Nvidia releases Ada Lovelace architecture GPU


NVIDIA officially released the Ada Lovelace architecture GPU. Huang Renxun said that Ada provides a qualitative leap for gamers and is paving the way for future games based entirely on simulation. The Ada Lovelace architecture GPU streaming multiprocessor has a shader capacity of up to 83 TFLOPS, which is twice the throughput of the previous generation. The effective ray tracing computing power of the third-generation RT Cores reaches 191 TFLOPS, which is 2.8 times that of the previous generation. The fourth generation of Tensor Cores has up to 1.32 Petaflops of FP8 tensor processing performance, which is more than 5 times the acceleration performance of the previous generation using FP8.

NVIDIA introduced Shader Execution Reordering (SER) technology, which improves execution efficiency by rescheduling shader loads on the fly to better utilize GPU resources. As a major innovation as the CPU's out-of-order execution, SER brings up to a 3x performance boost to ray tracing and up to a 25% increase in overall gaming performance.

Additionally, the Ada Optical Flow Accelerator delivers a 2x performance boost, enabling DLSS 3 to predict motion in a scene, and enabling neural networks to increase frame rates while maintaining image quality.

In terms of energy efficiency, the Ada Lovelace architecture GPU uses TSMC 4N custom process technology to achieve a performance-to-power ratio leap of up to 2 times.

In terms of encoders, the Ada Lovelace architecture GPU uses dual NVIDIA encoders (NVENC), which reduces output time by up to half and supports AV1. OBS, Blackmagic Design DaVinci Resolve, Discord, and many more are already using the NVENC AV1 encoder.

Nvidia also introduced a brand new ray tracing technology:

The new 3rd generation RT Cores can provide 2x the performance of ray and triangle intersection, and two new important hardware units. The Opacity Micromap engine increases the performance of ray-traced Alpha-Test geometry by 2X; while the new Micro-Mesh engine generates micro-mesh on the fly to generate additional geometry. The Micro-Mesh engine increases geometry richness without sacrificing the performance and storage costs of traditional complex geometry processing.

Post a Comment