Seems like so far, as far as anybody knows, only the Zotac “Solid” cards are affected.
First paragraph reads:
TechPowerUp has discovered that there are NVIDIA GeForce RTX 5090 graphics cards in retail circulation that come with too few render units, which lowers performance. Zotac’s GeForce RTX 5090 Solid comes with fewer ROPs than it should—168 are enabled, instead of the 176 that are part of the RTX 5090 specifications. This loss of 8 ROPs has a small, but noticeable impact on performance. During recent testing, we noticed our Zotac RTX 5090 Solid sample underperformed slightly, falling behind even the NVIDIA RTX 5090 Founders Edition card. At the time we didn’t pay attention to the ROP count that TechPowerUp GPU-Z was reporting, and instead spent time looking for other reasons, like clocks, power, cooling, etc.
ROP?
Basically, the ROPs are in charge of the final stage of rendering a frame.
They are discrete, physical components of the GPU.
GPUs at this point are quite complex boards of many specialized kinds of processors passing information to and from each other.
Tensor cores, RT cores, CUDA cores, ROPs, etc… these are all specialized processors, specializing in different kinds of computations.
ROPs, Raster Output Processors, do varying kinds of postprocessing on the almost final stage of a rendered frame, assemble the data from other parts of the GPU and then push the finalized, rastered pixels to the frame buffer.
A rough analogy would that ROPs are in charge of the final editing pass on a paper or article before it’s published, with the analagous ‘research’, ‘fact verifying’, and ‘rough draft’ having already been done by other parts of the GPU first.
Maybe another analogy would be that ROPs are the ‘final assembly’ of a frame, if constructing a frame was like building a car or aircraft.
A simpler, more literal explanation is that the ROPs perform the final stage of rendering a frame before the GPU actually pushes it out for you to see.
So… if the GPU is missing 8 ROPs… the GPU is basically bottlenecking itself, internally.
Wow thank you for the explanation! That made it very clear.
Thank you for the explanation!
The article is terribly written, you need to scroll way down in the article to find out what ROP means, despite the article using the acronym several times
Ok, it’s Raster Operations Pipeline for anyone who doesn’t want to read that far.
Edit: In this context it is probably Raster Output Processor, as sp3ctr4l points out.
Technically, the Raster Operations Pipeline is the entire process of actually rendering a frame, or, contextually and depending on what precise terminology is being used by what company for which architecture, it may only refer to the final stages of actually rendering the frame.
Raster Operations Pipeline is the process, the systematized flow of different stages of rendering, the verb or action that the physical Raster Output Processors actually perform. EDIT: Or perform a part of, a stage of.
In this case, the Raster Operations Pipeline is hampered by the GPU missing 8 out of 176 Raster Output Processors.
Ie, its missing an amount of discrete physical components from the GPU board, and thus is less performant at actually rendering the Raster Operations Pipeline.
Nvidia is … pretty much very obviously at this point going out of its way to make the terminogy around its GPUs as confusing as possible, so that they can more easily do the equivalent of Star Trek esque technobabble to waive away and dismiss anyone who tries to actually dive in and understand what they’re actually doing.