NVIDIA’s Lovelace GPU upgrades graphics with AI

Check out more coverage GTC Fall 2022.

A new major family of graphics chips developed by nvidia Powered by the next generation of Ada Lovelace architecture, which leverages Artificial Intelligence (AI) to create more realistic visuals in games.

At the company’s GTC 2022 conference, NVIDIA CEO Jensen Huang said the Lovelace architecture supports the latest GeForce RTX 40 GPUs. The architecture is named after the 19th-century mathematician considered one of the early pioneers in computer science.

NVIDIA said the top-of-the-line GPU in the gaming family, the RTX 4090, delivers double the performance and represents a major step up in power efficiency over the previous generation based on the Ampere GPU architecture circa 2020.

The Lovelace GPU is filled with 76.3 billion transistors, which is about 2.7 times more than its Ampere GPU and close to the same number of transistors as its Hopper GPU for data centers, plus over 16,000 CUDA cores.

These features give it one of the most advanced graphics chips on the market at a time when it’s feeling increasing pressure from AMD (with the impending RDMA 3 GPU architecture) and Intel (with high-performance Arc GPUs).

Lovelace architecture

The chips will be manufactured by TSMC on a dedicated “4N” technology node. That’s a huge step forward compared to NVIDIA’s latest generation of gaming graphics chips, which Samsung Electronics designed at an 8nm node.

The company said that the use of newer processing technology, combined with improvements in the infrastructure, gives Lovelace-based graphics processors twice the power efficiency of their previous generation using Ampere.

The Lovelace architecture stands out from the NVIDIA Hopper architecture announced at the beginning of the year. While Hopper will power the H100 GPU for high-performance computing and AI workloads, the Lovelace architecture is ideal for general-purpose and graphics-intensive workloads — everything from creating lighting and physically accurate objects in games to building digital twins using NVIDIA’s Omniverse software platform.

Digital Twins are large-scale simulations – for example, factory floors or cars – that give you a way to test and validate designs or processes in the safety of the virtual world before they are brought into the real world.

NVIDIA said the RTX 40 series of GPUs offer a range of innovations across the board. For example, the new generation of streaming multiprocessors are 3 times faster than the previous generation, according to the company. The units can provide up to 90 TOPS of performance for shaders, which are used to determine the correct levels of light, dark and color during scene rendering and are used in every modern game.

One of the most notable features of the Lovelace architecture is what NVIDIA calls a “shading implementation rearrangement”. This increases execution efficiency by rescheduling shader loads on the fly. The technology works in a seemingly similar way to an out-of-order implementation in the CPU. NVIDIA said the Lovelace architecture uses it to improve ray tracing performance by up to 3X and frame rates by up to 25%.

The chips also contain a new generation of ray tracing (RT) Cores that deliver up to 200 TFLOPS to create more accurate reproductions of light, making shadows and reflections more realistic in a real-time scene.

Graphics chips based on the Lovelace architecture feature a new video encoder with AV1 codec support.

AI-powered graphics

The Lovelace architecture also brings NVIDIA’s fourth-generation tensioner cores to the fold. These units Specifically designed to implement “matrix multiplication and aggregation” operations at the core of machine learning.

NVIDIA said its next-generation tensor cores are five times faster than the previous generation, providing up to 1,400 TFLOPS, or 1.4 quadrillion operations per second, for the company’s FP8 format for AI workloads.

New and improved inference processing nuclei It belongs to the same generation as the one used in the GPU Hopper. As a result, they Equipped with the same “Transformer Engine” as Hopper GPUs.

A new hardware engine called an optical flow accelerator completes the tensor cores. It uses machine learning to compare high-resolution pairs of frames and predict the movement of objects displayed in a 3D scene. This gives Lovelace the ability to display everything in the frame, from particles and reflections to shadows and lighting, at an earlier time, increasing the frame rate without affecting the sharpness of the image.

New tensioner cores and Hardware accelerators are what make one of the most advanced graphics technologies in Lovelace GPUs possible: NVIDIA’s 3rd Generation Deep Learning Super Sampling (DLSS).

Displaying every pixel in vast virtual worlds or in games with precise physics, bright lighting, and realistic materials requires a tremendous amount of computing power. But instead of trying to display everything in a scene, technology ignores it part of a pixel. It then uses machine learning to create new pixels that fill in the blanks, resulting in sharp, high-resolution graphics that run at frame rates that exceed the computational capabilities of NVIDIA’s GPUs.

Rather than just creating new pixels, DLSS 3 uses AI to create entirely new frames, increasing frame rates by up to 4X compared to without DLSS. The technology can give a performance boost even if the game is throttled by the CPU.

Everyone in the family

NVIDIA said the flagship RTX 4090 is one of the most advanced on the market, equipped with 16,384 CUDA cores, up from 10,752 in its predecessor, with the core clock frequency boosted by more than 30%.

All the new hardware features, along with a host of improvements to the Ada Lovelace architecture itself, mean the processor can render 4K gameplay at over 100 frames/sec.

The RTX 4090, paired with 24GB of high-speed GDDR6X memory from Micron Technology, has the same 450W power shell as the previous generation. The chips use PCIe Gen 4 lanes for communication.

NVIDIA said the RTX 4090 brings up to 4 times the performance of the current high-end graphics chip, the RTX 3090. It also delivers up to twice the speed of its predecessor with the same level of power consumption.

The semiconductor giant also introduced a new mid-range graphics processor for the gaming family, called the RTX 4080. The new chip comes in two parts Various memory configurations: 12 or 16 GB of GDDR6X memory.

While none of these configurations were as advanced as the RTX 4090, NVIDIA said that Lovelace-based GPUs can render high-quality graphics with more realistic lighting even faster than the current RTX 3090.

The high-end RTX 4090 will cost $1,599 when it comes to market next month, while the mid-range RTX 4080 will retail for $899 (for a 12GB GDDR6X configuration) and $1,199 (with 16GB of GDDR6X).

Check out more coverage GTC Fall 2022.