The mine tide has receded, new games are frequent, and AIGC has entered a new era. 2023 is also the year of the full layout of next-generation graphics cards, have you purchased a new graphics card this year?The end of the year has arrived, let's also take stock of this year's graphics card market.
The "new" graphics card is finally coming to the gamer's console
Putting an end to the chaotic period of the graphics card market, gamers finally got their hands on a new graphics card in 2023. On the Steam hardware list, the RTX 3060 has also become the most used graphics card, but the old GTX1650 and even GTX1060 are still very strong. For players, this is also quite frustrating.
It is worth mentioning that the share of the new generation of RTX 40 series graphics cards has risen rapidly, and the flagship-level RTX 4090 can even rank in the top 30. However, AMD is relatively bleak, if you don't count the core graphics, the highest usage rate of AMD graphics cards is still the RX 580 many years ago. This is also quite consistent with the performance of the market share of the two graphics cards.
Although from the perspective of graphics card shipments in the global market, this year's graphics card shipments are lower than last year's, which is related to the general environment such as destocking, the disappearance of demand for mining cards, the GPU ban and the decline in game business revenue, but in general, the graphics cards sold to players should be more.
The layout of next-generation graphics card products is perfect, and thousand-yuan products are still missing.
NVIDIA and AMD, the two major GPU** vendors, both released their flagship next-generation products at the end of last year, announcing the beginning of a new round of upgrades. And this year's task is to complete the product line separately.
The cancellation of the RTX 4080 also became the first time in the history of another graphics card.
On January 4, the beginning of the year, NVIDIA was the first to lift the ban on the RTX 4070 Ti in the RTX 40 series. This is perhaps one of NVIDIA's most controversial products in recent years. It first appeared in the official release sequence and was called the RTX 4080 12GB, which is now the actual RTX 4080 based on graphics card capacity. But in fact, not only does it have less memory capacity, the GPU core specifications are far from RTX 4080 (16GB), and the 192bit video memory bit width makes gamers complain that this is not only unprecedented in the "80" sequence, but even in the "70" sequence.
With the cancellation of the name, it seems that the FE version of the RTX 4070 Ti (pictured is Rog-Strix-RTX4070Ti-O12G-Gaming) has also disappeared
Due to too many complaints, NVIDIA finally announced that it had withdrawn the RTX 4080 12GB and changed it back to the RTX 4070 Ti. The name has been changed but the specifications can't be changed, the RTX 4070 Ti still only has a bandwidth of 504GB s even with 21GHz ultra-high-frequency GDDR6X video memory, which is far inferior to the previous generation RTX 3070 Ti, which leads to its very good performance in 2K games, and the performance even surpasses the RTX 3090 Ti, but its performance in 4K is more average. Of course, from the market point of view, Lao Huang's knife cut just right, effectively controlling the cost and power consumption, and Huang's knife method is once again accurately reflected in the product. However, if players remember that it is a 2K graphics card, then there is no problem at all.
The RTX 4070 FE Edition is starting to get very small.
On April 12, the NVIDIA RTX 4070 was unbanned, using the same AD104 cores as the RTX 4070 Ti, and the number of CUDA cores is exactly the same as the previous generation RTX 3070, both of which are 5888. It also has a memory bit width of 192 bits, compared to 256 bits for the RTX 3070. Previously, everyone speculated that its performance may only be at the level of RTX 3070 Ti, but its actual performance has caught up with or even surpassed the RTX 3080, and it can still maintain the highest image quality of more than 100FPS when it turns on 2K resolution in mainstream games, which is qualified as a 2K game graphics card.
The compact dual-slot, dual-fan short card has become the mainstream design of the RTX 4070 (ASUS Dual-RTX4070-O12G).
Surprising is its power consumption, although the TGP set by the RTX 4070 is 200W, but this upper limit is generally not touched in actual games, and the RTX 3080 is going to go above 300W. From the perspective of performance and power consumption, it is indeed a good product, but the price of 4799 yuan for the first time still discourages many players. However, with the passage of time, the RTX 4070 has now become the most used RTX 40 series graphics card on the Steam graphics card list.
The RTX 4060 Ti Fe has the same styling as the RTX 4070 FE, with only minor color differences.
On May 23rd, the NVIDIA RTX 4060 Ti was unbanned, which is the first "dessert" graphics card priced at 3000 yuan in the ADA Lovelace architecture. The RTX 4060 Ti uses the new AD106 small core, which has 4352 CUDA units. However, it has cut the memory bandwidth to 128bit, and although it gives 32MB of L2 cache to make up for it, this 128bit of video memory still makes gamers uncomfortable, after all, this is the first time that such a small bandwidth has been given in the "60" sequence. Although the L2 cache has been added, it is inherently destined to be basically out of line with 4K games. What's even more frustrating for gamers is that the RTX 4060 Ti is not too much of an improvement compared to the RTX 3060 Ti, only about 10%, and in a high-resolution environment, the RTX 3060 Ti G6X even has the power to fight. Fortunately, the power consumption of the RTX 4060 Ti is very good, and the gaming performance can be unleashed at about 130W, which is a leap forward compared to the 200W RTX 3060 Ti. But it's also worth mentioning that the RTX 4060 Ti is the only 60 Ti graphics card in Nvidia's history that hasn't outperformed the previous generation 70.
