1 Introduction Specifications Comparison Appearance Analysis
Introduction
If there are any earth-shattering changes in the DIY circle in the second half of the year, it must be the RTX4090 graphics card. Its performance is really impressive, AdaL The ovelace architecture and TSMC
N process have brought a new breakthrough in the energy consumption ratio of this graphics card. Lao Huang’s exclusive black technology DLSS3 has excited many gamers. The luxurious stacking has significantly improved productivity and efficiency. However, due to the high price and production capacity, it is difficult to find a card.
However, flagships are hardware that only a few people can use. Ordinary players need a graphics card with more suitable performance, power consumption, temperature and price. The RTX4080 may be the most suitable choice at the moment. Recently we just received the GeForce RTX4080 16GB Ice Dragon Super Edition from Inzhong . Let us go through the evaluation to see if the RTX4080 can live up to its positioning.
* In the following, "Inzhong GeForceRTX4080 16GB Ice Dragon Super Edition" is referred to as "Inzhong RTX4080 Ice Dragon Super Edition"
Specification comparison
Before starting, let's first understand the protagonist RTX4080 this time. It uses the AD103-300 core and is manufactured with the TSMC4N process. The chip area is 379 square millimeters , and the transistor density has reached 45.9 billion. The number of transistors has increased significantly compared to the previous generation product, almost doubling. The core of this generation has further increased the frequency, so it can bring better performance.
In terms of other parameters, RTX4080RTX4080 comes standard with 9728 CUDA, 128 third-generation RTCores, 512 fourth-generation TensorCores, and uses 16GB of GDDR6X video memory. The large video memory and performance improvements can bring a better experience to games and creators.
The AD103-300 core currently equipped with the RTX4080 graphics card is not a complete AD103 core. The complete AD103 core should include 7 GPCs (Graphics Processing Clusters), 40 TPCs (Texture Processing Clusters), 80 SMs (Streaming Multiprocessors), and a 256Bit memory bandwidth with 8 32Bit memory controllers. Therefore, the author speculates that RTX4080 may not be the ultimate product under the AD103 core, and the RTX4080Ti with a complete AD103 core should be launched in the future.
Let’s look at the core structure diagram of the RTX4080 below. Comparing it with the full version of the AD103 core, it is easy to see the difference. The RTX4080 core code is AD103-300, which has 4 complete specifications. GPC (graphics processing cluster, each with 6 built-in TPCs), and 3 incomplete GPCs (two with 5 built-in TPCs, one with 4 built-in TPCs), forming a total of 38 TPCs, leaving 76 SM units, and the memory bit width is still a complete 256Bit.
However, the AD103-300 core on the RTX4080 has a cut in the codec. It must be said that Lao Huang’s cutting skills are precise, and the number of codecs is cut in half. It has the same specifications as the RTX4090, and only two NVENCs are retained. encoder and an NVDEC decoder, so the full version of the AD103 core should have faster video encoding and decoding speeds. However, the RTX4090 has been tested before and the encoding and decoding speed has almost doubled compared to the previous generation. Presumably the RTX4080 also performs well.
Appearance Analysis: Inzo GeForce RTX 4080 16GB Ice Dragon Super Edition
The front of the packaging box is printed with a rendering of Inzo RTX4080 Ice Dragon Super Edition. From the above we can see that this graphics card adopts a three-fan design and is dressed in black and gray armor. It exudes a cool and low-key atmosphere from the outside to the inside.
The back of the packaging box is more simple, printed with the newly upgraded technical introduction and brand concept of Inzo RTX4080 Ice Dragon Super Edition.
After opening the package, it still has the familiar taste. This generation of Inzo RTX4080 Ice Dragon Super Edition continues the previous design, with black and gray as the base tone, supplemented by silver-white decorative strips, with outstanding texture.
The three huge cooling fans on the front are particularly eye-catching. The fans are surrounded by cool armor, and the thick armored wind blows towards your face. If you look closely, you can also see that the fan in the middle rotates in the opposite direction to the fans on both sides, which can prevent turbulence and effectively improve heat dissipation capacity.
took a closer look at the cooling fan. Each fan has nine 99mm fan blades, and the fan supports intelligent start and stop. When the graphics card is under low load, it can maintain a low speed and reduce noise.
comes to the back of the graphics card. The Yingzhong RTX4080 Ice Dragon Super Edition is consistent with most high-end graphics cards. It uses a full-size metal backboard, which can reinforce and protect the PCB. After all, the graphics cards are really too heavy now. The unique triangle pattern is printed near the core with the words "iCHILL" and "GEFORCERTX".
