Shulou(Shulou.com)11/24 Report--

As we all know, the energy efficiency of NVIDIA GeForce RTX 40 series GPU has been greatly improved due to the upgrade of architecture and manufacturing process. However, due to the previous release of high-end cards, the general performance has also improved a lot, so the power consumption is relatively not low. What really makes the author intuitively feel the leap in energy efficiency brought about by the Ada architecture is our protagonist GeForce RTX 4070, which is not only smaller, but also has a rapid improvement in performance. It can meet more than 100 frames to play light chasing games at 2K resolution.

CTOnews.com also received the public version of NVIDIA GeForce RTX 40 series GPU in advance. In this test, we will use a set of high-end intel configuration to test to avoid graphics card bottlenecks as much as possible. The specific configuration is as follows:

The packaging of the exterior design NVIDIA GeForce RTX 4070 follows the previous design of the high-end graphics card, giving people a sense of "speed travel", but in the huge package, the graphics card becomes quite small.

Graphics card body is a standard 2-slot design, the length is only about 24cm, the overall design is compact. Even the ITX chassis with A4 structure can be easily installed.

The backplane design of the graphics card still continues the X-type design of the public graphics card, full of metal style, low-key and cool.

The front is a thick heat dissipation fin, the whole graphics card all-metal design is also conducive to heat dissipation.

The graphics card fan adopts a unique arrangement of one left, one right and one up and down, which can form a vertical air duct and better take away the heat on the fins.

The periphery of the graphics card is surrounded by a silver-gray all-metal edge with an eye-catching GeForce RTX logo printed on top.

The screw interface is designed on the right side, and the whole card has a strong sense of integration.

In terms of power supply, it uses the 16Pin power supply interface of the latest ATX3.0 specification, and comes with the adapter from 2*8Pin to 16Pin in the package. in fact, its TGP power consumption is only about 200W, which can be driven by a single 8Pin power supply in theory. When we install the machine, it is best to use a power supply that supports the ATX 3.0 specification, which can make the chassis more beautiful. If you have to use an adapter, you can use an one-minute and two-8Pin cable.

The I / O interface bezel uses dark gray color matching, and the interface is equipped with 3*DP1.4a interface and 1*HDMI2.1 interface, and the highest 8K60Hz output is supported.

In terms of interface, NVIDIA GeForce RTX 4070 is still the interface of PCIe 4.0. with mainstream motherboard, all performance can be released.

Core parsing RTX 4070 is similar to RTX 4070Ti, both are based on the AD104 core, but the details are different. The specific core code is AD104-250. The complete GPC has changed from 5 groups to 4 groups, with a total of 5888 CUDA cores, 184 Tensor cores, 46 third-generation RT cores and 64 ROP units, which can basically be understood as retaining 80% of the core size of RTX 4070 Ti. However, the signboard NVENC unit has changed from two to one. This means that its video coding capability is not as good as that of RTX 4070 Ti and above, but it is much stronger than the previous generation of RTX 3070 and RTX 3070 Ti.

The BOOST frequency of RTX 4070 is 2475MHz, and the default frequency is 1920MHz. Video memory is consistent with RTX 4070 Ti. All are 12GB GDDR6X with 192-bit bit width, and the bandwidth is also 21Gbps. From the above positioning, we can see that RTX 4070, like RTX 4070 Ti, is a medium-and high-end graphics card for positioning 2K high frame rate games.

Of course, GeForce RTX 4070 still uses the latest NVIDIA Ada architecture, based on TSMC 4N NVIDIA custom process, thus achieving a leap in performance and power consumption of up to 2x. Its streaming multiprocessor throughput is more than twice that of the previous generation, and the ray-tracing computing power of the third-generation RT Cores is 2.8 times higher. In addition, the fourth generation of Tensor Cores added a FP8 engine with up to 1.32 petaflops of Tensor processing performance, more than five times that of the previous generation. SER technology brings up to three times the performance improvement for ray tracing, and up to 25% improvement for overall game performance.

The new Ada architecture brings many new features in specific functions: the introduction of an optical flow accelerator enables DLSS 3 to predict the motion changes between frames in the scene, thus realizing the frame generation technology through AI, which improves the frame rate while maintaining the image quality. In addition, the new architecture also supports the application of AV1 encoders, which reduces the video volume and achieves better picture quality. At the same time, the application of RTX VSR function also makes real-time video super-resolution possible.

