After the GeForce RTX 4070 Super and GeForce RTX 4070 Ti Super first reviews, Super's cost performance is becoming more and more solid, and it is just a flagship positioning Super to put an end to NVIDIA's GeForce RTX GPU SUPER update, and this heavy responsibility falls on GeForce RTX 4080 Super.
From the perspective of positioning and parameters, GeForce RTX 4080 Super is already quite capable of playing, not only cheaper than GeForce RTX 4080 The initial price is 200 US dollars, the official starting price of 8099 yuan is more affordable, and it is also in place in one step in terms of performance, touching the ceiling of AD103. Yes, the GeForce RTX 4080 Super could really fit into the RTX 4090 series if it were to go a little further further.
Jokes aside, the GeForce RTX 4080 Super's price/performance ratio once again makes us feel that NVIDIA has invested enough money to upgrade the Super series, which also means that this GPU has more powerful combat power in generative AI, DLSS and ray-traced games, and content creation.
So is this really the case? Now let's present our debut review of the NVIDIA GeForce RTX 4080 Super Founders Edition.
The ultimate evolution of AD103
The most impressive thing about the GeForce RTX 4080 Super is the codename AD103-400-A1 core. We know that the AD103 retains all of the main features of the AD102 and uses the ADA Lovelace architecture derived from TSMC's 4N custom process, with 45.9 billion transistors and a core area of 379mm.
In the ADA Lovelace architecture, the GPU contains several GPCs (Graphics Processing Clusters), GPCs contain several TPCs (Texture Processing Clusters), several Streaming Multiprocessors (SMs), and then CUDA and RT cores, tensor cores, and so on. The number of TPCs contained in each GPC is equal, and the GPC and TPC are physically shielded when the GPUs are positioned and differentiated.
The full AD103 contains 7 GPCs, but each GPC contains different TPCs. As shown in the figure below, while each of the 6 GPCs has 6 TPCs, there is also 1 GPC with only 4 TPCs. For example, at the time of the RTX 40 series release, the GeForce RTX 4080 was implemented by shielding 1 set of TPCs.
Interestingly, the GeForce RTX 4080 Super, which also uses the AD103, obviously will not make any concessions in the core configuration, just from the core code name AD103-400-A1, we know that this GPU is not simple. Yes, the GeForce RTX 4080 Super is the full AD103 with 10,240 CUDA Cores, 80 RT Cores, 320 Tensor Cores, 320 Texture Units, and 112 ROPS. On the memory subsystem, there are a total of 10,240KB of L1 cache, 20,480KB of register heap, and 65,536KB of L2 cache.
However, it should be noted that compared to the three 8th Gen NVENC and 3 5th Gen NVDEC configurations of professional GPUs, the GeForce RTX 4080 Super Encoder is configured with 2 8th Gen NVENC and 1 5th Gen NVDEC to make a difference, which is a common operation for consumer-grade GPUs and has little impact on gaming performance and AI acceleration performance.
Not only that, but the GeForce RTX 4080 Super also has a boost frequency of 255GHz, paired with 256-bit 16GB GDDR6X, the memory bandwidth reaches 736GB S, and the TGP is still the same as the RTX 4080 at 320W. Also worth mentioning is the uncensored 64MB L2 cache, which is 12% more than the GeForce RTX 3080's 5MB L2 cache8x, which makes this in the game a huge benefit from ray tracing, especially path tracing. The third-generation RT Core and **Tensor Core can better play the effect of L2 cache.
At this point, the three GeForce RTX Super released at CES2024 have been released, and here we might as well use the RTX 4080 for comparison, and the approximate specifications are as follows.
Black armor
The biggest difference in the SUPER version of the Founders Edition is that the black middle frame used for reinforcement has become a dark black, which suddenly looks much more advanced. However, the overall design idea has not changed much, mainly relying on the symmetrical design of the misplaced fan and special-shaped PCB to improve the heat dissipation of the GPU.
In terms of material selection, the GeForce RTX 4080 Super Founders Edition is one-piece molded, and the heat dissipation fins are made of 99% aluminum alloy, which combines the characteristics of light weight, reliable rigidity, and good thermal conductivity. At the same time, the heat dissipation fin is thickened again, becoming a three-slot design, and the bracket hole at the end of the bezel is hidden under the magnetic baffle, which can be pried open with only a fingernail, and when it is restored, you only need to put the patch back, and the magnetic force will be automatically adsorbed.
In terms of fan selection, the GeForce RTX 40 Series Founders Edition basically boosts the fan blade diameter to 116mm, which generates enough airflow per minute to fill 35 basketballs. In addition, the fan blades are made of reinforced glass fiber, which is elastic and can be used with dynamic bearings for better quietness.
