NVIDIA GeForce RTX 4080 SUPER first review: more powerful performance, more beautiful price

Finally, the last RTX 40 SUPER series graphics card is here, focusing on 4K high-brush gaming experience, and the starting price is 1,499 yuan lower than the RTX 4080, which is 8,099 yuan. If you happen to need to upgrade or install a computer recently, have a sufficient budget and want balanced performance in all aspects, you can get it done in one step. So today’s review of the NVIDIA GeForce RTX 4080 SUPER graphics card will definitely help you.

Before the evaluation, let’s introduce the test platform as usual. The specific configuration is as follows:

##Appearance Design

This time all models of the public version of the card have been changed to black paint, which can be said to be the biggest change in appearance. It is precisely because of this difference in details that the new version of NVIDIA GeForce RTX 4080 SUPER looks more handsome.

Unboxing the public version card itself is also an exciting thing. The packaging is very exquisite. When you put it on the table as shown above, it is said to be a metal The "ornaments" with a sense of technology are not an exaggeration.

The front of NVIDIA GeForce RTX 4080 SUPER still uses an Heyang carving processing, the actual look and feel is quite exquisite. However, the size has changed to a 3.5-slot card, so the volume is relatively large and the weight has increased a lot. Even when installing, it is recommended that you use a graphics card bracket.

#The heat dissipation fins and fan on the right side are all blackened, which makes it feel heavy in the hand and helps dissipate heat more efficiently.

Similarly, there is an identical fan on the left side of the back, which can form a vertical air duct with each other to better take away the heat inside the graphics card.

The side of the NVIDIA GeForce RTX 4080 SUPER graphics card has also been blackened, making it visually very unified. A slight concave arc has been made in the middle to enhance the texture. The power supply interface adopts 12 4pin design, making the power supply more stable.

The I/O interface bezel is dark gray, and the interface is equipped with 3*DP1.4a interface and 1*HDMI2.1 interface, supporting up to 8K60Hz output.

In terms of accessories, NVIDIA GeForce RTX 4080 SUPER is equipped with a 16pin to 3*8pin adapter cable, which has better compatibility with power supplies.

Core Analysis

Although the GeForce RTX 4080 SUPER is also based on the AD103 core, its core specifications are quite terrifying. For example, it is equipped with 10,240 CUDA cores. Compared with the RTX 4080’s 9,728 CUDA cores, the number of cores has increased by 5.2%. It also has 52 Shader Core TFLOPS units and 121 RT Core TFLOPS, and graphics/game performance has been upgraded again. In addition, its 320 Tensor Cores and 836 AI TOPS will provide powerful AI computing power and DLSS performance.

In terms of graphics memory specifications, the GeForce RTX 4080 SUPER is still 16GB GDDR6X, which is consistent with the RTX 4080. However, the 64MB L2 cache allows GeForce RTX 4080 SUPER to fully utilize gaming performance at 4K resolution. In terms of GPU frequency, the base frequency is 2295MHz and the acceleration frequency is 2550MHz, which are 40MHz and 80MHz higher than the RTX 4080 respectively. Judging from the graphics card specifications and official positioning, the GeForce RTX 4080 SUPER is a high-end graphics card positioned for 4K games and has powerful AI computing power.

Of course, GeForce RTX 4080 SUPER still uses the latest NVIDIA Ada architecture and is built based on TSMC 4N NVIDIA custom process, thus achieving a performance-power consumption ratio of up to 2 times. Streaming multiprocessors deliver 2x more throughput than previous generations. The DLSS 3 of the third-generation RT Cores can increase light tracing performance by 2 times. The fourth-generation Tensor Cores have a new FP8 engine with Tensor processing performance of up to 1.32 petaflops. DLSS can increase performance by up to 4 times.

Theoretical performance

Before testing the performance, we might as well pass the donut baking machine test to see how its power consumption performs. After baking for 20 minutes, the GPU power consumption remains at about 320W, the core temperature is about 69℃, and the memory temperature is 70℃. This temperature performance is relatively balanced. At this time, the GPU frequency can reach between 2325MHz-2450MHz, and the performance release is still Guaranteed.