The end of the metal back plate is a hollow window design, and you can vaguely see the heat dissipation fins of the internal . This design is conducive to improving the internal air convection , and the cooling fan in front can also bring away the internal heat more quickly. The top of the
graphics card is GEFORCERTX, and a whole row of air outlets spans the top of the entire graphics card, . Next to the RTX is the light of the ICHILL, which is looming when it is not lit. After
is lit on the machine, the light sign lights up, supporting the sync setting of the lighting effect. The perfect RGB effect is not too fancy, but it also looks cool and calm. A whole row of heat dissipation fins are also arranged at the bottom of the
graphics card to quickly discharge heat, reduce the core temperature, and ensure the stable output of the card. The
power supply interface uses the 12VHPWR interface under the ATX3.0 standard. The single port can provide 600W of power supply capacity, which is more than enough to use it to bring the RTX4080.
graphics card gold finger is a regular PCIe4.0x16 interface. It is worth noting that before using it, you must turn on the ResizableBAR function of the motherboard, so that you can run the bandwidth and fully exert the performance of the graphics card. The
interface is equipped with three DP1.4a and one HDMI2.1a, which supports up to 8K multi-screen output, and such an interface capability can fully meet the needs.
In addition, the size of this generation of RTX4080 is comparable to that of the big brother RTX4090. The Yingzhong RTX4080 Ice Dragon Super Edition is 3.5 slots thick. Pay attention to the chassis size when installing.
2 graphics card disassembly: Yingzhong GeForce RTX 4080 16GB Ice Dragon Super Edition
graphics card disassembly: Yingzhong GeForce RTX 4080 16GB Ice Dragon Super Edition
After watching the appearance of Yingzhong RTX4080 Ice Dragon Super Edition, let’s take a look at the inner part of this graphics card. Remove the screws on the metal back panel to separate the PCB and the huge radiator.
The interface, power supply, core and other positions on the front PCB are all ordered and arranged reasonably, but the PCB is obviously smaller than the RTX3090Ti and the density of electronic components is also greater, which is a considerable test for design capabilities. The back of
PCB is relatively simple, mostly with some control chips, and there are two POSCAPs (conductive polymer tantalum capacitors) on the back of the core. With them in charge, the overall electrical performance of the graphics card will be stronger.
AD103-300-A1 core is located on the front of the PCB, created by TSMC4N customization process, replaced with the brand new AdaLovelace architecture. With this powerful GPU core, Yingzhong RTX4080 Ice Dragon Super Edition can have flagship performance. Next to the core of
are 8 GDDR6X video memory, which comes from Micron , and the model is 2PU47D8BZF. As an upgraded version of GDDR6, the memory bit width is 256Bit, the speed reaches 22.4Gbps, and the single capacity reaches 2GB, and 8 pixels form a 16GB ultra-large video memory.
If you want to drive the core and video memory, you must also have a stable power supply scale. The Yingzhong RTX4080 Ice Dragon Super Edition uses 14+3 phase power supply, and also has high-quality components such as fully enclosed inductors and Fujitsu FP solid-state capacitors to inject strong power into the graphics card.
Each phase of power supply is equipped with an independent DrMos chip, the package model is BLN0, which should actually be AOS's AOZ5311NQI-03, and the continuous output current is 55A.
core power supply DrMos chip
video memory power supply DrMos chip
core power supply and video memory power supply are equipped with PWM power supply control chip, arranged on the back of the PCB, the models are: uP9512R, and uS5650Q respectively. Among them, the uP9512R manages core power supply, which can perform refined power supply management, while the uS5650Q is mainly responsible for the video memory power supply part.
uP9512R chip
uS5650Q chip
PCB is the power supply interface of the graphics card, which uses the latest 12VHPWR, with a maximum power of 600W on a single port. You only need a 12VHPWR power cord to complete the power supply of the graphics card.
Remove the remaining screws and you can see the cooling system of Inzo RTX4080 Ice Dragon Super Edition. Inzo calls it the eighth-generation Ice Dragon cooling system, which mainly consists of a radiator, cooling fan and metal backplate. The metal body of the
radiator is also quite large. It can not only dissipate heat for the GPU, but also for video memory, inductors and MOS tubes. The vapor chamber is equipped with a high-efficiency thermal pad to assist in heat dissipation. The heat dissipation effect is perfect.
Inzhong also adds a thermal pad on the inside of the metal backplate, which can provide additional heat dissipation.
Large-area cooling fins are arranged on the radiator, which is fully 6cm high, providing a heat dissipation area of 0.99m2, effectively improving the air flow efficiency, improving the temperature inside the graphics card, and ensuring that the graphics card can always operate stably. There are eight 6mm composite heat pipes sandwiched in the middle of the
radiator. They directly touch the vapor chamber above, which can quickly conduct the heat to the heat dissipation fins next to it and dissipate it out of the inside of the graphics card.
