There are a few key aspects that NVIDIA has been clear about when introducing new GPUs to the market and that AMD has begun to understand for over two years.
These things don't really work much more than the buildings themselves, but they have an interesting aesthetic that leaves a clear message that promises to change the look of the current graphics card.
Properties
We start from the premise that we understand GPUs as such, that is, they are great "big calculators" for FP functions and, therefore, are able to work in parallel. The majority of the computation is done by FPU units and unlike CPUs, these units are not programmed by software developers, but rather are both exciting and completely dependent on the driver supporting them.
This leaves AMD and NVIDIA to expand their products as few devices on the PC have. At the same time, this is just the beginning of the problem with the main issue of this article, and that is when NVIDIA allocated such a large number of resources when its development team was internally referred to as "NVIDIA Soldiers".
The number of software developers have is more than AMD and here's the peak of their GPUs. It should be noted that in architecture and early 2005, graphics cards have the same units to operate: ALUs / FPU TMUs and ROPs (with the exception of compatible addresses and VRAM specifications) and only Turing has installed new RT Cores . and Tensor Cores for different functions.
This also assumes that software and infrastructure development makes a huge difference in each generation if you don't make faster progress than your competitor. For specification, Navi as a build includes two blocks Shader engines the AMD is distinguished by its popularity Asynchronous Compute Engines (ACE), where each of them has 5 WGP and two CUs.
For pure comparisons, NVIDIA in Taring has 6 GPC with 6 TPC in each case two SMs for each block. This simple idea of each building's design allows us to see that Huang's similarity is much higher and more refined than AMD, which has very strong blocks together, but shows that at the same time less power is more effective than your partner's choice.
Finally, it should be noted that there are significant differences in the performance of both structures, arising in the past with similar evolution: NVIDIA operates on scale-killing units, with AMD in part using its neurotransmitter units.
What does this mean? It's a completely different access to work by developers and at the same time of its opacity, it's a wall that AMD is trying to break down to offer simpler programming units and better resources.
Use
The other problem is that AMD has had years, and that even the lithographic process is much more advanced than its predecessor, it does not manage to get ahead. Again, it is all a matter of design and construction.
NVIDIA is able to perform any group of TPCs and GPCs all in milliseconds, with greater workload variability and that by combining various technologies such as Color Prevention or High Levels makes their units more efficient and therefore able to increase performance by consuming less energy.
Efficiency is important and here NVIDIA with its unit performance has been able to outperform AMD in the same clock cycle. You shouldn't look at this too much from a performance point of view (which is obviously better) but from an application.
The scalar unit allows for one floating reading and one digit number at a time and for each clock cycle. The redesign of the built-in NVIDIA architecture allows the app to work with vector functionality in a much simpler way than AMD, especially now that Turing has three motors that are well-differentiated within each SM.
This allows rasterization to better focus on such engines, either INT32, FP32 or Tensor Cores, allowing, if not necessary, to make complete GPCs or other specified vehicles, saving use and efficiency in the work.
Prices
It's a deciding factor when talking about which GPU is "better". At NVIDIA, the high pricing strategy offers new technologies to work on this, but the fact is that both Ray Tracing and DLSS were never as big a step as they intended, and were never free from controversy or problems.
Offering a low-priced product doesn't make it any better, you have to know how to put it in an attractive way. The components of use and construction lead directly to it and make AMD seen as an expensive option for a large number of users.
Navi is surprising NVIDIA right now, with the technological advances being huge and the huge jump that has made Huang's introduction of a series of cards covering the gaps. But the reality worldwide is that the user views the technology, functionality and utility offered by NVIDIA at a higher price. It makes no sense that it holds more than 60 percent of the world market, so we are in the position of those who might choose to pay extra for the NVIDIA GPU to use its new technology and those who just don't want to pass on that radio for different reasons.
In any case, for over 15 years AMD has often been found in this category. NVIDIA puts prices on its new GPUs and AMD fills in the gaps with its GPUs.