After baking the machine, we conducted the 3D Mark Time Spy stress test on the GeForce RTX 4080 SUPER to see whether the graphics card performance is released smoothly under continuous high load. And durability, the GeForce RTX 4080 SUPER finally got a score of 99.6%, and the sustained performance output is very good.

Next is the GPU theoretical performance test. In the 3DMark TimeSpy Extreme DX12 test, the GeForce RTX 4080 SUPER's graphics card score was 14401 points, compared to the RTX 3080's 8602 points. An increase of about 67.4%, also surpassing the previous generation RTX 3090.

In the 3DMark Fire Strike Extreme DX11 test, the GeForce RTX 4080 SUPER's graphics card score was 34894 points, which was an increase of about 71.2% compared to the RTX 3080's 20382 points. The previous generation RTX 3090 Ti’s score of 27074 was also improved by 28.8%.

In the 3DMark Fire Strike Ultra DX11 test, the graphics card score of the GeForce RTX 4080 SUPER reached 17395 points, which was an increase of about 67.8% compared to the RTX 3080's 10365 points, and was also ahead. Compared to the 13855 points of the previous generation RTX 3090 Ti.

In the 3DMark Portal Royal light chasing test, the GeForce RTX 4080 SUPER scored 18,439 points, an increase of 64.1% compared to the RTX 3080’s 11,231 points. The performance of light chasing games has been improved. obvious.

Game Test

Now that we are talking about games, we have to mention that the NVIDIA platform currently has more than 500 games and applications that support RTX technology, and As DLSS 3 technology becomes more mature and popular, the number of games that support DLSS 3 is also continuously increasing.

For example, 14 new games supporting DLSS 3 technology have been added this year, including "Dragon's Dogma 2", "Horizon: West End", "Dragon Among Us 8" and "Dragon's Dogma 2". Well-known games such as Half-Life 2 RTX. In addition, the released "Alan Killer 2", "Cyberpunk 2077: Shadows of the Past", "Call of Duty: Modern Warfare 3", etc. also support DLSS 3 or DLSS 3.5 technology at the first time. At the same time, the current mainstream popular online competitive games all support NVIDIA Reflex technology. Whether it is for low-end players or e-sports players, these two technologies can greatly improve the gaming experience.

In the next game test session, we will test 6 games for you, including 2 online competitive games, 2 traditional raster games and 2 games that support DLSS 3 /3.5 ray-tracing games, all games are tested with 4K resolution to verify its 4K/4K ray-tracing game performance.

First of all, there are two online competitive games. In "CS2", we use the default high image quality option, conduct a game at 4K resolution, and pass Frame View records the average frame rate, and the average frame rate of GeForce RTX 4080 SUPER at 4K resolution can reach 305 frames.

In addition, by turning on the NVIDIA Reflex option in the game settings to enable enhancement, even at 4K resolution, the average delay of the game is only 9ms, which has a great impact on players. Small.

For the second online competitive game, we chose the popular "Rainbow Six: Siege", using GeForce RTX 4080 SUPER at 4K resolution and the highest image quality. The average frame rate has reached 350 frames, which is fully able to meet the needs of high-scoring and high-brush e-sports.

As for latency, "Rainbow Six: Siege" also supports NVIDIA Reflex technology. When the game is running, select the option with Vukan to find Reflex in the game. On enhanced option, after it is enabled to capture through Frame, the average delay at 4K resolution is only 10.2ms, and the delay performance is quite good.

Then there are 2 traditional raster games. In the classic narrative adventure game "Red Dead Redemption 2", choose the highest image quality and average frames at 4K resolution. The number can reach an average of 99 frames, and the RTX 4080 is 95 frames, an increase of 4.2%.

We also tested Ubisoft’s latest game “Assassin’s Creed: Visions” by running the game’s own performance benchmark test at the highest resolution of 4K. The average frame number of GeForce RTX 4080 SUPER has reached 100 frames, and the average frame number of RTX 4080 is 92 frames, an increase of about 8.6%. It is not difficult to see from the test results that in traditional raster games, the improvement of RTX 4080 SUPER compared to RTX 4080 is relatively limited.

