not so complicated, but simple and crude - take pictures
That's right, just find a "camera" to shoot the corresponding angle of the picture according to the angle of the player's operation, and then send it to the monitor after shooting. This so-called "camera" is actually your graphics card, and this whole process is called "rasterization" in computer graphics.
Specifically, when the graphics card "takes a picture", it will use different poses to "shoot" according to your settings for the game image quality, and the thing used to express the clarity of the picture is called "resolution".
In the case of display support, the higher the resolution of the picture output by the graphics card, the more delicate the picture will be on your visual experience, and the less obvious the jagged edges, and then further modify the picture effect through shadows, light effects, anti-aliasing and other functions, but the most expensive of the graphics card performance is still the resolution, so you can simply say that the resolution determines 80% of your somatosensory image quality.
But correspondingly, the higher the resolution, the more difficult the rasterization process becomes, and the higher the performance requirements for the graphics card.
As we all know, the old Huang family has a black technology called DLSS, which has now developed to the third generation. If you want to understand the working principles of the software mentioned in the opening article, you must first understand DLSS (Deep Learning Super Sampling).
The first generation of DLSS can be summed up in one word – and that is: "guess"
For example, if you enable DLSS 1.0 and then enable 4K resolution, the graphics card is actually lazy and only runs 1080p resolution, and then squeezes the AI computing core on the graphics card, and lets the AI forcibly "guess" what the complete 4K picture looks like based on the 1080 resolution picture and the database that was fed to him before
Guess doesn't that sound a bit unreliable, but it's actually ......
It's actually very unreliable.
The original DLSS was determined from the principle of operation, and if you want to have a better experience, you need to feed it with similar art styles, similar characters, and similar perspectives according to the specific type of game you are playing...... A lot of material with similar characteristics can guess a decent result.
Obviously, you can't let consumers do such a tedious work, right?So you can only rely on game developers to tune AI by themselves, but this is the most TM fucking place, for game developers, the commercial return attraction of tuning AI is really poor, and the picture optimization is barely enough, just don't make a big mess, in contrast, fix bugs, anti-cheat, make up DLC, and do money activities...... What's not more important than tuning the AI?
(*it seems that anti-cheat is less important @ all FPS games)
Therefore, the actual performance of the original DLSS in various games is very uneven, very mixed, and even some games are better than not open. Over time, NVIDIA can only think about the pain, they reimagined, is it possible to make an AI that does not need to be trained by game developers alone, and at the same time can make the AI more accurate when "guessing" the picture?
And so DLSS 2.0 came in
Compared with the previous generation, what remains unchanged is that the overall design idea of the second-generation DLSS is still open source and cost-saving for the performance of the graphics card. However, DLSS 2.0 no longer relies on a single low-resolution footage, but instead splits the target resolution into multiple sub-resolution pixels at runtime. For example, a single-frame 4K picture, the theoretical total rendering workload is 4 times 1080P, then DLSS 2.0 will split you into N 1080P pixels, and then the graphics card will still render the workload of 1080P during the actual operation, and finally decide which pixels to use with the assistance of AI, and then compose this picture for you according to the requirements of 4K.
Source: @硬件茶谈
What's even more awesome is that they really came up with a super AI that can optimize the picture for all types of games without separate training, and at the same time, when this AI is calculating, it will refer to the frame that has been calculated in the first few seconds, and complete the current picture that needs to be guessed what it should look like. Therefore, on the whole, the basis for AI "guessing" is more sufficient, the scope of "guessing" is smaller, and the final result is that AI "guessing" is more accurate.
That's why there are many games with DLSS 2.0 turned on, and the image quality looks even better than the original image quality directly output by TM... Because after turning on DLSS, what you see is not the picture designed for you by the developer's art designer, but the picture made for you by [art designer + AI], and the AI will fill in a large number of pixel-level details on the basis of the work completed by the designer, so that the total amount of information you see in the end is more than the amount of information directly rendered in 4K, so the somatosensory image quality will be better.
This is not over!Because the next DLSS 3.0 is even more beastly,This generation has additionally added"Frame generation"Technology,Take 2077 as an example,According to NVIDIA's official statement,Ideally, the image quality remains the same,The frame rate can be fully doubled by 4 times。。。 Then I manually searched for a bunch of extranet bigwig evaluations, and most of them could actually have an increase of 2~3 times, which was terrifying.
