NVIDIA CEO Jensen Huang unveiled the long-awaited next-generation gaming graphics cards from his kitchen. As expected, older solutions were announced today: GeForce RTX 3090, GeForce RTX 3080 and GeForce RTX 3070. Video cards are built on Ampere generation GPUs, made using Samsung's 8nm process technology, while their Turing generation predecessors were manufactured using 12nm technology TSMC.
GeForce RTX 3090
The transition to a new, thinner process technology allowed much more transistors to be placed on a chip, due to which the number of functional blocks increased significantly. Unfortunately, Jensen Huang did not specify the GPU configurations, and from the characteristics he named only the parameters of the memory subsystem. The GeForce RTX 3090 is equipped with 24 GB of GDDR6X memory with an effective clock speed of 19,5 GHz. Together with a 384-bit wide bus, this provides a memory subsystem bandwidth of 936 GB / s. The GeForce RTX 2080 Ti, recall, had only 11 GB of GDDR6 memory with about one and a half times less bandwidth.
The peak theoretical performance of the GPU of the new video card is 36 Tflops "with traditional rendering using shading (shaders)" (Shader-TFLOPS). Ray tracing performance is rated at 69 teraflops (RT-TFLOPS), while tensor core performance reaches 285 teraflops (Tensor-TFLOPS).
According to NVIDIA, this video card is designed for games at 8K resolution, in which it is able to provide a stable 60 FPS. It is also noted that this "beast" is about 50% faster at 4K resolution than the Titan RTX.
GeForce RTX 3080
The head of NVIDIA called the GeForce RTX 3080 video card the new flagship, bypassing the question of what the GeForce RTX 3090 is in this case. Nevertheless, the GeForce RTX 3080 looks quite βflagship-likeβ, both externally and in terms of performance.
Of the characteristics, the presence of 10 GB of GDDR6X memory with an effective frequency of 19 GHz was noted. The memory is connected via a 320-bit bus, which ultimately provides a bandwidth of 760 GB / s. Recall that the previous GeForce RTX 2080 had only 8 GB of GDDR6 memory with a bandwidth of 448 GB / s.
The performance of the GeForce RTX 3080 with traditional rendering using shaders is 30 TFLOPS (Shader-TFLOPS). When processing ray tracing, the video card provides 58 Tflops (RT-TFLOPS), and in operations on tensor cores it provides up to 238 Tflops (Tensor-TFLOPS). According to the head of NVIDIA, the GeForce RTX 3080 graphics card is twice as productive as its predecessor, the GeForce RTX 2080.
GeForce RTX 3070
The simplest card of the three presented, the GeForce RTX 3070, like its predecessor in the face of the GeForce RTX 2070, has 8 GB of GDDR6 memory.
The performance level of the GeForce RTX 3070 graphics accelerator was 20 TFLOPS with traditional rendering with shaders (Shader-TFLOPS). Ray tracing performance is rated at 40 teraflops (RT-TFLOPS), while tensor core performance reaches 163 teraflops (Tensor-TFLOPS). NVIDIA claims that the new GeForce RTX 3070 is more powerful than the previous flagship, the GeForce RTX 2080 Ti.
The first of the new products will go on sale GeForce RTX 3080. This will happen this month - September 17th. The cost will remain at the level of its predecessors - $ 699 (in Russia - 63 rubles). The more affordable GeForce RTX 500 will appear on the shelves sometime in October, and its price will be $3070 (499 rubles) - again, at the level of its predecessor. Finally, the GeForce RTX 45 will go on sale on September 500, immediately performed by both NVIDIA and its partners, at a price of ... $ 3090 (24 rubles). There has been no price increase here.
Source:
Source: 3dnews.ru