Theoretical performance We mentioned in the title that the power consumption of RTX 4070 is very low. How low can it be? Let's just bake doughnuts and test it. After 15 minutes, the core temperature is stable at 66.4 ℃, and the apparent storage temperature is about 80 ℃. The power consumption of the whole card is only 198W, which is even lower than the RTX 3070. If the power supply is not false, CPU with i5, R5 and other 100-watt CPU, as long as 450W power supply can drive the whole machine. Most crucially, it only needs a single 8Pin 1 / 2 output line to work (some non-public graphics cards even have a single 8Pin version), which means players with upgrade ideas don't need to change the power supply at all. Due to the small size of the graphics card, the chassis does not need to be changed.

Next, there is a 3DMark pressure test, which can detect whether the performance of the graphics card has declined under continuous running minutes. Generally, more than 97% can be regarded as a qualified graphics card. The score of measured NVIDIA GeForce RTX 4070 is 99.6%, and the performance release is extremely stable.

In the 3DMark TimeSpy DX12 test, the score of the video card reached 18140, which has exceeded RTX 3080 10G and is basically the same as RTX 3080 12G. It can be said that the DX12 performance of RTX 4070 is about the same as that of RTX 30 series high-end graphics cards, while the power consumption is only half.

In the 3DMark FireStrike Extreme DX11 test, the score of the NVIDIA GeForce RTX 4070 video card reached 21497, which was also similar to that of the RTX 3080 12G, while the power consumption was reduced by 50%.

In the 3DMark Portal Royal light pursuit test, NVIDIA GeForce RTX 4070 scored 11116, indicating that its light pursuit ability is about the same as RTX 3080 10G, which is also the level of being able to play light pursuit 2K games.

The actual measurement of the game, next, we will carry on the game measurement. It is worth mentioning that with the addition of counter-Terrorism Elite 2, nine of the ten mainstream competitive shooting games support NVIDIA Reflex. These games include "Apex Hero", "call of Duty: theater 2", "Destiny 2", "Escape from Takov", "Fortnite", "Watchman Vanguard", "return", "Rainbow 6: siege" and "Dauntless contract". NVIDIA Reflex can reduce operational delays. Improve the response speed and make the gun more sensitive. The author hopes that the CS2 version of Origin 2 will be updated soon and will also support NVIDIA Reflex technology.

Next, the author tested two classic 3A masterpieces. "Wild Dart 2: redemption" without turning on DLSS, GeForce RTX 4070 can reach an average of 103.6 frames, which is already enough to play at more than 2K 100 frames. This experience is undoubtedly excellent.

If we turn on the DLSS of the quality file, the number of frames will be further increased to an average of 119frame, and at this time there is almost no loss of image quality, and the experience is excellent.

We got a similar answer in another classic 3A masterpiece, Tomb Raider: shadow. Running its own BenchMark at the highest picture quality of 2K has reached an average of 174frames, satisfying the playfulness of the 2K165Hz screen.

When DLSS is turned on, the frame number is better, reaching an average of 206frames, and the frame number is increased by more than 30 frames. Generally speaking, the performance of RTX 4070 in traditional 3A masterpieces is very close to that of RTX 3080. Not to mention the RTX 3070 Ti and RTX 3070 of the previous generation.

These are just appetizers, and the most attractive thing about RTX 4070 is its DLSS 3 technology, which can further improve the frame rate. DLSS 3 is a performance multiplier driven by AI, powered by the new fourth generation Tensor Core and GeForce RTX 40 series GPU optical flow accelerator, is the latest version of NVIDIA's acclaimed deep learning supersampling technology, while providing groundbreaking optical multi-frame generation, will open a new era for NVIDIA RTX neural network rendering games and applications.

To put it simply, the past DLSS 2 technology improved image quality and frame rate by rendering a low-resolution image and then zooming in to a high resolution through AI. On the basis of compatibility with DLSS 2, the new DLSS 3 technology adds a frame generation function, which can insert a new frame calculated by AI between two real frames, so as to double the frame rate. At the same time, combined with the super-resolution function of DLSS 2, AI can reconstruct up to 7/8 display pixels, and the game performance can be up to four times better than without DLSS!