The eighth-generation Ice Dragon cooling system also has three cooling fans with a diameter of about 10cm for active cooling. It is not a problem to suppress the AD103 core with such specifications. Each fan of
has nine 99mm ultra-large fan blades and adopts a forward and reverse air duct design to enhance heat dissipation while preventing turbulence.
3 Test platform theory/game performance test
Test platform introduction
Let’s introduce our test platform. Since it is the flagship Inzo RTX4080 Ice Dragon Super Edition, it should be matched with it. The hardware cannot be defeated. We took out the strongest platform combination in the evaluation room. The CPU is Inteli9-13900K, the motherboard is ROGMAXIMUSZ790HERO, the memory is fully inserted, and it is directly connected to Kingston's 64GBDDR5-6000. Only with this configuration can the performance of this graphics card be demonstrated.
We all know that RTX4090 can even play 8K games, so smooth 4K with RTX4080 is not a problem. In order to better demonstrate the performance of Inzo RTX4080 Ice Dragon Super Edition in 4K games, we equipped it with a ceiling-level display The device - AGONPD32M co-designed by Aigong Porsche, 4K@144Hz high brush, and miniLED backlight blessing, exudes a sense of high-end from the inside out. I believe that with its support, we can see the ultimate gaming performance of Inzo RTX4080 Ice Dragon Super Edition.
International practice, you can see that Inzo RTX4080 Ice Dragon Super Edition has been correctly identified in the GPU-Z software. The base frequency of this graphics card is 2205MHz, and the Boost frequency can reach 2565MHz, which is 60MHz higher than the public version. This makes people full of expectations for its performance. In addition, you must turn on the ResizableBAR function of the motherboard before getting on the computer, so that the graphics card can maximize its full performance. In
GPU-Z, you can also learn about the temperature and power consumption strategy of Inzo RTX4080 Ice Dragon Super Edition. The TGP of the graphics card is set to 320W, which is consistent with the public version; the temperature wall is limited to 88°C, but with the heat dissipation capacity of Ice Dragon Super Edition, it is obviously impossible to hit the temperature wall limit.
Theoretical performance test
After using the machine, we first conducted the theoretical performance test of DMark. In order to more intuitively see the performance improvement of Inzo RTX4080 Ice Dragon Super Edition, we added the previous generation card king RTX3090Ti for comparison.
The first is the FireStrike series test represented by DX11. Inzo RTX4080 Ice Dragon Ultra Edition has exceeded the 6W mark. In the FireStrike Ultra test at 4K resolution, the performance is 40% ahead of the RTX3080Ti. In the TimeSpy series of tests represented by DX12, the performance of this graphics card also lived up to expectations. The previous generation of graphics cards could not touch the passing line of 1W points, while the score of Inzo RTX4080 Ice Dragon Super Edition has reached as high as 14186, and the lead has reached about 40%.
Ray tracing and DLSS are Lao Huang’s specialties. RTX40 series graphics cards have also made great progress in these areas. Inzo RTX4080 Ice Dragon Super Edition scored as high as 18137 in the PortRoyal test, which is already much better than The RTX4080 of other brands is even higher. If compared with the previous generation card king RTX3090Ti, the lead will be even greater. The RTX3090Ti is only about 1W4. The improvement in light tracking performance alone is more than 30%. After adding DLSS, the performance is even more explosive.
The above test is more about the theoretical performance of the game. This graphics card also performs well in terms of computing power. In the theoretical test of AIDA64GPGPU, the Inzo RTX4080 Ice Dragon Super Edition is more than 44% ahead of the RTX3080Ti. Compared with the card king RTX3090Ti, it also has a performance improvement of about 35%. It seems that Huang has squeezed out the toothpaste this time to make this graphics card have such an amazing improvement.
Game Performance Test
The theoretical performance improvement is outrageous. So how does Inzo RTX4080 Ice Dragon Super Edition perform in actual games? We selected a variety of games to test the performance of this graphics card at different resolutions.
1080P resolution, the Infinity RTX 4080 Ice Dragon Super Edition can be said to be the winner. After all, the previous generation RTX3080Ti can run smoothly at 100+FPS. Such high-end graphics cards obviously cannot show their true strength at 1080P resolution. At the resolution of
2K, Inzo RTX4080 Ice Dragon Super Edition has finally begun to exert its power. However, the tested games can basically run stably above 144FPS. Many game projects are nearly a hundred frames ahead of the previous generation. Even games like " War Machine 5", this graphics card can already meet the ultra-esports needs of 2K@240Hz. At
4K resolution, Inzo RTX4080 Ice Dragon Super Edition’s lead in gaming performance has gradually expanded, and its performance is already 40% higher than that of the previous generation. For example, the popular "Cyberpunk 2077" has extremely high hardware requirements. The previous generation RTX3080Ti could barely run at 59FPS, while the Inzo RTX4080 Ice Dragon Super Edition can already run smoothly at 90FPS with DLSS2 turned on. If DLSS3 is turned on, the frame rate will further increase, then it will be difficult for the RTX3080Ti to match it.
