NVIDIA has been at the forefront of supercharging AI use with their latest GeForce RTX 40 Series GPUs. AI as both a tool for work and for entertainment has seen a steady rise over the years. Let’s take a look at the current set of AI features that GeForce RTX 40 Series GPUs can give us. This article is presented in partnership with NVIDIA.
Tech companies, both hardware and software have all started offering solutions varying from boosting performance with their own hardware, or creating plugins to full featured apps that leverage the power of Artificial Intelligence. NVIDIA has also contributed to this effort with introducing proof of concepts before such as the GauGAN concept, which is now known as NVIDIA Canvas.
Table of Contents
NVIDIA GeForce RTX GPU Features
While it all began with GauGAN for creatives, there’s an ecosystem of tools and features that NVIDIA made around that utilize the capabilities of the NVIDIA RTX GPUs. Introduced in late 2018 on the ‘Volta’ microarchitecture, RTX GPUs feature Tensor Cores designed for deep learning.
A quick history on Tensor Cores: it is a step up of the existing NVIDIA CUDA cores (GPU) that have better deep learning capabilities. It is NVIDIA’s proprietary form of an NPU (AI Processor), which we’re now seeing more and more from other hardware chip vendors as well. And the best I can interpret this technology since these involve a lot of math; a typical GPU would be treated as a two-way highway going from point a to point b. Whereas an NPU would be a complex road network, that can all lead from point a to point b, but it’s sprawling, leading to different other points in between a and b.
NVIDIA GeForce RTX GPUs have several features and benefits that cater a wide range of users. These cover general productivity, arts (digital art, video production, image and video generation, streaming) to software development (includes gaming, AI characters, modding, automation)
Productivity Application
While Microsoft has its own chatbot tool called CoPilot, NVIDIA has its own called ChatRTX. Currently in its demo form, this application provides us a GPT large language model to interface with our own content in our PC, or through a large language model that we can choose and customize. It utilizes the RTX GPU to perform AI tasks and it will even recognize your local language (as long as the AI model used at configuration is used).
Arts – Image Creation
For the arts, a wide scope where multiple disciplines involved, NVIDIA has several tools made inhouse and support for popular applications. Starting off with Digital Art – remember GauGAN? Now called NVIDIA Canvas, it is an app that converts conceptual paintbrush strokes and it lets the app understand and convert into backgrounds. It decreases the time and effort in working on a concept backdrop of a scene, allowing for rapid updates.
Of course, for other image generation purposes, there are plug-ins and applications made for generating images based on provided words or phrases called “Stable Diffusion”. These Stable Diffusion apps are driven using the power of NVIDIA RTX GPUs mostly and its Tensor cores. Some examples of Stable Diffusion software that runs locally are: Stable Diffusion by Automatic1111 (GitHub), Adobe Firefly, Gimp (OpenVINO Plugin for Gimp – Performance Mode uses the system’s GPU by default). These applications are installed and run locally, and will utilize our RTX GPU.
NVIDIA Tensor cores can deliver up to 1300 TOPS (that’s Trillions of Operations per Second). Having a high number of operations is key in making effective use of time and being able to make multiple samples for review.
Other work that AI can be of help with photos and image editing work mostly revolves around filling in content to fix/improve/complete, removing elements in the image, and even resizing while keeping quality of the image intact (upscaling). Adobe Photoshop’s quite known for their AI features starting with their Content-Aware fill since the early 2010s and have now progressed with more AI features and better leveraged by solid GPU support.
When I create thumbnails for articles here, there are times that pictures I took have some form of blemish, which I would end up cleaning up. Be it dust, or an object that wasn’t intended to be part of would require an image cleanup. Using the Clone Stamp tool to edit out an object or small dust particles/blemishes on an image is the go-to procedure. But with Photoshop’s Generative Fill, which uses AI tech, it aids in cleaning up the image with less manual effort.
Arts – Video Production and Streaming
The same can be said for video production, visual effects which usually require additional time in an editors/s work to add in can be automated with the aid of AI. Applications such as DaVinci Resolve, Premiere Pro, Capcut and others have added AI features in to cut down the time and process of performing tasks like color matching between different shots of a scene to create a more consistent atmosphere.