The relatively good power-per-consumption performance makes the RTX 4060 Ti a perfect fit for ITX hosts. (Pictured is the iGame RTX 4060 Ti Mini OC).
It is also worth mentioning that the RTX 4060 Ti also has a 16GB video memory version, ** which is a few hundred dollars more expensive than the 8GB version, and was quietly launched in July, except for the increase in video memory capacity. Although everyone expressed dissatisfaction and incomprehension of this 16GB version at the beginning, with the gradual popularity of AIGC, it has naturally become a good choice for low-cost AI alchemy and painting. After all, for AIGC, it is necessary to have a large video memory to support "whether it can run", and then it will consider "whether it can run fast or not".
The public version of the RX7600 is also very delicate and compact.
AMD graphics cards were completely silent in the first half of the year, allowing NVIDIA to sing a one-man show all the time. However, AMD still rushed to the day after the RTX 4060 Ti was unbanned, and finally brought out the new RX 7000 series graphics card, but surprisingly AMD did not serve the dishes in order from high-end to low-end, and directly released the dessert-grade RX 7600. This is obviously to catch up with the opponent to come out with 2,000 yuan products to grab the market. RX 7600 in the case of more than 30% performance improvement compared to RX 6600, but the first ** is cheaper350 yuan, whether it is 1080p or 2K games can basically run smoothly, so it is still more conscientious. Later, everyone also confirmed that even the later released RTX 4060, although more expensive, still did not exceed the RX 7600 in raster performance, which can be said to be more conscientious. However, AMD's dilemma still exists, in terms of AI, FSR3, which is benchmarked against DLSSS3, has been slow to be implemented, and the performance of ray tracing games is still slightly poor. And when the RX 7600 was released, the RX 6650XT with similar performance had a more attractive **, and it was not so much that the RX 7600 was there to grab the market, but that it made its own product line more chaotic.
Moore Threads also released the "Zhiyu BMF" machine equipped with MTT S70 S80 graphics card.
On May 31, domestic GPU manufacturer Moore Threads held a press conference and released their new MTT S70 graphics card, which uses 3584 MUSA cores and a GPU frequency of 16GHz, equipped with 7GB GDDR6 video memory, video memory bandwidth 392GB s, FP32 computing power 112 tflops。The release of MTT S70 has improved the product system of Moore Threads, and everything is developing in a good direction. The DX11 driver has been upgraded again, which also gives the graphics card of Moore Threads more practical value.
For older GTX "60" players, the RTX 4060 is a good upgrade option. (Pictured is the GIGABYTE RTX 4060 Aero OC 8G).
On June 28th, the NVIDIA RTX 4060 was lifted, and the sale started at 2399 yuan, which was 100 yuan lower than the initial price of RTX 3060, and there was a price reduction on the day of listing. Although NVIDIA still positions the RTX 4060 as a 1080p gaming graphics card, it can actually support most games to run above 60fps at 2K resolution. Compared with RTX 3060, its performance is nearly 20% stronger, and with the blessing of AI technology such as DLSS3, the power consumption is also reduced by 1 3, and it can easily run full as long as 115W. It's fair to say that it's the perfect replacement for the RTX 3060 and earlier 60-series graphics cards in every way. It's a pity that in the early days of the listing, RTX 3060 has fallen into less than 2000 due to inventory clearance, and there is also AMD RX 7600 resistance, for players who pay more attention to performance at this price, there is still some resistance to buying RTX 4060 immediately.
The RX 7900GRE is positioned slightly lower than the RX 7900 XT
7month28day,AMD released the exclusive RX 7900GRE for the retail DIY market in China (only the whole machine is pre-installed abroad),It bears the name of "RX 7900",The overall specifications are similar to RX 6900 XT,The performance is not far from RX 7900 XT,The price has been pulled to 5299 yuan。 Compared to the expected RX 4070, it is 10%-20% stronger. However, although the price of 5299 yuan is cheaper than RTX 4070 Ti by 1200 yuan, it is still 800 yuan more expensive than RTX 4070 at that time, although it has 16GB of large video memory, but it is still difficult to change the choice of players.
The RX 7800 XT and RX 7700 XT are a bit late to market.