K resolution can be said to be the Shura field of graphics cards. We have tested a number of games and Inzo RTX4080 Ice Dragon Super Edition can still break through 60FPS in some games. , for well-optimized games such as "Horizon 5" and "Shadow of the Tomb Raider", this graphics card can reach 70FPS or even hundreds of frames. It seems that the game performance of Inzo RTX4080 Ice Dragon Super Edition is still good. Gamers should not miss it.
4DLSS 3 performance test
DLSS 3 performance test
As mentioned earlier, light tracing, as Lao Huang’s unique secret technique, has been launched on the RTX20 series graphics cards. However, the performance of the graphics card at that time was not enough to drive light tracing, so DLSS super-sampling technology was launched, which allows the graphics card to improve game performance without affecting the image quality, allowing players to experience more realistic game image quality. Now this technology has been updated to DLSS3.0, adding frame generation and NVIDIA Reflex technology, which can double the game performance.
DMarkDLSS3 Theoretical performance test
In the theoretical test of 3DMark, thanks to the blessing of DLSS3, the performance of Inzo RTX4080 Ice Dragon Super Edition has been further improved. At 4K resolution, the performance gap of switching DLSS3 can reach more than 3 times. The 8K resolution is even more outrageous. When DLSS is not turned on, it is only 1.5FPS. After turning on DLSS2, although the frame rate is improved, 45FPS is not enough to run smoothly. After enabling DLSS3, the frame rate directly reaches 70FPS, and the performance gap is close to 70 times. It seems that Lao Huang's DLSS3 really has something.
UnrealEngine5EnemiesDEMO Performance Test
Before the actual game test, we first tested the UnrealEngine5 engine needed to develop the game. As a brand-new game development engine, UE5 naturally puts a lot of pressure on the graphics card. Small, in EnemiesDEMO, the performance of Inzo RTX4080 Ice Dragon Super Edition soared after turning on DLSS3, reaching a level of 77AVG/661%FPS/55ms at 4K resolution, while turning off DLSS3 was only 22AVG/171%FPS/195ms, which is a full three times difference. I have to say that Lao Huang’s DLSS3 is really a black technology.
Some players may wonder whether the frame rate increase is achieved by reducing the image quality, so we also made a comparison. As can be seen from the video, even after DLSS3 is turned on, to the naked eye, there is almost no difference in image quality before and after the switch is turned on. And because DLSS uses AI for rendering, in some details, the image quality in DLSS mode is even clearer than the native image quality. Therefore, with DLSS3, while the smoothness of the game is improved, the image quality is not affected. It is a win-win situation.
"Reincarnation of Life and Death" game actual test
"Reincarnation of Life and Death" is one of the first games to support DLSS3. During the actual measurement process, we found that without turning on DLSS, the frame rate of Inzo RTX4080 Ice Dragon Super Edition is only 50FPS. Once we turn on DLSS2, the frame rate of the game can already reach 122FPS. By reducing some special effects, it can already meet the gaming requirements of 4K@144Hz.
When we turn on the latest DLSS3, the number of game frames increases at a speed visible to the naked eye, a full increase of 20FPS. At this time, the Inzo RTX4080 Ice Dragon Super Edition can already run an average frame of 143FPS at 4K resolution. If paired with the AGONPD32M Porsche co-branded display we used, this performance coupled with the high refresh rate of 144Hz, the gaming experience will take off directly.
K resolution DLSS performance test
We also tested a variety of games and programs. Judging from the actual test results, Inzo RTX4080 Ice Dragon Super Edition has no performance at 2K resolution. Pressure, many RTX3080Ti can only run 100FPS games, but on the Inzo RTX4080 Ice Dragon Super Edition, it can reach 200+FPS, far behind the previous generation graphics card, and with the addition of DLSS3, it can reach a terrifying 300FPS.
K resolution DLSS performance test
4K resolution can also be conquered by Inzo RTX4080 Ice Dragon Super Edition. After turning on DLSS3, the game The performance has been greatly improved, and the difference between the front and rear switches is the difference between playable and unplayable games in some games; and even at 4K resolution, many games can still run above 144FPS. It seems that 4K resolution has no pressure on the RTX4080 Ice Dragon Super Edition. We still have to try it at 8K.
