[Hardware] New NVIDIA and AMD graphics would break speed records

[FNX Magokiler]

Rumors and more rumors about what both AMD and NVIDIA will bring to the table. And it is that the new generation of graphics cards will arrive with great news compared to their predecessors, starting with two new lithographic nodes that apparently will push the frequencies of these graphics to record speed levels and totally as standard. What values are filtering through the Internet about this?

We actually know little to nothing about the new TSMC exclusive node for NVIDIA called 4N (not to be confused with the N4 that we always name) but it seems to be something really interesting. The problem of the frequency is linked to this as to another decisive factor: the number of cores that a graphics card has and here the scales are quite uneven according to the latest rumors.

NVIDIA would take the throne of performance: record speed graphics
Well, according to the latest leak from a well-known leaker such as Kopite7kimi, NVIDIA could give a more than interesting blow to this debate that has been with us for almost two and a half years. As we well know, Ampere has not meant a big frequency change in NVIDIA, while RDNA 2 has in AMD and hence part of its success in terms of performance.

But everything will change and the gap will apparently close. The filtering indicates something very interesting, and it is nothing more than the fact that the AD102 chip, a benchmark in the gaming sector, could reach 2.8 GHz completely as standard with its 18,432 Shaders. But of course, this chip would be used for the supposed RTX 4090 Ti that could also arrive as a TITAN if necessary, and that from the outset and as we know, will not go on the market because NVIDIA keeps that ace up its sleeve.

 

 

What we will see is an AD102-300 cut to 16,128 Shaders that due to the cut it suffers in its SM will consume 450 watts and possibly maintain that figure of 2.8 GHz, which will also mean a record in the Huang graphics card segment serial speed.

AMD would disappoint in "almost" all aspects
TSMC's 5nm for Lisa Su will represent a more than notable jump in performance compared to the current 7nm and 6nm. It is speculated with a figure well below the expected in terms of total cores for Navi 31: 12,288 Shaders. Apparently, AMD is going for a monolithic chip with 6 MCDs surrounding it with an interposer made by TSMC with its 65nm CoWoS technology, so it wouldn't be the kind of MCM architecture we expected, but at the same time it's revolutionary because it leaves out of the main chip important parts of it, such as the L3.

The problem in terms of frequencies is that AMD would not have taken such a big step as NVIDIA, where rumors point to 3 GHz as standard for the alleged RX 7900 XT. Therefore, and except for last minute surprises in architecture, AMD starts with too great a disadvantage in Shaders to compensate for those supposed 200 MHz more than it could achieve. Curiously, nothing has been said about the consumption of this Navi 31, nor do we know anything about the price.

What it does seem is that AMD is betting on a model that is more complex in terms of elements of its architecture, but cheaper to manufacture in general terms and, above all, easier to dissipate because it has individual units where the heat output can be separated. to the sink. What we can be sure of is that it will be very interesting to see the two bets officially on the table, because in any case these graphs will break speed records once again.