It has been delayed until September 6th, AMD's 4099 yuan RX 7800 XT and 3699 yuan RX 7700 XT are belated, the former opponent is RTX 4070, and the latter is RTX 4060 Ti. The pricing of these two graphics cards is probably still quite tangled for AMD, mainly because the RX 7900XTX and RX 7900 XT released last year have gone through multiple rounds of price reductions. Both graphics cards are N**i 32 cores and feature an MCM design similar to the N**i 31. Performance-wise, they have achieved their goals, surpassing the RTX 4070 and RTX 4060 Ti in raster performance, and thanks to the large video memory, they can also run some 4K games well. But on the whole, at the time of its release, AMD's promised FSR3 had not been fulfilled, and at this time AIGC had also become popular, and the lack of AI functions made it difficult for players to start when buying, after all, everyone wanted to buy a product with more features.
The RX 6750 GRE was launched at the same time as the FSR3.
To the surprise of gamers, AMD lifted the ban on the RX 6750 GRE on October 17, and there are two sub-models of 10GB and 12GB. Just like its model, they are not part of the latest RX 7000 series, but are built on the RDNA2 architecture. This also proves from the side that RDNA2 is indeed a relatively excellent architecture, while RDNA3 is more like a facelift. The RX 6750 GRE fills the huge product gap between the RX 7600 and RX 7700 XT. In addition to the different memory capacity, the two sub-models also have slight differences in specifications, with the 12GB version having 40 CUs, while the 10GB version has 36 CUs. In terms of performance, the RX 6750 GRE 12GB is very close to the RTX 4060 Ti, but** a bit cheaper than the RTX 4060. The RX 6750 GRE 10GB is benchmarked against the RTX 3060 12GB, but in reality its performance is basically the same or slightly better than the RTX 4060, and their capacity is also very generous at this level. For the 2000 yuan game market, there is still a lot of lethality. What's more, along with the RX 6750 GE, there are also FSR 3 super-resolution technology and AFMF frame generation technology, and AMD's pie has finally landed.
Intel ARC A580 Index 8G
It's a pity that it's the end of the year, and players are likely not to see this generation of thousand-yuan desktop graphics cards this year, and RTX 4050 and RX 7500 are now unknown. If you want to say it, the last member of the Intel ARC A series graphics card family, ARC A580 debuted in October, priced at 1599 yuan, which can be regarded as a unique seedling of 1,000 yuan. It has 24 XE cores and 256-bit 8GB video memory, and its theoretical performance is on par with the RTX 3060, but the only problem is the old problem of driving.
dlss3.5 With FSR3, a new direction for AI-driven graphics cards
In fact, from the perspective of the overall architecture, whether it is NVIDIA's RTX 40 series or AMD's RX7000 series, it is not a new thing compared to the previous generation of products, and it can be regarded as a facelift with the blessing of more advanced process technology. In order to face the increasing performance demand, NVIDIA pioneered the launch of DLSS3 technology based on AI frame generation at the launch of the RTX 40 series, bringing real-time frame generation to game image generation, providing a new direction for future high-frame-rate games.
Nearly a year after the release of DLSS3, NVIDIA officially released the new DLSS3 on September 215。Brand new DLSS35. Ray reconstruction technology is introduced, which can effectively improve the problem of artificially designed denoiser processing the details of the picture. dlss3.5Training with 5 times the amount of data as DLSS 3, it can recognize different ray-traced effects, make more informed decisions, and process game footage more accurately, making it a perfect alternative to human-designed denoisers.
It is a new technology that has been named "drag". Because DLSS3 is bound to the optical flow accelerator of the RTX 40 series, neither the RTX 30 nor RTX 20 series graphics cards can enjoy it. But dlss35. The new Ray Reconstruction technology supports the full range of RTX graphics cards. So, full DLSS35 Composed of Super Resolution, Frame Generation, and Ray Reconstruction, the RTX 40 Series is capable of supporting all DLSS35 features, while the RTX 30 RTX 20 series supports SR and RR, in fact, there is a bit of confusion in the naming, but since it was previously named 30, with the blessing of new technology, it can't be returned and renamed 2x bar.
After NVIDIA launched DLSS3, AMD immediately expressed its ambition to launch the benchmark FSR3. However, the reality is cruel, and it wasn't until almost a year after the pie was painted that FSR3 landed with the release of the RX 6750GRE. FSR 3 uses the same computational frame generation technology as DLSS 3, which also requires game support. However, to the comfort of players, Su Ma is still so generous, the opponent's DLSS 3 only supports RTX 40 series GPUs, while FSR 3 can not only support RX 6000 7000 series, but also support opponent (Intel NVIDIA) graphics cards.
Along with FSR3, there is also AMD's AFMF (AMD Fluid Motion Frames) technology, which is intended for games that do not support FSR 3 DLSS 3.
