5 things you might have missed about NVIDIA RTX 4000 series GPUs

NVIDIA's new flagship, the RTX 4090 GPU, with up to 4x the performance of its predecessor, was revealed today. It is part of the GeForce RTX® 40 Series of GPUs, which is intended to bring breakthrough performance for gamers and artists.

The RTX 40 Series heralds a new age of real-time ray tracing and neural rendering, which employs AI to produce pixels. They are the first GPUs in the world to be built on the new NVIDIA Ada Lovelace architecture and provide enormous generational jumps in performance and efficiency.

---

Here are 5 key takeaways from the NVIDIA GTC Keynote

1) What do the new GPUs bring to the table?

When playing fully ray-traced games, the RTX 4090 with DLSS 3 is up to four times quicker than the RTX 3090 Ti, with DLSS 2 from the previous generation. In modern games, it is also up to twice as quick while using the same 450W power draw.

With 16,384 CUDA cores, 76 billion transistors, and 24 GB of fast Micron GDDR6X memory, it coughs up a stable 100 frames per second and up for 4K gameplay. Sales will start on Wednesday, October 12. The RTX 4090 will cost $1,599.

NVIDIA also unveiled the RTX 4080, which will be available in two variations. The RTX 4080 16GB is twice as fast in modern games as the GeForce RTX 3080 Ti and features 9,728 CUDA cores and 16GB of high-speed Micron GDDR6X memory.

The RTX 4080 12GB is faster than the RTX 3090 Ti, the flagship GPU from the previous generation, thanks to DLSS 3. It contains 7,680 CUDA cores and 12GB Micron GDDR6X memory. Prices for the two RTX 4080 versions will start at $1,199 and $899, respectively, in November.

---

2) Primary innovations

The RTX 40 Series GPUs include a number of significant technological advancements under its belt, such as the following:

1. Up to 83 teraflops of shader capability in streaming multiprocessors, which is a 2x improvement over the previous generation.

2. Up to 191 effective ray-tracing teraflops in third-generation RT Cores, which is 2.8 times more than in the previous iteration.
3. Fourth-generation Tensor Cores use FP8 acceleration to reach up to 1.32 Tensor petaflops, which is a 5x improvement over the previous generation.
4. Shader Execution Reordering (SER) is a technique for increasing execution efficiency by dynamically rescheduling shading tasks to make greater use of the GPU's resources. SER increases ray-tracing speed by up to 3x and in-game frame rates by up to 25%, which is an invention as significant for CPUs as out-of-order execution was for CPUs.

---

3) DLSS3

Huang unveiled NVIDIA DLSS 3, the company's newest advancement in its Deep Learning Super Sampling neural graphics technology for video games and innovative apps. For significantly fluid gameplay, the AI-powered system may produce complete frames.

By enabling the GPU to produce full frames independently, it can get around the game's CPU performance restrictions. More than 35 games and applications will soon use the technology, which is being added to some of the most widely used gaming engines in the world, including Unity and Unreal Engine.

---

4) RTX Remix

For NVIDIA Studio developers, the RTX 40 Series GPUs and DLSS 3 bring a whiff of fresh air. Without the need for proxies, 3D artists can render completely ray-traced worlds with precise physics and realistic materials and see the changes as they happen.

The addition of new dual, eighth-generation AV1 encoders and better GPU performance also helps to improve live streaming and video editing. Three upgrades to the NVIDIA Broadcast software development kit are now accessible to partners and include enhancements to Virtual Background quality, Face Expression Estimation, and Eye Contact.

NVIDIA RTX Remix, a modding platform to produce amazing RTX remasters of vintage games, will soon be included in NVIDIA Omniverse, part of the NVIDIA Studio software package.

With the aid of RTX Remix, modders may quickly and simply activate RTX using ray tracing and DLSS as well as automatically upgrade materials with potent AI tools.

---

5) Revamped Ray Tracing

The new third-generation RT Cores contain two significant new hardware components and have been improved to offer 2x quicker ray-triangle intersection testing. The Micro-Mesh Engine creates micro-meshes on the fly to generate more geometry, and the Opacity Micromap Engine accelerates ray tracing of alpha-test geometry by a factor of 2x.

The advantages of higher geometric complexity are delivered by the Micro-Mesh Engine without the usual performance and storage drawbacks of complicated geometries.