So how does GeForce RTX 4080 SUPER perform in light chasing games? This site has selected two ray tracing games that support DLSS 3/3.5. One of them is the very well optimized "Forza Horizon 5". Select the extreme image quality preset in the game and turn on DLSS 3 DLSS 2 quality. file, the NVIDIA Reflex option will also be turned on by default. As usual, we select Reflex ON BOOST and run the performance benchmark test that comes with the game.

Enter the game and select 4K resolution extreme image quality, the average number of frames can reach 149 frames, and the average delay of the game is only 16.4ms. The average frame rate of the RTX 4080 is 132 frames, an increase of about 12.8%.

So if we only enable light tracing, do not enable the DLSS option, and turn off the Reflex low-latency option, what is the performance of the GeForce RTX 4080 SUPER? It can be seen that in the well-optimized Horizon 5, even without DLSS, the average frame rate under 4K light tracing still reaches 125 frames. However, without the blessing of Reflex low-latency technology, the average game delay soars to 24.1ms.

The last one is "Cyberpunk: 2077", which has very high requirements for graphics card performance. This time we tested it 3 times. For the first time, we selected super ray tracing quality and turned on DLSS 2 quality level DLSS 3 frame generation, running the built-in benchmark performance test at 4K resolution, the average frame rate can reach 83 frames, which is not bad performance.

The second time we chose the ultra-fast light tracing quality, turned on DLSS 2 quality file DLSS 3 frames to generate DLSS 3.5 light reconstruction, at this time at 4K resolution, GeForce RTX The average frame rate of 4080 SUPER is exactly 60 frames, which can be regarded as holding the bottom line of 4K / 60 frames.

Of course, if DLSS is completely turned off and only light tracing is turned on, the average number of frames in the game under 4K will be only 29 frames. After all, not all games can be optimized for Horizon 5 Level and image quality also have different loads on performance. Through comparison, it is not difficult to find that DLSS technology is definitely the most practical technology for gamers.

Turn off light reconstruction

Turn on light reconstruction

So what impact does not enabling DLSS 3.5 have on the game image quality? Look at these two pictures. The reflection of the slippery road after rain in the same scene in the game gives us the answer. After turning on the light reconstruction, the clarity of the reflection is significantly improved, and the artifacts that were common in the previous pictures are also much reduced. It is obvious DLSS 3.5 can be said to be the lubricant for high-level light tracing. It can retain the highest picture effects while also achieving the best frame rate performance.

creative production

We mentioned earlier that the L2 cache of GeForce RTX 4080 SUPER has also reached 64MB, the memory bandwidth is as high as 23Gbps, and there are as many as 836 AI TOPS cores. This is GeForce RTX 4080 SUPER provides a strong guarantee for creative production and AI performance. Users can also install NVIDIA Studio drivers to accelerate more than 110 of the most popular creative applications, such as Optix, DLSS and Maxine, as well as the full suite of NVIDIA Studio creative applications, including NVIDIA Omniverse, Broadcast, Canvas and RTX Remix.

In modeling and rendering software, GeForce RTX 4080 SUPER's 10240 CUDA cores can provide hardware acceleration, improving work efficiency. Almost all modeling software is optimized for NVIDIA GPUs, so it can take into account efficiency, stability, and compatibility. For example, in the commonly used rendering tool V-Ray, you can use the RTX accelerated ray tracing function to achieve high-performance final frame rendering. In addition, a GPU equipped with AI noise reduction can further accelerate interactive rendering and provide a smoother work experience.

A practical look at the professional productivity performance of GeForce RTX 4080 SUPER. The first is the benchmark test part of the V-Ray Benchmark modeling software. The GPU CUDA test item scored 3103 points.

In the GPU RTX test project, the final score of the GeForce RTX 4080 SUPER graphics card was 4292 points.

In the Benchmark test of another modeling software blender, the three-scenario scores of the GeForce RTX 4080 SUPER graphics card are as shown in the figure. The three-scene scores are 4574 and 2162 , 2264, also has a significant improvement.