Despite being late, FSR3's image quality and frame rate performance were still appreciated by players. From now on, frame generation based on AI computing will provide new possibilities for high-resolution and high-frame-rate games.
The hot AIGC, continuous driver upgrades and bans
Since last year, AIGC has entered the public eye, including ChatGPT, Midjourney, Stable Diffusion, SVC and other AIGC-related applications have begun to reach ordinary users on the C side, and the popularity of AIGC-related content in the community has continued to rise. The localized deployment of stable diffusion and various language models has set off a new wave of learning Xi, which has triggered a large round of graphics card upgrades.
Using Stable Diffusion is one of the most popular ways to play it right now.
Due to Lao Huang's continuous layout for many years, the CUDA ecosystem has an extraordinary leading edge in AIGC, and many artificial intelligence Xi frameworks are based on the CUDA ecosystem, and the AI market can be said to be one-sided, and the ecological barriers have been very solid. Although AMD and Intel have been catching up and coming up with ROCM and OpenVino to fight against it, it is obviously still in its infancy, and it will take a long, long way to go to perfect. It can be said that if you want to develop and learn Xi AIGC at present, the N-card is almost indispensable.
On another track, Intel is still working hard to update the driver for its own ARC A series of independent graphics cards. After more than a year of continuous driver updates, Intel's graphics cards have been much improved in terms of usability. But as long as the driver problem is there, the Intel graphics card will always be a niche choice. Coupled with the turmoil within Intel's graphics card department, judging from the current situation, the three-point world situation that players are expecting may come later.
Moore Threads is the first domestic GPU manufacturer to support DX11 API.
Like Intel, who works hard to engage in drivers, there are also domestic GPU manufacturers Moore Threads. Domestic GPUs have a long way to go, and pioneers like Moore Threads will obviously encounter many problems, and the drive is the biggest problem. With the efforts of Moore Threads, the first adaptation of DX11 API in China has finally been realized. It's quite gratifying that more and more DX11 games can run smoothly on Moore Threads GPUs, and at the same time AIGC support has been added.
The sprout of a new birth is always easy to stifle, and when everything is going in a good direction, the United States has put Moore Threads on the "entity list", which can be described as a slightly heavy blow. However, we also believe that at this point in time when challenges and opportunities coexist, this is not necessarily the "darkest moment" of domestic GPUs, it may also be a new starting point for the sea of stars.
At the same time, the RTX 4090 ban is slowly fermenting. After arguing for a long time, the RTX 4090 has officially taken effect due to the high computing power in China, and it cannot continue to be produced and shipped in China. Nvidia's Chinese official website has removed the RTX 4090 graphics card information. At present, RTX 4090 on major e-commerce platforms are off the shelves and out of stock, and it is difficult to find a card. Although there is currently news that NVIDIA plans to launch the RTX 4090D with reduced computing power to meet the domestic market demand, it is always an uncomfortable thing, and the reason for it is embarrassing, and I can only say "hehe".
Regardless of performance, the second innovation of graphics cards
ASUS back btf2The 0 solution is a very bold and innovative design.
The colorful graphics card adopts a special hidden line back plug design.
Players always say that the DIY market is silent and a little less passionate. And this year's graphics card products may bring a little surprise to gamers, and manufacturers have incorporated a little care into the design. With the advancement of the rear plug series products, the graphics card has also followed further, and this design of reducing the front cable has also been recognized by some players. Among them, Colorful adopts a back-plug hidden cable design, while ASUS is more radical and directly designs the back-mounted BTF20 solution, a new hidden interface is designed for the graphics card to replace the original external power supply interface, and the new interface is directly plugged into BTF20 The power supply interface of the graphics card of the motherboard realizes a "completely wireless" design.
It is a very peculiar idea to take advantage of the reduced PCIe lanes of the graphics card.
Another flower job is also from ASUS. As we all know, the PCIe interface of the RTX 4060 Ti has been reduced to PCIe 40 8, which means the regular PCIe4 on our motherboardThe 0 16 interface has half of the bandwidth left. ASUS came up with an ingenious solution by designing an M. on the RTX 4060 Ti's graphics card cooler2 expansion slots, on motherboards that support PCIe channel splitting, you can install the SSD on the graphics card and use it normally. Since this channel is directly connected to the CPU, it may be faster in read and write performance than PCH-based slots, and it can also support PCIe 50。I still want to praise the innovation.
Summary: The new year is also a new starting point for the market
In general, the graphics card market in 2023 is mainly to innovate and improve the layout. After years of mining turmoil, players are finally able to buy graphics cards normally, upgrade configurations, and play new games. In this year, the popularity of AIGC has given more new uses to the graphics cards that were originally used to play games. For the market, graphics cards are no longer just for gamers, and the popularity of AI has brought more new explosions to the market. So how will the next 2024 develop, let's wait and see.