Under the fan, a design of 6 heat pipes with heat dissipation fins is used, and the power supply continues to use the latest 16pin (12vhpwr) interface, and the machine is packaged and distributed with a 3x8pin adapter cable, and the TGP is 320W.
In real-world gaming tests, the GeForce RTX 4080 Super Founders Edition showed particularly high performance, with power consumption generally kept below 250W in 4K gaming environments.
The interface design includes three DisplayPort 14A and 1 x HDMI 21 interface.
In stress tests, the GeForce RTX 4080 Super Founders Edition didn't have high GPU temperatures and surface temperatures. For example, the GPU temperature is only 58, and infrared detection can see that the entire case is in a fairly cool state.
Activate Super Combat Power
Now let's get to the test, which includes the Core i9-14900K, ROG Maximus Z790 Dark Hero, and GSkill Trident Z5 DDR5-7200 16GBX2 with 1250W power supply as support.
In the basic performance test, the main reference is 3DMark Time SPY, 3DMark Time Spy Extreme, 3DMark Fire Strike Extreme, 3DMark Fire Strike Ultra, and Port Royal, which measure the performance of DirectX 11 and DirectX 12. You can see that the GeForce RTX 4080 Super is about 20% better than the RTX 4070 Ti Super, and if you compare it with the previous generation RTX 3080, the increase is up to 70%.
So if you're using DLSS 2 and earlier regular games, the GeForce RTX 4080 Super starts at 4K max quality and is up to 80% faster than the RTX 3080 Ti. And with the GeForce RTX 4080 Super compared to the GeForce RTX 2080 Super, the improvement is quite exaggerated, and can range from 150% to 250%.
DLSS 3 and DLSS 3 brought to the GeForce RTX 40 Series5 technology is the most important, due to the huge performance improvement of Tensor Core, supplemented by the combination of optical flow accelerator (OFA) and DLSS, a more three-dimensional analysis method is built, so that the actual execution process of the GPU is compressed to the original 1 8.
dlss 3.5 is a set of AI-based algorithms to improve image quality and frame rate at the same time, including DLSS frame generation (FG) and ray reconstruction (RR) technology, plus super resolution (Super Resolution), DLAA, frame generation (Frame generation) and other technologies integration.
If you use the RTX 3090 as a comparison, you will find that even with the blessing of 24GB of video memory, the RTX 3090 is weaker than the GeForce RTX 4080 Super in terms of picture smoothness regardless of 4K or 8K resolution.
In DLSS 3 and later games, the GeForce RTX 4080 Super is a testament to its combat prowess. Here we directly turn on the 4K highest image quality, turn on ray tracing or path tracing. As you can see, the GeForce RTX 4080 Super can be improved by up to 60% to more than 200% compared to the RTX 3080 Ti, and compared to the RTX 2080 Super, the maximum is more than 800%, of course, because the GPU cannot push it, and in terms of general results, the GeForce RTX 4080 Super is not a problem with a 300% increase over the RTX 2080 Super.
Even the fusion of DLSS 3 and Reflex makes the GeForce RTX 4080 Super even more impressive in competitive games. Especially at 1080p resolution, games such as Valorant and Counter-Strike 2 achieved a frame rate of more than 300 fps, and the system latency was reduced to less than 10ms with reflex.
Here we use DLSS 35 of "Cyberpunk 2077" for a comparison of picture quality, you can see in DLSS 35 With the help of light reconstruction, the wet reflection becomes clear and natural, and the edge of the object still shows the effect of light reflection in the dark shadow, which is mainly achieved under the premise of obtaining a higher frame rate.
AI and creation are very capable of fighting
Next up is the AI testing session. GeForce RTX 4080 Super with 16GB GDDR6X is a huge step forward for generative AI. Here we might as well take Stable Diffusion XL as a reference, Stable Diffusion XL itself needs to rely on Transformer transformers and convolutional neural networks to operate, both of which require GPUs for massively parallel computing, which is undoubtedly what NVIDIA does best. Because of this, the rapid development of AI in recent years has been largely driven by the development of NVIDIA GPU technology.
The Stable Diffusion XL chosen here is quite stable Diffusion 15 is 3 times larger, resulting in a higher amount of data to be calculated, but also better image quality. This is configured with Stable Diffusion Web UI + TensorRT to describe the generation of 10 interstellar voyage images with a resolution of 768x768 in shorter language**, 1 per operation**, that is, a total of 10 executions, and the number of sampling steps is set to 50.
