The deadline for the launch of Ampere graphics cards is approaching and rumors are closing in on details and giving, in theory, a clear idea of what we will have in the market this year.
Before starting, it should be clear that what has been seen so far has been in the HPC branch, namely, data centers, Deep Learning, IA etc … What we will see next is the GPU gaming data, which is the most interesting part of the current market because it is the one used users.
TSMC vs. Samsung: NVIDIA will distribute chip production in all its different locations
The controversy over lithographic processes is back, but at the same time and because of rumors it seems that everything we see is being given a quote. We have already seen how the GA100 will reach 7 nm with amazing performance and features, where the TSMC would take care of its chips.
In contrast, NVIDIA's low-end GPUs outperform Samsung's 12nm due to its low hardness and chip size and because Koreans would be offering Huang's most competitive price per employer.
Well, new rumors say that these chips can be GA106 and GA107, but some chip won't get to those nanometers and that's where the surprise is. NVIDIA will require high performance and low density, but at a lower cost to compete with AMD and be more profitable at the same time.
Therefore, it is stated that GA102, GA103 and GA104
SLI limited to + 40% (RTX 3080 Ti)
The sources provide a lot of news, and that is what NVIDIA looks like to take the house out the window as AMD plans to do with RDNA 2. First, all Ampere graphics cards will definitely have RT Cores and definitely Tensor Cores. supporting RsHardware and DLSS Tracking.
SLI technology will only be enabled by the GA102-enabled cards, that is, the only high-end, which is not usually true of the company, has been slowly reducing its performance.
On the other hand, the introduction of the RTX 2080 Ti (thought to come as the RTX 3080 Ti) will have 40% more performance than its predecessor. The features of the chips are not really surprising, because despite the large number of SM, Shaders and memory installed, only the fastest model will be able to jump the best performance.
Based on (100%) RTX 2080 Ti, GA102 will achieve a 40% additional functionality, new and complex GA103 a 10% and GA104 will be left with a 5% below the reference at this time.
To translate it to, at the very least, the GA104, with 20 SM less will remain 5% of the TU102 with the RTX 2080 Ti. This means that the increase in the performance of the middle and upper grade may be a 15% – 25% more.
