Another case of missing ROP was observed and confirmed, but this time it is GeForce RTX 5080, which confirms our fear about a comprehensive problem.
Despite the non-exposure by NVIDIA, Geforce RTX 5080 is affected by missing rops
We have another confirmation case of the missing ROP (Rekhapunj Operations Pipeline), which is the fifth one in a row in two days if we consider such reported cases, including RTX 5090 and RTX 5070 TI GPU. It seems that more users verify their card glasses, the more they know that they have a Nerfed GPU.
This is the first reported case of GeForce RTX 5080 GPU, which was launched on 30 January with Geforce RTX 5090. This is the second fastest GPU in the RTX 50 series and is officially priced at the base model of the founding version and other vendors. However, a heavy price tag cannot save it from shedding some rope, as we have a user on Reddit who is spotting it on the GPU-Z program.

User U/Gingerf 90 posted a screenshot of its GPU-Z window, revealing that its RTX 5080 is only facilities 104 ROPS Instead of specified 112 ROP. Glasses are valid, which proves that RTX 5080 is present with the missing ROP. This indicates that Nvidia tried to hide that RTX 5080 could be affected by this problem.

It recently issued a statement that only RTX 5090/5090D and RTX 5070 TI are affected by this problem and the total number of affected GPU is less than 0.5% of the total units produced. Even though this is just a case of missing rops on Geforce RTX 5080, it is likely that we will face something else in the coming days.
If you do not know, Rops play an important role in razoring performance and directly affect the pixel filler. This can result in noticeable performance regression, and the end results will be inferior to what you expect. Nvidia says that the performance difference will be around 4% on an average, which is still to be furious about something, given that users are paying the entire price.
At this point, it needs to be investigated, given that the issue is growing, and over time, this general consumer can sign some more problematic for the GPU industry.
News Source: Reddit