New storyline twist from what we saw less than a month ago. And it is that the configuration of the cores of the RTX-4090 Well, it won’t be the one we’ve discussed and seen, but NVIDIA seems to have moved towards a little more power in its final design. Because of this, performance was assured to be slightly higher than two RTX 3090s, an unheard-of jump in performance between series, especially considering that the total core count of the RTX-4090 Is equivalent to 16,384 shaders.
It is true that the current leaked configuration hardly differs from the previous one. In fact, although we discussed it, it was weird to leave a TPC hanging in SM, when the current setup is rounder and would be more what you expect.
NVIDIA RTX 4090, from 16,128 Shaders to 16,384, an increase over AMD?
The final specification may be 16384 instead of 16128. One small difference.
– kopite7kimi (@kopite7kimi) June 12, 2022
The reality is that the increase between the two figures by NVIDIA corresponds to a 1.58% GAP between them, showing that it may have been more of a leaking error than a last-minute change on Huang’s part. Although it cannot be ruled out that this increase is a means of balancing the performance balance between CPTsince, as we have already said, there were too many gaps in the GIC handicapped as such.
Logically, this move also minimally closes the gap for the eventual TITAN RTX 40, or the RTX 4090 Ti, and possibly even both, as the former might have the chip AD102-300 in its entirety, while the second could be an intermediate version with just over 17,000 cores.
Curiously, with these new RTX 4090s opens a debate in the series that has less to do with performance and more with consumption. And it is that any modification of the chips upsets NVIDIA, where the consumption of this GPU has been estimated at 450 watts of the 600W who has the full chip.
Can NVIDIA limit consumption?
We know nothing of the frequencies, which is curious, but logically if we knew them we would already have many indications on the real consumption and not on what is filtered as such. Comparatively speaking, it would be very interesting and we would draw conclusions, but we still don’t know.
The issue here is whether this minimal increase in Shaders affects consumption or frequencies as such. Why is this so relevant? Especially because he talks about 450 watts for the RTX 4090 with 16,384 shaders and 420 watts for the RTX 4080, but other rumors claim that the last of these values would be lower: 400 watts. So either NVIDIA increases the TGP for the RTX 4090, or those 130 watts or 150 watts represent a lower shader number for the RTX 4080.
Keep in mind that chips are different, as most of them use the AD102while the miner goes for the AD103 Nvidia thus segments the ranges to make the seconds more profitable and the first more expensive. In any case, there are still too many things in the air to draw definitive conclusions.