Starting from 12,999 yuan, the NVIDIA GeForce RTX 40 series redefines graphics cards

time:2023-02-08 02:35:13 source:scripttoolbox.com author:Loudspeaker box
Starting from 12,999 yuan, the NVIDIA GeForce RTX 40 series redefines graphics cards

At the 2022 fall GTC conference, NVIDIA officially released the latest generation of desktop graphics cards - the GeForce RTX® 40 series, which will use NVIDIA's new Ada Lovelace architecture core, and CEO Huang also introduced us to many new Ada Lovelace architectures. Features including: TSMC 4N custom process, DLSS 3, Streaming Multiprocessor, 3rd Gen RT Cores, 4th Gen Tensor Cores, Shader Execution Reordering (SER), Ada Optical Flow Accelerator, and Dual NVIDIA Encoders (NVENC) Wait, it really feels like a bit of toothpaste squeezed out. The first graphics cards to be launched include the GeForce RTX 4090 and GeForce RTX 4080 (16GB/12GB) versions. NVIDIA will launch the RTX 4090 and RTX 4080 (16GB) FE editions in limited quantities. The RTX 4090 will be available on October 12 with a suggested retail price of ¥12,999; the RTX 4080 (16GB/12GB) will be available in November with a suggested retail price of ¥9,499 and ¥7,199, respectively. From the perspective of performance and retail quotations, the pricing of these three graphics cards seems to be a bit reasonable. After all, the performance improvement is basically twice that of the previous generation graphics cards with the same positioning, and the power consumption is also kept at the same level. The flat retail quotation is also relatively Ideal too. In addition to NVIDIA's FE version, domestic top graphics card suppliers such as ASUS, Colorful, Gainward, GALAXY, GIGABYTE, Yingzhong, MSI and Zotac will launch GeForce RTX 4090 and 4080 GPU standard and overclocked versions. The non-public version of AIC has also been released. Now let's take a look at the information that can be provided to you. Since the NDA is on October 12th, we don't have too much information at present, or the information we can tell you is quite limited, so we can directly take the ready-made GALAXY graphics card specification sheet as an example. The first three graphics cards, GeForce RTX 4090 and GeForce RTX 4080 (16GB/12GB), all adopt the latest TSMC 4N custom technology, and the architecture is the Ada Lovelace architecture (abbreviated as "ADA") that we have just mentioned. The GeForce RTX 4090 is currently the highest specification, the core code is AD102-300, with 76 billion transistors, 16384 CUDA cores and 24 GB of high-speed Micron GDDR6X memory, 8K HDR games are not a problem at all. GeForce RTX 4080 16GB uses another core, the core code is AD103-300, with 9728 CUDA cores and 16 GB high-speed Micron GDDR6X memory, the memory bit width is also reduced to 256Bit, but the memory frequency is the highest among the three. GeForce RTX 4080 12GB specification is lower, the core code is AD104-400, with 7680 CUDA cores and 12GB Micron GDDR6X memory, the memory bit width is only 192Bit. According to the official information given by NVIDIA, it is estimated that there will be no new products in the near future, and the 40 series and 30 series will compete in the same field for a period of time. Going down is actually difficult. GeForce RTX 4080's video memory bit width has reached 192Bit, so let's guess, or maybe the future GeForce RTX 4060 will be 128Bit? ? ? There's a lot of nonsense, let's come again. Ada Lovelace The naming of the Ada Lovelace architecture is really interesting. According to the information from Baidu Encyclopedia, Ada Lovelace, known as the "Queen of Numbers", wrote the first computer program in history and was recognized as the world's first computer programmer. I don't know if this naming means that NVIDIA wants to redefine graphics cards with the appearance of GeForce RTX 40 series graphics cards. TSMC customizes the 4N process technology, and the AD102 core can plug 76 billion transistors, and the official said that there are more than 18,000 CUDA cores, which means that the AD102 core used by the RTX 4090 is not a full-blooded version. But even so, the Ada Lovelace architecture core still has about 70% more transistors than the previous generation Ampere architecture core, while achieving a performance-to-power ratio of up to 2 times. It can only be said that the Ada Lovelace architecture + TSMC 4N is really fierce , but the 450W TDP really puts a lot of pressure on the heat dissipation of the graphics card. The first is the stream processor. GTC2022 Lao Huang introduced us 90 TFLOPS, but NVIDIA's official press release introduced that the RTX 4090 has a shader capability of up to 83 TFLOPS, so the full core AD102 will be 90 TFLOPS. Peak computing power; compared to the full specification GA102 core 40 TFLOPS of the RTX 3090 Ti graphics card, the theoretical performance has been more than doubled. The second is the third-generation RT Cores and two important hardware units: the Opacity Micromap engine and the new Micro-Mesh engine, which can provide us with twice the performance of ray and triangle intersection. Speaking of which, GeForce RTX 40 series graphics cards will have more powerful ray tracing capabilities, enabling physically accurate graphics calculations even if the rendering geometry of environments and objects is more complex. The third improvement is the fourth-generation Tensor Cores. The FP8 tensor processing performance has been improved to 1.32 Petaflops, which is 5 times stronger than the previous generation. Stronger computing power means that the stronger the graphics card's ability to try and learn, the greater the AI ​​computing power will be, including the use of NVIDIA Omniverse and NVIDIA Broadcast, which can help us achieve more efficient computing. The fourth shader performs reordering (SER), and it is easy to understand the role of SER (scheduler) by looking at the picture. According to NVIDIA's official statement: "Improve execution efficiency by rearranging shader loads on the fly, so as to better Utilize GPU resources. As a major innovation like the out-of-order execution of CPU, SER brings up to 3 times performance improvement for ray tracing, and the overall game performance can be improved by up to 25%." This means that SER can sort out computing tasks in advance. , and the SM unit completes the task according to the demand. It can be seen that each generation of RTX graphics cards has significant technical improvements, especially the GeForce RTX 40 series graphics cards, as the third-generation RTX graphics cards, can provide more powerful ray tracing performance, and also achieve faster and more efficient real-time computing. As mentioned above, it is actually to introduce NVIDIA's latest generation of Deep Learning Super Sampling deep learning super sampling technology DLSS 3 for games and creative applications. The original DLSS technology has already led the industry, but under the Ada Lovelace architecture DLSS 3 will be the next revolution in neural network graphics technology. DLSS 3 can also overcome CPU-bound games by allowing the GPU to generate entirely new frames, leveraging artificial intelligence-driven technology that can generate entirely new frames to dramatically improve game performance. And this technology has already been used in the world's most popular game engines including Unity Engine and Unreal Engine and has been supported by many of the world's leading game developers. In addition, we can use NVIDIA RTX Remix in NVIDIA Omniverse to add RTX effects to classic games. After we capture the game material to some game material, we can use the powerful computing power of GeForce RTX 40 series graphics cards through RTX Remix software, and use the deep learning model included in the AI-assisted toolset to improve the resolution of textures and materials; we can also use An AI model converts materials into materials with precise physical properties, such as adding lighting effects, etc., so that we can make an old game more RTX. Of course, there are still many highlights at the 2022 Autumn GTC Conference, including Omniverse applications and cases, Thor processors, NVIDIA DRIVER platform, Jetson Orin Nano micro-robotic computer, etc. Those who are interested can watch the replay on NVIDIA official website. As for the performance of the GeForce RTX 40 series graphics cards, we will know when the ban is lifted on October 12.

(Responsible editor:Monitor)

Related content