So far, more than 280 DLSS games and apps have been released. There are a total of 30 released DLSS 3 games. In terms of the progress of the release, DLSS 3 was adopted seven times faster than DLSS 2 in the first six months of the release of DLSS 2 and DLSS 3 respectively. It seems that the difficulty of adaptation is quite low. I believe that more and more games will adapt to DLSS technology in the future.

Let's first take a look at the frame number of the most stressful "cyber punk 2077" at + 2K resolution. By default, RTX 4070 still has no way to play, with an average of about 38 frames. If DLSS 2 quality is turned on, the average number of frames can reach 69 frames. If you open the exclusive cool techs DLSS 3 of RTX 40 series, it will soar instantly to an average of 103 frames, which can satisfy the look and feel and fluency at the same time.

Legendary 3A masterpiece "Wizard 3: hunting" has also recently updated the next generation version, the configuration requirements have been greatly improved, of course, the picture quality has also kept up with the trend, and it is not out of date today. At the same time, it also provides DLSS 3 technology support, which is undoubtedly good news for RTX 40 series graphics cards.

In the next-generation version of "Wizard 3: hunter", the average GeForce RTX 4070 is only 38 frames when DLSS is not enabled at 2K. If the quality file DLSS 2 is turned on, it can reach an average of 64 frames. If AI-enhanced DLSS 3 technology is enabled, the number of frames will soar to an average of 95 frames, and most scenes can be played at 2K frames.

In another next-generation 3A masterpiece, the Legend of the Plague: Requiem, the optimization is relatively much better. Even if you do not turn on any AI technology, you can play smoothly at 2K 60 frames, and you can play with the number of e-sports frames after DLSS 3 is added.

The measured RTX 4070 can already satisfy 71.2 frames at a resolution of 2K. If you turn on the DLSS 2 quality file, you can achieve an average of 104.5 frames, and if you turn on DLSS 3 inserting frames, the number of frames will soar to an average of 147.2 frames, meeting the needs of 2K144 high-brush displays.

So, how much GPU will be improved from generation to generation? The author replaced this configuration with RTX 3070 Ti to test a set of data to see how much performance will improve between generations and how much will be improved with the blessing of DLSS 3 technology. When DLSS 3 is not enabled, RTX 4070 has a frame number advantage of about 10%. If you turn on the DLSS 3 technology exclusive to the RTX 4070 series, the number of frames will lead by about 70%. This is the charm of AI technology.

As we mentioned earlier, the power consumption of RTX 4070 is quite low, which can actually be driven by a single 8Pin power supply. So we also use Nvidia's official FrameView tool to count the average power consumption in each 3A masterpiece. It can be seen that the actual power consumption of RTX 4070 in most 3A masterpieces is about 160W-190W, so low power consumption means that the choice of power supply is very abundant, and the budget can be saved a lot, which may become a new choice for ten thousand yuan machines. And ITX enthusiasts may be able to look forward to the RTX 4070 card with a single fan.

Creative production video memory scale, NVIDIA GeForce RTX 4070 and RTX 4070Ti specifications are completely the same, both have 12G large video memory, the speed has also reached 21Gbps. Therefore, it has also been given a certain creative production capacity. As a result, the graphics card also supports the installation of NVIDIA Studio drivers to accelerate more than 110 of the most popular creative applications. Proprietary SDK makes these applications run faster and provides exclusive features such as Optix, DLSS, and Maxine. NVIDIA Studio has a full range of creative applications, including NVIDIA Omniverse, Broadcast, Canvas and RTX Remix. At the same time, the blessing of big show also makes it have a certain computing power of generative artificial intelligence, so there is no problem to use it for simple AI painting training.

In modeling and rendering software, the 5888 CUDA cores of NVIDIA GeForce RTX 4070 provide hardware acceleration for increased productivity. Almost all modeling software optimizes NVIDIA's GPU, so it can take into account efficiency, stability and compatibility. For example, in the commonly used rendering tool V-Ray, you can take advantage of the ray tracing feature of RTX acceleration to achieve high-performance final frame rendering. In addition, GPU with AI noise reduction can further accelerate interactive rendering and provide a smoother work experience.

We also measured the performance of V-Ray Benchmark, and GeForce RTX 4070 got a score of 1847, which can handle most medium-scale modeling and rendering requirements.

