It’s 2024, and AI is easily the hottest buzzword of the year with the amount of products that makes at least one mention of it. Does the emergence of AI concern you, the average Joe, you may ask – there’s a good chance that some parts of your smartphone is currently being powered by AI to some degree, and soon your laptops will feature onboard AI capabilities as well. For those who need more AI-powered oomph, there’s the GeForce RTX GPUs – here is one such example in the form of Palit’s GeForce RTX 4060 Ti Dual card.
I’m Out Of The Loop – What’s AI, Again?
AI is short for “artificial intelligence” – to simplify it, the ‘true’ implementation of AI involves the ability to interpret information and perform calculations that are not explicitly written in the code by developers. Think of it like dogs getting trained by police to sniff certain things, except it’s a robot, and it’s a lot faster at it. (Here’s a quick explainer from NVIDIA.)
So what’s good about training AI this way compared to plain old coding a software to do exactly as it says? In certain applications, this can be very useful in things like image recognition, medical diagnostics, image and video enhancement, and more. The next step is what makes the word “AI” a commonplace since ChatGPT boomed in popularity – generative AI. Instead of interpolating the data, generative AI creates new data, making it capable of generating texts, images, and even videos.
The AI-Powered Advantage Of The Palit GeForce RTX 4060 Ti Dual
While you might not be in a hurry to use generative AI features that everyone is talking about lately, AI is still capable of doing things in the background that helps you work faster and play smoother in more ways than one. Here’s a few examples of what you can do with PALIT GeForce RTX 4060 Ti Dual:
Gaming
One of the best-known features from NVIDIA’s modern-day GPUs is the DLSS 3.5, which has seen many iterations over the years as the training data improves over time and more AI-powered features get integrated into this performance and quality-boosting feature.
DLSS 3.5 includes several things: Super Resolution (previously DLSS 2), Frame Generation (part of DLSS 3), and the new Ray Reconstruction feature. The Super Resolution feature upscales the content on screen to your display’s native resolution via the GPU’s Tensor Cores, while Frame Generation inserts generated frames in between to boost overall framerate (NVIDIA Reflex must be applied to compensate for increased latency).
Ray Reconstruction, however, is less of a performance-boosting feature and is more of a visual enhancement feature. To make ray tracing work, the game engine has to shoot a huge number of rays to the player’s view. This is an extremely computationally heavy task, so to ensure playable framerates, the total amount of rays is reduced. This, however, introduces noise – while existing denoisers can solve this, it incurs development time. With supercomputer-trained data, DLSS 3.5 leaves this task to the AI to do the heavy lifting, which means less work for developers, and more image quality for players.
NVIDIA is also finding use cases for AI beyond just graphics. The NVIDIA ACE (Avatar Cloud Engine) puts the AI into “AI” whereby the NPCs are completely generated on the fly with capabilities to hold conversations with players. From the developer’s perspective, you don’t have to record every line and movement – just give it some basics like character background and certain features, and players can interact with them naturally, giving more variety of how players get responses based on the choices they make.
Creating
It’s not just gamers that get the benefits, though. Creators also get their fair share of improvements with AI-capable RTX GPUs. Many professional apps today come with AI-assisted features that simplify the workflow: for example, in Adobe Lightroom and Photoshop, AI can be used to better upscale images compared to traditional methods that rely on spatial upscaling. In Autodesk 3DS Max, you get access to NVIDIA Omniverse AI that provides powerful physics simulation usable in digital twin applications.
Stable Diffusion is also another use case where it taps into your GPU to generate images. This is done on-device, meaning less reliance on cloud-based solutions and less chances for any sensitive information to be stolen. Generative AI can use quite a lot of memory to do its job, so the 16GB variant of the GeForce RTX 4060 Ti can be useful in models such as Stable Diffusion XL.
Developing
Of course, you can also use the GPU to run large language models – LLMs for short – without the need to use cloud services or pay for monthly fees, in some cases. Developers can build software with AI models built from NVIDIA AI Workbench, which allows scalability from home PCs all the way up to cloud. This can be further optimized via NVIDIA TensorRT, and in the case of LLMs, there’s TensorRT-LLM that improves performance for most language models, including Google Gemma, Meta Llama 2, Mistral, and Microsoft Phi-2.
Daily Usage
You most certainly have heard of the new Copilot+ PCs launched recently, which comes with AI capabilities among other things. Does that mean every system out there is not AI-capable? That’d be incorrect. As a matter of fact, NVIDIA has included Tensor Cores since the RTX 20 series GPUs, which means all GeForce RTX GPUs today have some levels of AI compute capabilities. That’s over 100 million of them, in case you’re wondering.
The AI can be used for features like background blur in video streaming & conferencing (via NVIDIA Broadcast), video upscaling through RTX Video, and even a local chatbot through ChatRTX that functions as your PC’s personal assistant, all without relying online services which sometimes can charge monthly fees for usage.
How Powerful Is It?
Your GPU can function as an AI compute engine like NPUs, and it’s significantly more powerful at it too. Currently, most laptops have the TOPS (trillion operations per second) figure under 50, while something like the Palit GeForce RTX 4060 Ti Dual packs 353 TOPS – that’s more than seven times the AI compute power than your typical laptop. Going up the ladder and you’ll find RTX 4090 packing 1,321 TOPS to crunch through even the most demanding AI workloads.
Conclusion
Of course, this is just scratching the surface on the kind of things AI can do with the help of GeForce RTX GPUs, and we’re still fairly early in the implementation of AI-assisted applications, so expect developers to come up with more ways to utilize the performance on tap offered by these GPUs. NVIDIA is certainly hedging its bets on AI for the foreseeable future, and it’s ready to capitalize on it to accelerate compute with more applications.
