GeForce RTX 4070

The GeForce RTX 4070 is based on the new GeForce RTX 4070 GPU based on the AD104 architecture. At the heart of this powerful device are 5,888 CUDA cores capable of delivering 29 teraflops (TFLOPS) of computing power for rendering bitmap graphics and 12 gigabytes of GDDR6 video memory. The new GPU features 184 Gen 4 Tensor Cores delivering 466 tensor TFLOPS (with sparsity) for AI and DLSS processing, and 46 Gen 4 Ada RT cores capable of delivering 67 RT-TFLOPS for next generation ray tracing processing.

NVIDIA Ada architecture features include a new shader execution engine (Shader Execution Reordering (SER), new RT cores, and an improved version of DLSS, making the RTX 4070 a great solution for gaming at any graphics level.

Ideal for content creation, the GeForce RTX 4070 GPU delivers superior gaming performance. The RTX experience is accelerated in top content creation applications, along with NVIDIA Studio drivers designed for maximum stability. In addition, exclusive tools are available that leverage the power of RTX to accelerate creative processes with the help of artificial intelligence.

NVIDIA Ada Streaming Multiprocessor

The RTX GPU has three core processors: a programmable shader that was introduced almost 20 years ago, RT cores for accelerating ray and triangle intersections, and an artificial intelligence processing pipeline called Tensor Cores.

The Ada architecture improves on all three RTX processors:

1. Programmable Shader: 29 Shader-TFLOPS, compared to 22 TFLOPS on the RTX 3070 Ti. SM Ada includes a new Shader Execution Reordering (SER) technology that redistributes tasks on the fly, doubling the speed of ray tracing shaders. SER is as significant an innovation for GPUs as out-of-order execution has become for CPUs.

2. Gen 4 Tensor Cores: The new Tensor Cores in Ada include the NVIDIA Hopper FP8 Transformer engine, which delivers over 466 sparse tensor terraflops in the RTX 4070 for AI workloads. That's double the 174 tensor teraflops on the RTX 3070 Ti. Compared to FP16, FP8 reduces storage requirements and doubles AI performance.

3. Gen 3 RT Cores: New Opacity Micromap Engine speeds up alpha-tested geometry ray tracing by a factor of 2, and new Micro-Mesh Engine allows additional geometry to be created on the fly, allowing for increased geometry complexity without the cost of building and storing BVHs . Ray and triangle intersection performance on Ada reaches 67 RT-TFLOPS, compared to 42 RT-TFLOPS on the RTX 3070 Ti.

4th Generation Tensor Cores

Tensor cores are high-performance computing cores specialized in matrix multiplications and accumulations that are used in artificial intelligence and high-performance computing applications. Tensor cores provide breakthrough performance for matrix calculations, which are critical for training and inferring deep learning neural networks. Sample output applications include NVIDIA DLSS 2.1 technology for gamers, which introduced a new 9x upscaling mode with 1440p internal rendering and upscaling to 8K resolution. In addition, many creative applications have begun to use artificial intelligence features to speed up the work of artists. NVIDIA Broadcast also offers noise removal tools,

Fourth-generation tensor cores of the Ada architecture build on the capabilities of previous Ampere GPUs, which supported many new data types and introduced sparsity that doubled the throughput of tensor cores compared to Turing tensor cores. Ada's fourth-generation tensor cores support the new FP8 data format, which was first introduced in the NVIDIA Hopper GPU architecture. Compared to FP16, FP8 data format reduces data storage requirements and doubles AI performance. With the new FP8 format and sparsity feature, the GeForce RTX 4070 delivers 466 teraflops of performance for AI inference tasks.

3rd Generation RT Cores

The third generation of RT cores in the Ada architecture are specialized hardware blocks to speed up BVH traversal and perform ray and triangle intersection calculations, which is critical for speeding up ray tracing.

RT cores free up the resources of the programmable SM shader cores, allowing them to perform other pixel, vertex, and compute tasks. Ada RT cores provide ray and triangle intersection test acceleration up to 2x faster than Ampere architecture GPUs, allowing developers to add more detail to their virtual worlds. The Ada architecture also features the Opacity Micromap Engine, which accelerates alpha-tested geometry ray tracing by a factor of 2, and the Micro-Mesh Engine, which generates micromeshing on the fly to generate additional geometry, allowing for more geometric complexity without the expense of creating and storing BVHs.

AV1 video encoding

The GeForce RTX 4070 offers hardware support for video encoding using the AV1 codec. AV1 provides improved image quality at the same bit rates as H.265/H.264, which is a big advantage for game streaming. With AV1, users can achieve the same image quality at lower bit rates, resulting in smaller file sizes and faster video loading. AV1 Ada encoder is 40% more efficient than H.264 encoder in Ampere architecture.

So, thanks to the improved technologies and architecture of Ada, the GeForce RTX 4070 is an excellent choice for content creation professionals and gamers who demand high performance and visual quality. With support for the latest technologies such as ray tracing, DLSS and AV1, the RTX 4070 graphics card delivers superior image quality and performance for today's and tomorrow's games and applications.