As for its working principle, it is very easy to understand,For example, every 10 frames generated by the graphics card,AII3 frames will be generated according to the existing picture,These 3 frames do not need the graphics card computing unit to work at all,So the graphics card has gained additional performance release space,But the corresponding price is that AI must wait for the graphics card to output the material before it can start calculating,So there will be additional physical delay,But according to the test of netizens,Most games can be controlled within 10ms,So it's not a big problem。 The real problem is that in order to sell cards, Lao Huang's B even has to be locked for his own users, and DLSS 3.0 technology can only be used with NVIDIA 40 series graphics cards.
In general, the AI graphics computing optimization technology represented by DLSS has indeed succeeded in achieving higher benefits (image quality) with less cost (performance), and with the update and iteration of DLSS, more and more developers have begun to pay attention to this technology, and there are currently more than 50 games that support DLSS 3.0, and more games that support DLSS 2.0, a total of more than 500......
But careful friends should have noticed that the above briefly described the operation principle of the entire DLSS technology, and its key part is to use the "AI computing core", which refers to the core at the hardware level, Nvidia officially calls it Tensor core, which is something that only RTX series graphics cards have.
So is there a new AI algorithm that can achieve DLSS-like effects without the need for specialized Tensor cores?
At this time, a beautiful woman who graduated from MIT with a Ph.D. at the age of 25 and made AMD's stock price rise more than 50 times in less than 10 years after taking office in Goose City shrugged her shoulders:
"AMD wants light"
As a result, AMD has its own AI graphics computing technology - FSR
As of the release of this article, FSR has also been iterated for 3 versions, in order to facilitate understanding, it is crude to say that the iterative path of FSR is similar to DLSS, the 1st generation and 2nd generation are all by rendering low-resolution pictures, and then let the AI brain make up the high-resolution pictures, and the 3rd generation is also a frame generation technology, and after testing by netizens, the effect is also quite good.
Source: Post-90s technology said
But the difference is that FSR does not lock a specific model of graphics card, not only can all A cards be used, but even N cards, Intel, and any brand of graphics cards in the future can be used, because - he is open source.
But open source is not invincible, even if it is open source, it is necessary to negotiate with the game developer one by one to persuade the other party to support FSR, so maybe Su Ma's funds for the FSR project are limited, or maybe the progress of business negotiations is too slow. FSR 3.0 is only supported in 4 games so far.
To sum up, the current situation of AI graphics computing technology is very painful, and although DLSS 3.0 has the best game compatibility, it is too closed because it only allows 40 series N cards. FSR 3.0 is open enough.,But there are few games that support it.,So there's no place to use the strength.。。
But!Don't worry~, because since this kind of problem can be thought of by ordinary players, the dads of the open source community will also think of it......
A developer named @Nukem9 made a plugin that can replace DLSS 3.0 with FSR 3.0, which roughly integrates the advantages of DLSS 3.0 compatibility and the advantages of FSR not picking devices.
That is to say, all RTX 20 and 30 series N card users can now experience the frame generation technology of FSR 3.0 in all games that support DLSS 3.0, and some extranet bigwigs have tested 11 games in one go, and the frame rate has increased by 50%~100% on average without a significant drop in image quality, which can be said to be very good!
The name of this plugin (mod) is called "dlssg-to-fsr3", no need to rush to find the download link and tutorial, rest assured that I have already helped you pack it, let's read the other treasure software first, and then you will pick it slowly after reading.
Let's go on to say that the above plug-in is only applicable to N-card users of the 20 & 30 series, so for A-card users and N-card users of 10 series and below specifications, are there any easy-to-use plug-ins that can also improve their own experience?
At this time, another big guy named @LukeFZ came up with a plug-in called "FSR2FSR3", his thinking was the same as the big daddy upstairs, but he also replaced DLSS 2.0 and DLSS 3.0 with FSR 2.0 and FSR 3.0, so that the DLSS support library games instantly all support FSR, and Jing A is ecstatic.
What's more, thanks to FSR's device-agnostic advantage, you can theoretically enjoy the thrill of skyrocketing frame rates with this patch as long as it's a DLSS-enabled game, no matter what kind of device it's running on, such as ROG handhelds and Lenovo Legion Go...... If you know how to decompile mods, or are willing to play the win system, even SteamDeck and switch can double the performance in an instant.