The text description is: Epic Space Battle Over a Water Planet, Many Ships, Cinematic Lighting, Depth of Field, Hyper-Detailed, Beautifully Color-Coded, 8K, Many Details, Chiaroscuro Lighting. By recording the time, the estimation efficiency per minute is calculated, and the reference formula is 60 (totaltime (batchsize * batchcount)) = images per minute.
The GeForce RTX 4080 Super easily outperforms many GPUs that have participated in recent reviews, and even compared to the RTX 4070 Ti Super with the same video memory, it has a performance improvement of more than 25%, which is quite a significant improvement. This means that the GeForce RTX 4080 Super can already be used for local AI deployment, whether it is the current Wensheng graph acceleration, graph generation, or even ** generation, or large language model acceleration, GeForce RTX 4080 Super has provided a very rich computing power foundation.
At the same time, the domestic software D5 renderer also has a good impact on DLSS 35 AI acceleration provides support. Due to DLSS 35 It is trained with 5 times more data than DLSS 3, so it can also better identify ray tracing effects and make more reasonable judgments in temporal and spatial data, so as to achieve efficient scaling.
In the Minimalist Villa 1 scene tested in the real world, the GeForce RTX 4080 Super made the D5 renderer run extremely smoothly, easily surpassing the 60 fps smoothness, which is more than 35% better than the RTX 4070 Ti Super a week ago.
Not only that, DLSS 3The addition of 5 also made a qualitative change in the scene. When DLSS 35. After turning off, the reflection on the ground becomes very rough, the glass on the left side also loses its refraction effect and transparency, and the frame rate is reduced to less than half of the original.
The last AI test is to use On1 Resize AI 2022 to improve the ** resolution. It can be seen that the GeForce RTX 4080 Super has completed the resolution increase in the shortest time, which is half the time compared to the RTX 3080.
The large number of Cuda Cores and RT Cores is also a strong point of the GeForce RTX 4080 Super. In the V-Ray 5 Benchmark, Cuda Cores and RTX are tested separately, and with more Cuda Cores, the GeForce RTX 4080 Super has a strong performance in both CUDA performance and ray tracing, which is 60% better than the RTX 3080.
The other Octane Render RTX is used to test the ray tracing performance of the GPU, which tests the sample rate generated by the GPU in complex scenes, and the GeForce RTX 4080 Super is 50% better than the RTX 3080.
In addition, the GeForce RTX 4080 Super is up to 80% higher than the RTX 3080 in the three reference scenarios introduced by the Blender Benchmark.
In the professional software creation session, we cited SpecViewPerf 2020, which is a special test of the 3D graphics performance of systems running under OpenGL and Direct X APIs, including 3DSMAX, Catia, Creo, Energy, Maya, Medical, SNX, and SolidWorks. The GeForce RTX 4080 Super delivers a 50% to nearly 90% increase in frame rates compared to the RTX 3080 for professional applications.
Write at the end: define the superceiling
The great thing about the GeForce RTX 4080 SUPER is that it has achieved the current Super Ceiling with a cheaper pricing strategy than the RTX 4080, and the full AD103 GPU does not disappoint, and the combat power consumption of the RTX 4080 is not improved, and the DLSS 3 game performance at 4K resolution is amazing.
In fact,GeForce RTX 4080 Super is already a GPU that can be full of special effects,Run all the masterpieces smoothly at 4K resolution,All you have to do is enter the game,The game's detection system automatically adjusts the highest configuration,Or manually pull the image quality slider to the head,Experience playing games and never look at the peak of life with picture quality settings,At this time, the pricing of 8099 yuan will be particularly cost-effective。
Outside of gaming, the GeForce RTX 4080 Super has shown itself to be an all-rounder, whether it's accelerating generative AI with Stable Diffusion XL or DLSS 35 Accelerated D5 renderer, or massive CUDA Cores to accelerate professional software such as 3DSMAX and CATIA, or replaced with NVIDIA Studio drivers to access the NVIDIA Omniverse metaverse.
The GeForce RTX 4080 Super is a PC that can easily be turned into an all-rounder, so even if you're a professional artist, designer, or game developer, you can accelerate the thrill of this GPU or professional app. In particular, the excellent AI performance of the GeForce RTX 4080 Super with 16GB GDDR6X large video memory allows users to no longer pay for AI services, and rely on the local AI performance of GeForce RTX 4080 Super to easily realize Wensheng graph, ** generation, and large language model acceleration, providing a sufficient computing foundation for the upcoming localized AI era.
And this product, which represents the highest ceiling of the current Super series, also interprets the cost performance very well. If you want to get plenty of AI performance and run 3A titles at 4K at maximum quality, and you want to work with **, graphics, or professional software with ease, the GeForce RTX 4080 Super is the right GPU for you.