Then let’s take a look at its performance in video rendering and encoding and decoding. GeForce RTX 4080 SUPER is equipped with 2 NVENC encoders and supports the new generation of AV1 video encoding. technology, it can provide faster video encoding and higher quality streaming media transmission performance under the same video size. Domestic Cutting Professional Edition has implemented acceleration support for the NVIDIA AV1 video encoder. We also tested this. We imported a 2 minute and 30 second 4K video material into the Cutting Professional Edition, and then used H .264 encoding and AV1 encoding for export.

You can see that when exporting the same 4K / 60 frame specifications, the size of the video exported by AV1 encoding is 609MB, and the size of the video exported by H.264 encoding is 800MB. The volume of AV1-encoded videos is nearly one-third smaller than H.264-encoded videos.

And in terms of picture quality, by intercepting the picture details of the two videos and observing them, there is basically no difference between the two encoding methods. They are small and high-definition, AV1 It is bound to become the new favorite among encoding formats.

There is also RTX video super-resolution technology, which can upscale online 1080P videos to 4K resolution through GPU AI calculations to achieve the effect of improving clarity through algorithms. We only need to find the video in the NVIDIA control panel - adjust the video image settings, and then we can customize the adjustment. There are 4 super levels.

↑ From left to right are 480P, VSR1, VSR2, VSR3, VSR4

We play on this site B I took a video posted by the website account, adjusted the resolution to 480P, and then successively increased the super score level to conduct a screenshot test. It can be seen that the characters become clearer at levels 1-4.

Finally, we briefly tested the performance of this card in AIGC through Stable Diffusion. The model we used is sd_xl_base_1.0_0 based on the SDXL open source model. 9vae.safetensors, draw 21 images of girls on snowy mountains in the form of Vincentian diagrams, the image resolution is 1024*1024, the sampling method is Euler a, the number of iteration steps is 50, the total number of batches is 10, and the final cost is 1 minute Completed in 55 seconds.

The second time we opened the TensorRT Engines optimization engine and ran it again with other settings unchanged. This time it only took 1 minute and 01 seconds, saving Nearly half.

In addition, check the video memory usage when running graphs in Stable Diffusion. Even under the relatively simple drawing parameter settings in the current evaluation, 8GB of video memory usage is normal, instantaneous The graphics memory usage will also soar to more than 10GB\12GB, not to mention the drawing conditions with higher resolution and more complex parameter settings. If the video memory of this card is only 8GB, then for professional AI creators, the video memory explosion is likely to happen frequently. Once the video memory is insufficient, the drawing time will be greatly extended, which is very delayed. Therefore, the 16GB large memory of GeForce RTX 4080 SUPER plays a very important role in AI. It is directly linked to the efficiency of AI content generation. Under the same architecture and similar performance, the larger the memory, the more advantageous it is.

Summary

Compared with the GeForce RTX 4070 SUPER we reviewed before, this GeForce RTX 4080 SUPER is obviously aimed at users who pursue the ultimate 4K gaming experience, that is, full picture quality for mainstream games The special effects can also reach 4K 100FPS game performance. Players do not need to choose between game image quality and special effects, and can just enjoy the game. Users who are still using old cards from the RTX 40 series and want a one-step gaming experience can buy it with their eyes closed. However, if you already have any RTX 40 graphics card, you can postpone the upgrade cycle. In addition, users with AIGC creation or other AI needs can also consider this card. Currently, it is also the most suitable for AI-related work.

On the other hand, the excellent workmanship and texture of the public version of GeForce RTX 4080 SUPER, as well as its unique appearance design, are also one of the reasons why many players are pursuing it. Not to mention that the initial price is a full 1,400 yuan cheaper than the original RTX 4080. Interestingly, if we compare the previous generation RTX 3080Ti, you will find that the number of CUDA cores, Tensors Cores, and RT Cores of both is the same, but because of the Ada Lovelace architecture, the performance in all aspects is much stronger, and the performance is much better. The consumption is even lower than 30W.

The above is the detailed content of NVIDIA GeForce RTX 4080 SUPER first review: more powerful performance, more beautiful price. For more information, please follow other related articles on the PHP Chinese website!