K resolution DLSS performance test
8K resolution The RTX4080 graphics card is not enough. According to actual measurements, only a small number of games or programs can run smoothly at around 60FPS. The author believes that this is due to insufficient video memory. Subsequent optimization by game manufacturers is required to allow this RTX4080 to perfectly demonstrate its due strength at 8K resolution.
5 creative productivity dual NVENC encoder performance test
creative performance test
creative productivity is also the focus of many users, so this time we chose PugetBench and PCMark10, two common test software, to test the performance of Inzo RTX4080 Ice Dragon Super Edition in daily office work, video content production and other aspects.
is first tested by PCMark10. This graphics card has made great improvements in digital content creation and gaming projects. The overall performance is more than 8% ahead of the previous generation, and is even 50-50 better than its own big brother RTX4090. The above performance improvement is also confirmed in the creative productivity software test. In the Benchmark test of the DaVinci editing software alone, the performance is about 16% ahead of the RTX3080Ti.
It is also worth mentioning that the Inzo RTX4080 Ice Dragon Super Edition is equipped with 16GB of large video memory. In actual testing, no project crashes were encountered. Precisely because of the blessing of large video memory, this graphics card can run more and more complex projects, while the previous generation only had 12GB of video memory, so the video memory often exploded.
Let’s take a look at the performance of Inzo RTX4080 Ice Dragon Super Edition in modeling, rendering, industrial design, etc. In the Blender rendering software, 9728 CUDA cores are still strong, leading RTX3080Ti and RTX3090Ti by 52% and 45% respectively, which shows the strength of this generation of graphics cards.
This is also true in the 8 industrial software tests integrated by SPECviewperf2020. The performance improvement has also reached about 32%. That is, for the same model rendering, replacing it with Inzo RTX4080 Ice Dragon Super Edition can allow you to shorten the rendering time by nearly one-third, and the efficiency is significantly improved.
Dual NVENC encoder special test
Lao Huang’s black technology is not only DLSS3, he also arranged dual NVENC encoders on the RTX40 series graphics cards to support the popular AV1 encoding, and AV1 is the next generation Mainstream video encoding technology has its own unique advantages. It has faster encoding speed and higher quality streaming media transmission performance. Commonly used editing software such as DaVinci, Wondershare, and Cutout already support AV1 encoding. Mainstream video websites such as Bilibili have also added AV1 decoding. AV1 will become a new trend in the future. Since
has dual NVENC encoders, we have to test the encoding capabilities of this graphics card.We used the 8K chip source and engineering files provided by NVIDIA to test the encoding time in AV1 format and H.265 format respectively. The actual coding capability of Yingzhong RTX4080 Ice Dragon Super Edition can be on par with the big brother RTX4090. When using AV1 encoding, the speed is significantly better than the H,265 format. Even if it is replaced with H.265 encoding, with the support of dual encoder, the efficiency of Yingzhong RTX4080 Ice Dragon Super Edition is more than 62% faster than the previous generation.
From the above comparison test, we can see that AV1 codec is significantly ahead of H,265 format in terms of export speed, but it also has advantages in occupying space. From the figure below, we can see that whether it is 4K or 8K resolution, after using AV1 encoding, the video file size can be reduced by more than 25% on average.
AV1 has a considerable advantage in encoding and exporting and occupying space, and its image quality is not inferior to H.265 format. Here we have taken a few pictures for comparison. From the naked eye, the image quality is actually almost the same, which also means that AV1 can achieve image quality performance with a smaller space occupancy and H.265. It has to be said that video workers can really change graphics cards. The addition of AV1 codec gives Yingzhong RTX4080 Ice Dragon Super Edition super export performance and ability to fight the future.
6 Power Consumption Overclocking Evaluation Summary
Power Consumption and Heating
TSMC4N Process and Architecture Innovation, the performance of RTX40 series graphics cards has been greatly improved. This time, Yingzhong even used a super-spec cooling system for its own RTX4080 graphics card, which makes the author look forward to the temperature and power consumption performance of the Yingzhong RTX4080 Ice Dragon Super Edition.
In the 15-minute test of Furmark donut single baking, the occupancy rate of Yingzhong RTX4080 Ice Dragon Super Edition reached 99%, but the GPU core temperature at this time was only 65℃, and the GDDR6X video memory, as a major heating provider, was as low as 48℃, which is enough to prove that this generation of graphics cards has really put a lot of effort in architecture design and radiator.
In addition to the pleasant temperature performance, the Yingzhong RTX4080 Ice Dragon Super Edition also performs well in noise control. The temperature is lower than the previous generation RTX3090Ti, and the fan speed has always remained at low levels, and the fan noise is almost impossible to feel. Overall, the radiator of this graphics card has excellent performance, and it is no problem to deal with appropriate overclocking operations, and it is also far ahead of the previous generation of products in terms of energy consumption ratio. This is what high-end graphics cards should do.
