© ROOT-NATION.com - Use of content is permitted with a backlink.
One of the most critical epithets about the NVIDIA GeForce RTX Blackwell generation is the phrase, that the company has shifted its focus from gamers – to artificial intelligence. On the one hand, this is true, because the growth rate of NVIDIA GPU performance in TOPS exceeds anything NVIDIA could show in raw raster performance from generation to generation. On the other hand, using the example of the Asus PRIME GeForce RTX 5070 Ti OC, I can tell you why the focus on business, artificial intelligence, and media rendering is actually just as important.
Read also: How useful the RTX 5070 Ti REALLY is?
We thank Telemart.ua for providing us with a video card ASUS PRIME GeForce RTX 5070 Ti OC
To explain how exactly important this is and how useful it really is for everyone – editors, gamers, and AI architects – I will structure this material in three directions.
TABLE OF CONTENTS:
CUDA
Let’s start with what actually kicked off the use of graphics cards for more than just rendering 3D models in Quake 3 Arena. Let’s start… with CUDA cores. The term stands for Compute Unified Device Architecture. This technology appeared in NVIDIA in 2006, with the 8800 GTX and 8800 GTS series graphics cards.
The invention of CUDA can only be compared to the establishment of an international system of measurement. No more arshins, shapels, and megalithic yards – only metres. This allowed the world to industrialise and unify tools, so a ruler purchased in Ukraine will be just as useful and precise in Australia.
The same happened with CUDA. Standardised micro-cores, similar to x86 processor ones, allowed – and still allow – parallel computing on multiple threads simultaneously. At the time, this was absolutely revolutionary, because the first dual-core processors – Intel Pentium Extreme Edition 840 and AMD Athlon 64 X2 – had been released just a year earlier, and optimisation for multiple threads simultaneously would appear MUCH later.
With CUDA, everything is simpler. Not because it is advantageous to distribute game rendering across multiple threads simultaneously, but because the unification of the standard made it possible to use video cards for other tasks. Initially, CUDA was used for weather forecasting. Then, it was used to process huge amounts of scientific data.
The real boom in CUDA usage began when software manufacturers started using these cores for video encoding and decoding. And of course, all these tasks can be done by a CPU. For example, on modern AMD Epyc processors, which have 100+ cores, games can be rendered exclusively on processor cores. But the efficiency of such rendering will be non-existent, because games are optimised for the parallel use of tens of thousands of low-power cores, and a few hundred, even powerful ones, will not improve the situation, only make it worse.
In this regard, almost the entire computer world today is tied to CUDA. Want to play a match in Counter-Strike 2? This is done by CUDA, because they perform the basic raster rendering of 3D engines. Want to apply several effects to a BRAW video file in DaVinci Resolve? This is done by CUDA. Want to render a three-dimensional scene in Blender or Cinema4D? CUDA cores will help you.
Not only the number of these cores is important, but also the generation, because the more recent it is, the better the cores work. And the newer the generation, the smaller these cores can be. The GeForce RTX 5070 Ti, for example, has 8960 CUDA cores, while the RTX 3090 has 10496. But in terms of performance, the RTX 5070 Ti completely destroys the previous flagship, in some cases by 50% or more, even in raster performance. Even before we mention DLSS 4 with Multi Frame Generation, which I described at length in my recent article, that you can find here.
Media engine
Of course, using CUDA is not always hyper-efficient. For simple video decoding – like playing ready-made files without any processing – CUDA cores are too powerful and consume too much energy. For this purpose, the RTX 5070 Ti, for example, has a separate media engine that takes on this task and consumes a minimal amount of power. This is extremely important for laptops, but also useful for PCs.
The Blackwell architecture not only reduces the power consumption of video cards to 10-15 W when watching videos, but also has hardware support for YUV 4:2:2, which was not available in previous generations. H264 and H265 encoding and decoding is now 50% faster. This is due to the fact that the new generation has three encoders and two decoders.
This means that if you are just watching videos on YouTube, your RTX 5070 Ti will not consume 100+ W, as video cards did a few generations ago. And even more importantly, it will not glitch like video decoding on processor cores (in case the processor does not have its own encode/decode video engine).
A lot of media creative powers can be unleashed through NVIDIA App, where streamers and reviewers can get the footage from games, blur the background with object separation, and reduce audio noise. No mandatory login into NVIDIA profile needed! Also, Studio Drivers will be much more useful to creators, due to increased stability with apps like Blender, DaVinci Resolve and Cinema4D. When new version comes – be sure to check new driver updates!
Tensor cores
Where else do CUDA cores not perform as efficiently as they could? In artificial intelligence processing. Yes, some AI models run on CUDA cores, such as LLM in StoryToolkitAI for creating subtitles. But more serious plugins, programs, and entire companies rely on tensor cores.
These are the same cores that are responsible for DLSS, because DLSS does not run on CUDA. This means that they are also very important for gamers, but their main task is to accelerate artificial intelligence in hardware.
However, you may be misled by the usefulness of tensor cores, as the vast majority of programmes use them as an auxiliary tool. Topaz Video AI Pro connects them for general video processing, DaVinci Resolve has several plugins, including slow motion using the Optical Flow/Speed Warp. There, without tensor cores or analogues, plugin simply refuses to work. Tensor cores speed up work in Blender, and they are the basis for work in generative AI models.
How are they similar to CUDA cores? The more there are, the better. The newer they are, the better. The RTX 5070 Ti has a total of 280 5th generation tensor cores, while the RTX 4070 Ti, for example, has 240 of these cores, but they are 4th generation. As a result, the RTX 4070 Ti delivers up to 640 TOPS, while the 5070 Ti delivers as much as 1400.
Also – one of the most important results of streamlining and optimising with tensor cores, is undoubtedly the Blueprints and NIM Microservices. Latter ones are the curated list of the best AI modules to use inside Chat RTX or any similar app, like ComfyUI. You can have LLM, media generator or animation, all in one place and with peak optimisation.
What are the Blueprints? They are all-in-one instruments to build your own AI instrument. They consist of basic functions, UI to control, reference code and architecture visualization. The simplification of development process probably will lead to an exponential boom in AI variety and spread.
Conclusions on the productive use of the NVIDIA GeForce RTX 5070 Ti
The main thing you need to understand is that it is important not to focus on where the RTX 5070 Ti will be useful for non-gamers. Because one aspect or another of this graphics card will be extremely useful in almost all scenarios for using PCs and laptops running Windows. Even if you are just printing text, by connecting an offline grammar checker, sooner or later it will be optimised either for CUDA or for tensor cores.
Therefore, the correct question would be the following. HOW beneficial is it, compared to the previous generation or competitors, to use the Asus PRIME GeForce RTX 5070 Ti OC for tasks not related to gaming? The answer is simple. Very, very beneficial!