New technological innovations:
A new generation automatically means a lot of technological innovations and this RTX 4060 Ti, like its big sisters, the RTX 4090, 4080, 4070 Ti and 4070, is no exception to the rule. We will try to mention the most important ones in this more theoretical page.
New Ada Lovelace architecture :
On September 21st at 5pm, NVIDIA unveiled part of its future range of graphics cards built around the “ADA Lovelace” architecture. Technically, this chip is out of the ordinary with its 76 billion transistors, its N4 engraving by TSMC, its Shader Execution Reordering, its 90 TFLOPS, its 200 RT TFLOPS, its 1400 Tensor TFLOPS and a maximum of 18432 CUDA Cores in a single GPU. That’s quite impressive compared to the previous generation Ampere.
The chip is organized into 12 Graphics Processing Clusters (GPCs) each containing 72 Texture Processing Clusters (TPCs) with 144 Streaming Multiprocessors (SMs). The Streaming Multiprocessors benefit from a new technology called Shader Execution Reordering (SER) which allows Ray Tracing operations to be multiplied by two. It is also worth noting that the RTX 4090 chip has been amputated, perhaps allowing for a future RTX 4090 Ti to be introduced.
The new RTXs also come with 4th generation Tensor Cores, which will be useful for DLSS and DLSS 3 AI, and 3rd generation RT Cores, which will double the performance of Ray Tracing. In addition to the generation evolution, the number of cores also increases compared to the previous generation depending on the model.
By the way, here is a comparison table to judge the differences between these cards.
DLSS 3 is coming and is already operational:
One of the announcements of NVIDIA during its conference is the arrival of DLSS 3. However, this is only compatible with the RTX 4000. Owners of a previous generation card will be happy with DLSS 2, which NVIDIA will continue to develop in the coming months.
As a reminder, until now, the principle of DLSS (Deep learning super sampling) was to allow a gain in terms of FPS by replacing the TAA antialiasing. The idea is to calculate the rendering in a lower definition than that chosen in the parameters and then to replay it in the said definition using an AI applied via the Tensor Core while keeping the most perfect image quality possible.
Today, with the arrival of DLSS 3, things are changing even though the fundamental principle of increasing FPS is still the main reason. What’s new is that DLSS 3 allows you to add whole intermediate frames instead of pixels. This increases the number of frames per second and consequently the fluidity. To achieve this, DLSS3 uses an optical flow accelerator integrated into the new RTX 4000 architecture. This is why previous generations that do not have this will not be able to benefit from DLSS 3. We will therefore have to be careful during our tests and compare what is comparable because yes, of course DLSS 3 will be more powerful but it is difficult to judge it on the Ampere generation which does not support it.
Even if the main function is to increase the number of FPS, NVIDIA also reminds us of the importance of latency. We had already dealt with this subject when we spoke to you about Reflex. To sum up simply, the principle is to reduce latency as much as possible in games, i.e. the time that elapses between the moment you click on your mouse button and its manifestation on the screen. In order to be able to carry out all the readings, a compatible mouse will be necessary.
NVIDIA’s latest developments could well turn the world of e-sports upside down as we could see 1440p 27″ displays replacing the current 1080p 25″ displays. Finally, NVIDIA has already announced over 35 games that are compatible and ready to use DLSS 3. We will of course be adding games that support DLSS 3 to our tests. Of course DLSS 3 games are backwards compatible with DLSS 2 technology.
AV1 hardware compression:
NVIDIA is now integrating its 8th generation NVENC encoder with AV1 which is 40% more efficient than H.264. The arrival of AV1 will be very useful especially for streaming.