Overclocking Experience
Since it is a secondary flagship graphics card, in addition to standard performance testing, overclocking testing is naturally indispensable. We use Yingzhong's own INNO3DTuneIT overclocking software, which can adjust the core frequency, video memory frequency, unlock power consumption and temperature wall operations on this software.
Click the "Smart Scan" button on the INNO3DTuneIT interface. The software starts to perform intelligent scanning of the graphics card for a long time. After the scan is completed, it can also be intelligently overclocked, which is quite convenient to use. Even beginners can easily get started and gain the performance improvement brought by overclocking.
However, this time, the author used manual adjustment to increase the GPU core by 160MHz, and the video memory was overclocked by 1250MHz. The final overclocking result was in the TimeSpy test project in 3DMark. The GPU reached 2970MHz, and the video memory also reached 1556.5MHz, with a score of 29692. Compared with the silent frequency state, the performance improvement was nearly 4%, and the performance was not bad.
Of course, this is not the limit of this graphics card. If you have better cooling conditions, the overclocking ability of this graphics card can be further improved. Interested players may wish to try it out and explore the limit of this graphics card.
evaluation summary
After testing, it can be said that Lao Huang is really serious about making RTX40 series graphics cards this time. The AdaLovelace architecture of the replacement has brought a huge improvement in the ray tracing of this generation of graphics cards. The dual encoder and support of AV1 allow productivity to feel the efficiency improvement brought by technological progress. With the support of DLSS3 black technology, the gaming performance cannot be underestimated. In the past, 8K was even dared not think about it, but now it is close. In general, this generation of RTX40 series graphics cards are full of surprises everywhere.
The Yingzhong GeForceRTX408016GB Ice Dragon Super Edition, which was given this test, is a model. It has exquisite appearance, radical material piles, excellent texture, and excellent heat dissipation performance. It also reserves a certain amount of overclocking space for players. With its own overclocking software, Xiaobai can also easily get started.The performance of the
graphics card is also remarkable. It can play all kinds of game masterpieces without any problems, and many games can even meet the needs of 4K@144Hz e-sports. The 16GB large video memory also puts it at the forefront of productivity graphics cards. It has the performance of productivity graphics cards and is very cost-effective.
Inzo GeForce RTX 4080 16GB Ice Dragon Super Edition is currently on the market, priced at 10,699 yuan. Obviously this is a graphics card for game enthusiasts and productivity creators. However, with its strong performance and outstanding performance, it still deserves the title of high-end graphics card. Interested players may wish to pay attention.
7 technical explanation: Ada Lovelace architecture
Ada Lovelace architecture explanation
Turing and Ampere's two core architectures are named after characters. The former is the father of computer science - Alan Mathison Turing ; the latter is the "Newton of electricity" - Andre Marie Ampere , the international unit of electric current ampere is named after his surname. Then Ada Lovelace must be extraordinary, and after a quick test, it turned out that this is Ada Lovelace , known as the "Queen of Numbers", wrote the first computer program in history, and was recognized as the first computer programmer in the world. She is indeed more awesome than the previous generation. PS: Her father is the author of "Don Juan" and the poet Byron .
Starting with the Turing architecture, NVIDIA added the RTCore unit to accelerate ray tracing and the TensorCore unit for AI reasoning to the graphics card for the first time. This revolutionary innovation makes real-time ray tracing possible. The Ampere architecture is a comprehensive architectural improvement. In addition to the new generation of second-generation RTCore and third-generation TensorCore, there is also a more advanced SM unit design, which doubles the efficiency of the graphics card. When it comes to the AdaLovelace architecture, efficiency improvement is the main premise. Naturally, the latest third-generation RTCores and fourth-generation TensorCores units are introduced, and many novel black technologies are added. In terms of execution efficiency, the AdaLovelace architecture is more than twice that of the previous generation Ampere architecture, and even the ray tracing capability has reached a terrifying 4x performance.
The new SM streaming processor
One of the biggest highlights in the AdaLovelace architecture: the new SM streaming multiprocessor. Each SM contains 128 CUDA cores, 1 third-generation RTCores, and 4 fourth-generation T ensorCores (tensor core), 4 TextureUnits (texture units), 256KBRegisterFile (register file), and 128KBL1 data cache/shared memory subsystem, so this new SM unit has twice the performance of the previous generation.
In the past, the number of Turing architecture INT32 computing units and FP32 was the same, and the two were added together to form 64 CUDA cores. But starting with the Ampere architecture, the computing unit on the left has implemented concurrent execution of FP32+INT32 computing units, which means that the number of CUDA cores has doubled to 128.