Video upscaling (similar to what we have in gaming for DLSS) is another feature that AI acceleration, by de-artifacting and replacing missing spots with appropriate pixels improving overall resolution. Moreover, these apps support NVIDIA’s encoder (NVENC) or other encoders like AV1 to ensure videos are rendered at a faster rate.
For the streamers, NVIDIA has Broadcast that offers many features including setting virtual backgrounds, audio and video noise cleanup, and unique to the app features like “eye contact” and “auto frame”, which uses AI to keep our eyes affixed to the camera (even if we’re taking glances to our script, or chat window), and track our movements as if our camera is moving but in reality AI crops and zooms the frame dynamically. And it works with Streaming applications such as OBS as a webcam/audio source.
While Broadcast works on improving the audio and video output, RTX GPUs also aid streaming applications like OBS Studio, Streamlabs, etc. Same with rendering videos or production work, RTX GPUs offer the same NVENC encoder for streaming video and it is still able to play games.
Gaming
Gaming is a core component of a GPU’s use. With GeForce RTX GPUs the latest in graphics technology like ray tracing, high textures on 1440p to 4k gaming is possible. With the introduction of Tensor Cores in the RTX 20 series gamers are able to experience immersive visuals in major games like Cyberpunk.
With the latter releases in RTX 30 and RTX 40 series AI support was introduced in the form of Deep Learning Super Sampling (DLSS). As a gamer, it’s paramount to get what we paid for, for our graphics card to be able to perform well in its task and some bonus features like DLSS. As experienced with owning and reviewing some NVIDIA GeForce RTX graphics cards in this site, performance for games is impressive.
NVIDIA DLSS aims to aid in improving frame rates by way of resampling lower resolution graphics to higher resolution by reconstructing them through AI (DLSS Super Resolution), while DLSS Ray Reconstruction aids in improving Ray Tracing output by learning a scene and adding more detail, hence lower blurring and increasing quality. And finally, DLSS Frame Generation, which uses AI to generate additional frames to improve game performance, while working with NVIDIA Reflex to reduce lag.
Programming
Everything in this article is tied into programming, even if there’s AI support for apps, who in turn writes the features for other applications, chatbots, 3D NPC, 3D Sceneries have Ray Traced supported updates of games? As a software programmer by trade, companies are shifting to utilizing AI more in aid of handling tasks and finding ways for our clients to do their work more efficiently. And having a discrete RTX GPU on a work laptop makes a fair reason to utilize its capabilities for developing software.
NVIDIA hosts libraries and builds for learning and developing Large Language Models (LLM) as we’ve seen with ChatRTX, it utilizes NIM for inference modelling, but that being a large-scale operation, may end up utilizing enterprise level GPUs.
RTX AI ToolKit, RTX Remix and NVIDIA ACE (for rendering an NPC sample: Mecha BREAK ) are tools NVIDIA has in the shed for helping programmers and artists in setting up AI powered games, apps and chatbots that leverage NVIDIA RTX GPUs.
GeForce RTX GPU Performance Lets it Do it All
GPUs’ performance lets it do everything under the sun, most especially when AI is involved. While we know that general NVIDIA GPU performance in games is great, it also shines for AI tasks. Here is some information shared by NVIDIA on their effort with AI development.
For Desktop GeForce RTX GPUs, the RTX 4090 being the flagship GPU performs at a whopping 1321 AI TOPS. Going down the line, there’s a gradual decrease in Trillions of Operations per Second, but not that while the GeForce RTX 4060 has the lowest score for this generation, it still outdoes any GPU from its peers (Note: We’ll most likely start adding in AI Tests via Stable Diffusion for the next batch of graphics cards).
While desktop GeForce RTX GPUs have the best scores, the laptop RTX GPUs still hold their own with strong performance, albeit power and thermal is limited for its form factor. Differences between their desktop counterparts scale at a different range with the more mainstream RTX 4060 desktop and mobile GPUs having a closer TOPS gap. The higher end mobile RTX GPUs meanwhile have a wider difference between platforms due to the maximum power supplied to a laptop (240-250W).
To sum it up, an NVIDIA GeForce RTX GPUs can handle now handle AI development and aid more in gaming and creation work more than its previous generations or peers.
| Develop | Game | Create |
| 30x | 8x | 13x |
| Faster AI Model Training | Higher FPS in Games | Faster Image Generation |
Source – AI Benefits NVIDIA
