NVIDIA Ampere could be presented on March 23 and will arrive with great technical news
For several weeks now, there has been talk about NVIDIA Ampere graphics cards . This new graphics card architecture will replace the Turing GPUs currently on the market, manufactured using a 12nm process. Not much is known about Ampere, but there is talk of a 30-50% performance improvement thanks to the jump to 7nm. It is also rumored to be unveiled at GTC 2020, which will take place in March.
At the same time, AMD is talking about its Big Navi graphics. Not much is known about AMD's high-end solutions, but they indicate that it will revolutionize 4K gaming. It is also said that they will come to compete with the RTX 2080 Ti, although it may not end up being this way if NVIDIA surprises with something brutal.
[amazon box="B07MJGCPW5"]New details on the NVIDIA Ampere
The latest rumors regarding Ampere is that they may not replace Turing architecture graphics. Some rumors indicate that it could replace Volta, NVIDIA's architecture for the professional segment. But there is also talk of a presentation on March 23, shortly before the GTC that will be held on March 26. Although they will now be presented, these graphs will not reach the market until September.
Additionally, they are talking about a better segmentation of their silicones in order to offer an improved GPU structure. It seems that NVIDIA has no desire to return to the strategy of the Super to face the competition from AMD. According to rumors, we could go from five silicons to a total of eight to ten silicons for the RTX 3000 Series.
Another modification would be the big increase in SM and CUDA Cores. The most powerful silicon would be the GM100 with 128SM and a total of 8192 CUDA Cores, a truly spectacular leap. As a simpler solution we would have the GA107 with 20SM and 1280 CUDA Cores, which is not bad at all.
SM with some news
Along with the preliminary data, an image has been leaked of what the SMs of the new graphics could be like. The SMs will receive an update, although this is better than expected, a relevant power jump is expected. There are no changes in the number of units and it provides the same. The Warp Scheduler, Dispatch are also kept, and the log files and Clock per cycle are also kept.
The big change is in the FP32 units, which are doubled in order to keep the number of shaders at an optimal ratio. What this offers us is an increase in system performance. Something also relevant is that the Tensor Cores have been doubled, which will greatly improve the performance of DLSS.
We go to the L1 cache that goes from 96KB to 128KB, without giving more information about it. We do not know how much will be for data and which for shared memory, but it is said that it is improved to support more complex functions.
Of course, Ray Tracing has an important role with an improvement of the RT Cores. These new cores for ray tracing are more advanced, more efficient and faster for work under the BHV algorithm. The improvements have not been specified or if they have managed to make it stop being a hybrid process with shading.
Source: PCGN