Let’s take a look at the SM of the AdaLovelace architecture. The FP32/INT32 computing unit combination also realizes the design of each SM containing 128 CUDA. It seems that the improvement is not big, but when you understand that GeForceRT The X4080 has 76 SMs and 9728 CUDA cores, so you should understand how the shader capability of 82.6TFLOPS is achieved, which is more than twice the 40TFLOPS of the previous generation RTX3090Ti graphics card.
In addition, the AdaLovelace architecture has also undergone major improvements in terms of cache. First, each SM unit is equipped with a 128KB cache, so that the RTX4080 graphics card achieves 97MBL1/shared memory. Secondly, the core's L2 cache has been redesigned, and the complete AD103 core and RTX4080 both have 64MB L2 cache, which can be said to be a qualitative leap compared to the RTX3080Ti.
Technical explanation: The third generation RT Cores and the fourth generation Tensor Cores
thought that the number of CUDA and the large L2 cache were already very strong. The biggest improvement in the AdaLovelace architecture is still in the third generation RTCores and the fourth generation TensorCores.
The third-generation RTCores
RTCores are used for ray tracing acceleration. The effective ray tracing computing power of the third-generation RTCores reaches 191TFLOPS, which is 2.8 times that of the previous generation.
In the Ampere architecture, the second-generation RTCores supports BoxIntersection testing and TriangleIntersection testing, which are used to accelerate BVH traversal and perform ray triangle intersection test calculations. Although ray tracing processing capabilities have been more efficient than the original Turing architecture core, as the geometric complexity of the environment and objects continues to increase, traditional processing methods are difficult to more efficiently and correctly reflect light in the real world, especially the accuracy of light movement.
Therefore, two important hardware units have been added to the third generation RTCores: OpacityMicromapEngine and DisplacedMicro-MeshesEngine engines. OpacityMicromapEngine is mainly used for alpha channel acceleration, which can increase the ray tracing speed of alpha test geometry by 2 times.
In traditional raster rendering, developers use some Alpha channel materials to achieve more efficient picture rendering, such as Alpha channel leaves or complex-shaped objects such as flames. But in the era of ray tracing, this traditional approach will lead to a lot of ineffective calculations for ray tracing. For example, if a moving light passes through a leaf multiple times, and each time the light hits the leaf, a shader will be called to determine how to handle the intersection, which will cause serious execution costs and time waiting costs.
while OpacityMicromapEngine is used to directly parse opacity state
triangles with non-opacity ray intersections. The alpha channel is processed according to three different block states: opaque, transparent and unknown: if the transparent block is directly ignored and continues to find the next one, the opaque block is recorded and notified of the hit, and the unknown block is left to the shader to determine how to process it. In this way, most of the GPU does not need to debug the shader and can achieve more efficient performance.
DisplacedMicro-MeshesEngine
If OpacityMicromapEngine accelerates surface processing, then DisplacedMicro-MeshesEngine is an accelerator for geometric surface details. As shown in the figure above, in the AdaLovelace architecture, a highly detailed geometric grid can be created through 1 base triangle + displacement map, which requires lower resource usage and is more efficient than the second-generation RTCores.
uses the example of creating a 14:1 coral crab given by NVIDIA. Here we need 17,000 micro-grids and 1.6 million micro-triangles. In the AdaLovelace architecture, the BVH creation speed can be accelerated by 7.6 times, and the storage space can be reduced by 8.1 times. DisplacedMicro-MeshesEngine plays a key role. It divides a geometric object into micro-network processing with different densities according to different details. The density of red is extremely high, and the more complex the detail processing is. The corresponding low-density micro-network area can release more resources and storage space, so that DisplacedMicro-MeshesEngine can help BVH accelerate the process and reduce build time and storage costs.
At the same time, ShaderExecutionReordering (SER) has been added to the AdaLovelace architecture SM. This is because ray tracing no longer only deals with strong light or shadow rendering. In the future, it will focus more on the movement of light, so the light will become more and more complex. If you want the third-generation RTCores and the fourth-generation TensorCores to have higher execution efficiency, you must arrange a housekeeper for them. Shader Execution Reordering (SER) is designed to instantly rearrange shader loads to improve execution efficiency, provide 2x acceleration for ray tracing, and better utilize GPU resources. However, there are currently no examples. To realize this function, we need the support of games and development tools.
Fourth generation TensorCores
TensorCores are specialized execution units designed to perform tensor/matrix operations, which are the core computing functions used in deep learning. The fourth generation of TensorCores adds a new FP8 engine, which has a tensor processing performance of up to 1.32 petaflops, 5 times more than the previous generation.