Speaking of which,This story is already cool.,The skyrocketing frame rate even makes me feel a little numb.,I used to toss for half a day overclocking + pick the hardware physique may have a frame rate that has risen by 5%、10%,This move your fingers to patch it will rise by 50% or even double,It's really cool!。
However, for players whose performance is already sufficient, and the patching soars to more than 200 frames per second, or even 300 frames, in fact, the benefits of the frame rate increase are no longer important, so think about it the other way, are there any plug-ins that can improve the image quality with the help of AI algorithms?
Then what you need may be the "DLSSTweaks" developed by the @emoose boss, simply put, this MOD can use an anti-aliasing optimization technology called DLAA in all games that support DLSS, and supports manual control of sampling coefficients, balancing frame rates and anti-aliasing effects, and has also been tested by a number of bigwigs on the Internet, and the anti-aliasing effect is very good, better than traditional FXAA, SMAA, and TAA.
Okay, after introducing these mods, whether you want to maintain the frame rate or want to increase the image quality, I believe it can meet the needs of the vast majority of people, but because the act of playing mods itself will involve modifications to the root directory and source files of the game, so if you want to play a game with PVP gameplay, you need to evaluate whether there is a risk of being misjudged and cheating, which is very important!
But!what if...... If you're a little more greedy, is there a frame rate optimization tool that is not easy to ban?
For example, the Lossless Scaling developed by the "THS" boss, in addition to the same use of FSR technology, its biggest advantage is that the author claims that the API (programming interface) used by the software is the same as the famous live broadcast software OBS, so as long as it is a game that will not lead to a ban on the number of people using OBS, you will not be banned by using this software. And because it uses FSR technology, this software doesn't need the game itself to support DLSS to be usable, so it is theoretically suitable for all games. And this big guy also said that he is studying the porting of FSR 3.0 technology, and it will be updated for free when the time comes.
Coincidentally, I just found out that this big guy opened a steam purchase channel.,I flipped through dozens of reviews one by one.,I didn't see one who said that he had been banned.,On the contrary, many netizens shared that the measured XX game will not be banned.,So the security is still okay.。
But the price to be paid with the corresponding is,According to the feedback of netizens,The overall image quality loss of this software is slightly higher than the several mods introduced above,The reason is not clear for the time being,So if the above mods can solve your needs,Try to use the above。
Finally, as mentioned at the beginning of the article, I have sorted out all the software introduced in this article, and you can refer to this picture to choose which software to use↓↓↓↓
At the end of the article, let's talk about my personal understanding, if it is said that whether it is A or N, the reason why they have disagreements on strategic issues such as technical routes, open source or non-open source, is that the essential reason is to consider their respective business interests. There is such a group of people, they seem to be pure "stupid". Those open source bigwigs, they could have used this to explode some gold coins, or entertain themselves and play for themselves after the development is finished, in short, you can't think from the perspective of commercial interests, why do they bother to send it to the community open source, and even continue to update it for free for netizens.
The only answer I can think of is that they really feel happy to help others. This is the brilliance of humanity that I see in the spiritual core of open source, and this is also the unique golden spirit of the Internet age. Especially in the context of global economic deflation in the past two years, but the rapid development of AI technology, I will always be involuntarily "cyber anxious", such as what if the human technology tree is crooked, what if one day there is really an evil AI to enslave all mankind, or be privatized and monopolized by a small number of people?
But it was these unassuming open-source mods that made me suddenly less anxious. Take the "dlssg-to-fsr3" MOD of the @Nukem9 boss as an example, it can be said that it is holding the sword of the sworn enemy AMD, overturning the DLSS 3.0 blockade carefully designed by NVIDIA, and to output NTR fiercely. But it's been almost 1 month since the release of the MOD version 0.1, and Nvidia doesn't dare to put a fart, and I guess it won't be long before DLSS 3.0 will be open to all N-card users, and it may even be partially open source, as for why I say that, I have to first figure out how important it is for NVIDIA to lock the old graphics card with DLSS 3.......
The famous "Moore's Law"
Let me first talk about a "violent theory" - the bottleneck of the development of the chip manufacturing industry can already be seen with the naked eye. This bottleneck is not a lack of investment, nor is it a political risk, but fundamental physics. Because human beings have been making extreme use of physical space in the matter of chip manufacturing. Specifically, the smallest computing unit in the chip, the "MOS tube", can no longer continue to be miniaturized, and continuing to force a small MOS tube will cause problems such as "leakage" and lead to inaccurate calculation results.