Thanks to the improvement in light-chasing performance, the score of GeForce RTX 4070 in V-Ray GPU RTX has also increased to 2631, which can also meet the needs of some medium-load light-chasing modeling rendering.

In the later part of the video, GeForce RTX 4070 is also equipped with a NVENC encoder. And RTX 4070 also supports the next generation video coding technology AV1,AV1, which can provide faster video coding and higher quality streaming media transmission performance while occupying the same space. As the major video platforms want to save the cost of server traffic, AV1 coding will become the mainstream coding method in the future.

NVIDIA works closely with a number of industry partners to ensure that its GPU fully supports today's mainstream multimedia applications. At present, the domestic professional version of editing and projection has achieved accelerated support for NVIDIA AV1 video encoders. The commonly used video color editing software DaVinci also provides support for AV1 coding. This means that NVIDIA GPU will certainly be able to take full advantage of its AV1 acceleration when dealing with video editing and rendering tasks in the future, providing more efficient performance and a smoother work experience.

CTOnews.com also tested the PugetBench For Adobe family bucket to see if it could do the video editing job. Measured GeForce RTX 4070 in the Adobe Premiere commonly used by creative workers, we enable GPU Cuda acceleration, and then use PugetBenchmark to test. With a final score of 1475, there is no pressure to browse 4K videos on the timeline.

We also use PugetBenchmark for testing in Adobe Effects, another more stressful video special effects software. The final score is 1640, which can be used to produce some more complex visual effects.

In addition to content production, RTX 4070 also has new features in the area of content consumption. That's RTX VSR technology. The full name is RTX Video Super Resolution (RTX Video Super Resolution Technology). It can improve the definition of online 1080p video to 4K resolution at most through GPU's AI calculation. At present, this technology has been adapted to Chrome browser and Edge browser, and will also be adapted to local players led by VLC.

The use of RTX video super resolution is very simple, as long as the driver of the RTX 30 series / 40 series is updated to version 531.14 and the Chrome / Edge browser is updated to the latest version. Enable the path as follows: NVIDIA Control Panel-Video-adjust video image settings. There are four gears available under this option box. The higher the gear, the more obvious the super-resolution effect, but it will also consume more GPU resources. At present, it already supports domestic mainstream video platforms (bilibili, iqiyi, Douyu and Huya), as well as some foreign video platforms (Youtube, Twitch, Netflix, Hulu and Disney+).

In the following test, we play a video released by CTOnews.com on bilibili with a resolution adjusted to 480p. After exceeding the resolution of the leftmost native 480p, the next 1-4 files can be seen more clearly, and the actual look and feel is comparable to that of 4K.

↑ is 480p, VSR1, VSR2, VSR3, VSR4 from left to right.

In addition to Edge and Chrome browsers, well-known VLC native players also provide adaptation of RTX VSR technology in the Beta version. Just like a web page, just open it in the driver control panel. When we use a 480p local video to demonstrate, it obviously feels clearer on the right. Later, when you enjoy the local movie, you can also use this technology to further optimize the picture quality.

In the recently popular AI painting (generative AI), RTX 4070 also has a place for World War I. For example, using the v2-1_768-ema-pruned.ckpt model of stable-diffusion, we tried to draw 20 images of old yellow of AI. It took a total of 2 minutes and 18 seconds for RTX 4070.

On the other hand, the RTX 3070 Ti side takes 2 minutes and 57 seconds, which is quite obvious compared to the drawing speed gap of RTX 4070. If you just want to play with AI, RTX 4070 is good enough. Of course, if you want complex research, you'd better go to RTX 4090 or higher GPU.

To sum up, the author has no resistance to small and powerful things, and NVIDIA GeForce RTX 4070 is such a small and powerful small steel gun. It is about 50% lower in size and power consumption than the RTX 3080, but its performance is very similar. At the same time, a series of AI technology, such as DLSS 3 and RTX VSR, can meet the requirements of 2K 100 + frame. It is a card that is very suitable for RTX 3070 and RTX 2070Super players to upgrade.

If you are going to install a compact mainframe, then the public version of NVIDIA GeForce RTX 4070 is almost the best choice right now, and none of the other non-public RTX 4070 launches are that small. This graphics card will go on sale on the evening of April 12 with a suggested retail price of 4799 yuan. If you are interested, you can follow it.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.