8 technical explanation: DLSS 3
technical explanation: DLSS 3
or the fourth generation TensorCores is so hard-core that you don’t know what it is? What’s the point of improvement? But you will definitely know the most classic application of TensorCores, DLSS. This time the AdaLovelace architecture supports NVIDIA's latest DLSS3 technology.
https://images.nvidia.cn/cn/youtube-replicates/r-hu006p23I.mp4
We have also talked about DLSS technology before. Its design was originally designed to make up for the performance loss after ray tracing technology. The specific manifestation was that the game frame count dropped significantly after turning on ray tracing technology, and it was even difficult to ensure smooth operation of the game. Therefore, DLSS uses low-resolution content as input and uses AI technology to output high-resolution frames, thereby improving the performance of ray tracing.
contains three technologies in DLSS3: DLSS frame generation, DLSS super resolution (also known as DLSS2), and NVIDIAReflex. You can understand that DLSS3 has added DLSS frame generation technology based on DLSS2; among the last two technologies, DLSS super resolution only requires GeForceRTX graphics cards to be used, while NVIDIAReflex is used for graphics cards after GeForce900 series.
It is not easy to generate DLSS frames, which requires the GeForceRTX40 series graphics card with the AdaLovelace architecture. The principle of DLSS frame generation technology is: use AI technology to generate more frames to improve performance. DLSS will use the new optical stream accelerator equipped with the GeForceRTX40 series GPU to analyze continuous frames and motion data, thereby creating other high-quality frames without affecting image quality and response speed.
Starting from the Ampere architecture, NVIDIA graphics cards have already supported optical flow accelerators, and the optical flow accelerators of the AdaLovelace architecture have been upgraded to the second generation, which provides up to 300 TeraOPS (TOPS), which is more than 2 times faster than the first generation optical flow accelerators of the Ampere architecture (OpticalFlowAcceleration, OFA). In order to realize DLSS frame generation, OFA plays an important role, and it combines with the new running vector analysis algorithm to achieve accurate and high-performance frame generation capabilities within the framework of DLSS3 technology.
In addition, since DLSS frame generation is performed as a postprocessing on the GPU, we can also obtain better game performance improvements even when the game is limited by CPU performance. Especially for physically compute-intensive games or large-scale scene games, DLSS2 allows GeForceRTX40 series graphics cards to render games with up to twice the performance computable by the CPU.
Finally, since DLSS3 is based on DLSS2, game developers can quickly integrate this function in existing games that already support DLSS2 or NVIDIAStreamline, DLSS3 has been widely used in the game ecosystem, and more than 35 games and applications are about to support this technology.
Read highlights: NVIDIAReflex
NVIDIAReflex is also a link in DLSS3. It can synchronize the GPU and CPU, ensuring optimal response speed and low system latency.
To achieve end-to-end minimum latency, you need to ensure that the game, monitor and mouse all support and enable Reflex technology at the same time.
When GeForceRTX40 series graphics cards and NVIDIAReflex are paired, they can directly achieve a 1440p resolution of 360FPS experience, which is really a bit powerful.
was revealed at the GTC2022 conference that there will be four new G-SYNC e-sports monitors with 1440p resolution to be released, including AOCAG274QGM-AGONPROMiniLED, MSIMEG271QMiniLED and ViewSonicXG272G-2KMiniLED, all with 300Hz refresh rates, while the strongest one is ASUSROGSwift360HzPG27AQN, with the refresh rate directly reaching 360Hz.
But the only problem is that some display manufacturers believe that such products have a small audience, which will reduce the production capacity of such monitors, and even the products have been PASSed in the internal, so 1440p360Hz is very beautiful, but the reality is also quite skinny.
9 technical explanation: Dual NVIDIA encoder
technical explanation: Dual NVIDIA encoder (NVENC)
GeForceRTX40 series graphics cards also have a brand new upgrade, that is dual encoder NVENC. The eighth generation NVENC dual encoder not only supports H.264 and H.265, but also supports the open video encoding format AV1.
. Since AV1 is a royalty-free video encoding format, upstream software manufacturers and the supporting ends of the game are vigorously promoting this encoding format. We will also see more and more hardware and software supporting the AV1 format, including the Chuangying Professional Edition, DaVinciResolve, and the more popular Voukoder plug-in of Adobe PremierePro, all of which can be used for dual encoder through encoding presets, so that the time we wait for video export is reduced by nearly half.
is not only a video production software, the AV1 format will also be the new favorite of anchors and game live broadcasters. While ensuring the highest quality of the picture, the AV1 encoder can increase the efficiency by 40%, and the graphics card usage is also lower. The first-generation software including OBSStudio will also add support for the AV1 format. In addition, we can also record content up to 8K60 through GeForce Experience and OBS Studio, which will make it easier for us to record games.
, including the in-game recorded videos we used for subsequent tests, all support the AV1 format, and the dual encoder NVENC is doing better and better in terms of resource usage and adaptation.