"Moore's Law grows 10 times every five years and 100 times every 10 years. And now the performance of chips can only increase by a few percentage points per year. Only 2 times every 10 years...... The expense and complexity of cramming more transistors into a small area makes it difficult to double the performance of a chip on a regular basis. So Moore's Law no longer applies. - Jensen Huang
In Lao Huang's own words, "Moore's Law" is about to fail. This is a very fatal problem for a graphics card manufacturer, because the core selling point of all chip products is computing power. When the growth of computing power is destined to become weaker and weaker, chipmakers can only toss some new moths as selling points.
Thinking about this, I came to my senses and understood why, starting from the 20 series graphics cards, Lao Huang bit the "ray tracing" and "DLSS" to death...... At that time, the concept of immature technology was so heavily promoted that DLSS 3.0 had to be locked only for 40-series graphics cards. Because there is no doubt that DLSS 3.0 is a very valuable opportunity for users to visualize the strength of the new generation of graphics cards.
I also understand why AMD and Intel, as latecomers in AI graphics computing technology, are willing to open source their own technologies (AMD is called FSR, Intel is called XeSS). Because from the perspective of business competition, open source can not only convene the world's heroes for my use, but also gradually dismember NVIDIA's sales strategy, and more importantly, from the perspective of market public relations, open source naturally stands on the moral high ground, so AMD and Intel can be said to have stepped on the door of NVIDIA's life.
Intel's XeSS is only partially open source
And it is extremely "coincidental" that just two days ago, AMD suddenly announced that it would integrate a frame generation technology called AFMF into the official version of the graphics card driver on the 24th of this month, and all A-card players above the 6000 series can directly unlock this function for free when upgrading the driver.
Wait a minute...... That's right, you guessed it, the core of this so-called AFMF technology is the open source FSR 3.0...... Combined with the time when several bigwigs introduced above sent MODs, I can only say that AMD is really ruthless, and it doesn't give Lao Huang a chance to breathe.
Speaking of which, you should be able to understand why I said that Lao Huang is very likely to have DLSS 3.0 technology in the future, or even partially open source. Because for NVIDIA, instead of forcing its old users to ride the wall and use AMD's FSR or Intel's XeSS, it is better to delegate it to old users as a part of the welfare, cultivate the user ecology of DLSS, and if you are lucky, you can get back some reputation.
Finally, I hope you don't get me wrong, I'm not accusing Nvidia of slamming the door, nor am I deliberately hacking him to make money. In fact, on the contrary, after combing through the whole article, you will find that NVIDIA, as a pioneer in the industry, has made great contributions in the field of AI graphics computing, and even without DLSS, there may not be FSR, XeSS and other AI graphics computing technologies at all. As a pioneer in this field, he has the advantage of leading technology, but he also has to bear the countermeasures of late-mover competitors, so no matter how much it is, it is nothing more than morality, and nothing more in business.
I take Nvidia as an example, simply because it is too representative, and Nvidia is very rare, and at the same time in a company that is very strong in AI hardware and AI software. To tell you a little trivia, the market capitalization of one Nvidia is equal to 7 Intel, 5 AMD, 4 Tencent, and 2 and a half TSMC...... Total market capitalization of $1.39+ trillion......
What is this concept? Even the popular fried chicken OPEN AI, which made Chat GPT, has a total market value of less than $90 billion, and even from the perspective of the global economy, as of the end of 2023, there are a total of 197 countries in the world, of which only 19 countries have an annual GDP of more than 1 trillion US dollars, accounting for less than 10%......
But it is such a rich multinational group, in the face of competitors open source digging into the wall, in the face of private users open source self-sufficiency to do MOD, he has no way at all, and the existence of these mods is completely legal, completely just, and with the support of netizens all over the world, it will only become stronger and stronger.
So I think that even if the day really comes when there will be an evil super AI, we will still have a righteous open source super AI that belongs to all mankind. This is not only because I choose to believe in human nature, but also because from the course of history, the outcome is likely to be the same.
Finally, thanks to the spirit of open source, let the progress of science and technology benefit all people, and also thank the existence of a community like the small black box, so that the information gap will no longer become a monopoly tool!