The filtering by Geebench 5.0 leaves two new graphics cards in line with their functionality and where NVIDIA would obviously be able to jump at 7 nm, something we will discuss later.
These two cards look like they are prone to the server market due to their suspension in VRAM, at the same time surprising.
NVIDIA Ampere: New RAM configuration or HBM2E release?
Starting with the features of a high-power model, this can obviously come with the layout of 7552 Shaders included in 118 SM, continuing to throw 64 Shaders per SM it is similar to Turing and it is nth certification that NVIDIA will maintain its low-level architecture with the same parameters as today.
The speed instead would be far less than what we used to do with Huang, because he could only choose 1.11 GHz It shows the suite. Your VRAM may pop up 24 GB and benefit from Open CL points for 184096 points.
A little of the two models will also come with it 6912 shaders at 108 SM with a clock of 1 GHz and some surprises 47 GB VRAM total 141654 points under Open CL. This information serves as a gunpowder, until the estimates for the implementation of the theory are already on the table: 16.7 TFLOPS first and 13.9 second TFLOPS.
In addition, rumors about the increase in frequency are also evident, since there would not be too many changes in the architecture and the lithographic connection should be able to match at least the current density, which would cause performance.
In any case, big data is already a 40% faster than the current TITAN RTX offer.
New lithographic and VRAM process
After the controversy and deception of the interpretation of AMD's new graphics cards by SK Hynix, it's clear that there is no denying that VRAM stacks are up to 12 GB, at least for that company.
Therefore, 24 GB and 47 GB leave NVIDIA's second option: Samsung. If Geekbench learns well we are facing a clear step forward for a product with a low density stack but that will allow for more Huang processing on their GPUs.
If this happened it wasn't a stick hard enough for SK Hynix (and it is undeniable that they work with some models) TSMC is enough to say in a battle when it doesn't compete, but how?
Obviously, TSMC will continue to work with NVIDIA, but Samsung is once again making a piece of the cake. Obviously, the Koreans will also make low-end Ampere GPUs for gaming with Huang's low-quality lithographic process, but at the same time it will stand up to the jumps and widespread use.
This lithographic process is little known as 8LPP, which is one step behind the 7 nm TSMC and two steps after the first EUV procedures.
The advantage is that the Shaders NVIDIA low-cost chips can benefit from a small-size lithographic process depending on the capacity and performance per chip, because they are the most mature and efficient monthly. on active chips.
So and in short, what we saw today with Ampere represents its Tesla GPU at 7 nm, so we could see chips with a lower Shaders count, but higher desktop speeds, which is what really matters for gaming.
