Investigating NVIDIA’s Defective GPUs: RTX 5080 Missing ROPs Benchmarks
jimmy_thang
March 10, 2025
We take a look at an RTX 5080 with missing ROPs and benchmark its performance against a non-defective model
The Highlights
- The RTX 5080 is supposed to have 112 ROPs, but we purchased one with 104
- NVIDIA shipped some RTX 5090, RTX 5090D, RTX 5080, and RTX 5070 Ti GPUs with missing ROPs
- This is a major defect that can have a performance impact
Table of Contents
Grab a
GN15 All-Over Print Component Mouse Mat for a high-quality mousing surface that'll fit your keyboard & mouse. These mouse mats use a high-quality yellow rubber underside, a blue stitched border for fray resistance, and are covered in PC parts. This is the best way to support our work and keeps us
ad-free to support consumer-first reviews!
Intro
Today we’re looking at a deficient RTX 5080 Founders Edition (read our review) that’s missing ROPs, meaning it came straight from NVIDIA and that NVIDIA had complete and total end-to-end control over the entire process, just like they want, and they still somehow f***ed it up. The normal 5080 is as much as 10-11% better than this deficient one. Sometimes it’s 0%, often it’s 3-8%, but the swings can be large. This is a huge problem because most users won’t ever notice if their GPUs are missing ROPs. Unfortunately, the masses probably won’t even know about this problem to begin with.
Here’s what’s going on:
Editor's note: This was originally published on March 2, 2025 as a video. This content has been adapted to written format for this article and is unchanged from the original publication.
Credits
Test Lead, Editing, Host, Writing
Steve Burke
Writing, Web Editing
Jimmy Thang
We traded a functional Zotac 5080 (read our review) to our viewer, Mason, for his defective 5080. Thanks for the trade, Mason.
We have confirmed and validated that this GPU is missing 8 of its ROPs, down at 104 ROPs from the expected 112 ROPs. That means the proper card has nearly 8% more ROPs than the defect.
This is a huge problem. There are no obvious signs that this issue is present without knowing to look for it, which screws mainstream owners. There will absolutely be defective units out there without people knowing, and we don’t think NVIDIA has done enough to draw attention to this issue. You’d have to know to launch GPU-Z, then know to check the ROPs, then recognize that the count is wrong. That knowledge and skill will be way more common for this audience than most, but even then, most people aren’t going to feel a need to validate the GPU they bought has each individual fixed function unit present.
NVIDIA needs to do better about notifying customers. NVIDIA may claim 0.5%, but from first-hand experience in our inbox alone, we have a hard time believing the count is that low -- especially since it didn’t name the 5080. Either NVIDIA didn’t know about the 5080, in which case it’s wrong about the defect rate, or it did know and it was disingenuous at best by leaving it out. We’re not sure which is worse. We’ve received dozens of emails and messages about units deficient in ROPs count, which seems awfully high for a focused audience with seemingly low distribution of the card so far.
That’s the backstory. Let’s get into the testing.
A “ROP” is a raster operations pipeline (or render output unit) and is a core part of the GPU.
We already explained this in-depth in a video about this topic. Here’s the basics: The NVIDIA Blackwell architecture for gaming GPUs looks like this block diagram at most.
Each GPC has 8 TPCs, with two groups of 8 ROPs assigned separately to groupings of 4 TPCs. In some 5090, 5070 Ti, and now 5080s, one of these banks of 8 ROPs appears to be disabled, or at least not functioning.
As we said before, the weird thing is that this shouldn’t have been possible on the 5080. The 5080 should be a full GB203 die, and so there are no disabled SMs on a GB203 RTX 5080 unlike a 5090 or a 5070 Ti where some of the stuff is turned off and there could be collateral damage under traditional understanding.
Maybe some poor TSMC or NVIDIA employee knocked over a coffee mug and hit the KILLROPS.EXE key on the 5080 production run or something. Whatever the case is, it has an impact.
Over the years, we’ve seen GPUs with a ROPs advantage and iso other conditions typically show their benefit in higher resolution scenarios or in heavily anti-aliased testing. ROPs perform some of the final stages in the rendering pipeline. Of the 3 resolutions we test, 4K will show the biggest impact, but scenes with a lot of blending or some types of anti-aliasing will also reflect the change.
NVIDIA acknowledged this issue shortly after it emerged, but only for the 5070 Ti (read our review) and 5090 (read our review). It almost immediately somehow knew exactly how many units were affected, which further reinforces our belief that NVIDIA would have known about this, with the company pinning it to 0.5% of units. The company did not name the 5080 -- it also didn’t really apologize and we felt it downplayed the performance impact to the lowest number, which would be 4% on the 5090. It didn’t mention that the 5070 Ti and 5080 impact would be greater.
After Mason’s card showed up on Reddit, NVIDIA issued a second statement that we’re going to call their “oopsies” statement, where they confirmed the 5080 was also affected.
RTX 5080 Missing ROPs Benchmarks
Grab a
GN15 Large Anti-Static Modmat to celebrate our 15th Anniversary and for a high-quality PC building work surface. The Modmat features useful PC building diagrams and is anti-static conductive. Purchases directly fund our work! (or consider a
direct donation or a
Patreon contribution!)
Let’s get into some simple performance numbers. We only really need two basics:
1 - The impact to performance against the normal 5080, which can be done in a simple A/B chart
2 - The change in relative positioning versus nearby alternatives
We’re going to keep the charts really focused and simple because we don’t need much to show the evidence of performance impact.
Performance Recap: 4K Raster
This chart shows the head-to-head in average FPS for the two 5080 cards. Some games are almost exactly identical: Baldur’s Gate 3 predictably is CPU-bound, but it’s nice to know that a CPU-bound scenario didn’t force a gap. Black Myth: Wukong was remarkably consistent and Final Fantasy 14 was within 1 FPS for this testing, but there are some differences.
Total War: Warhammer 3 is the most concerning of these. This one has always rooted-out the most erratic behaviors in testing and that’s why we keep it around. Across all 3 resolutions, we saw major swings. At 4K, we observed an 11% improvement with the actual RTX 5080 rather than the deficient one. That is a difference as big as the gap between some of NVIDIA’s models entirely.
Dying Light 2 also consistently showed a gap: The full 5080 ran 8.7% higher framerate for average FPS than the deficient one. F1 24 showed a 3.3% improvement with all ROPs, with Resident Evil 4 at 1.6%, which is outside our run-to-run variance and makes it a real result, and Starfield at 2.3%.
Performance Recap: 1440p Raster
At 1440p, we saw an 8.8% improvement with all ROPs in Dying Light 2, which matches our 4K results. Final Fantasy is at about 2%, Dragon’s Dogma 2 is at 2.5%, and F1 24 is at 0.8%.
Comparative Charts
Let’s look at how this impacts the relative ranking versus other cards. We’ll look at only the games with the largest impact for a worst-case scenario.
Total War: Warhammer - 4K
Here’s Total Warhammer III result at 4K. The 5080 “Reduction of Performance” variant ran at 82 FPS AVG, a significant reduction from the correct result of 91 FPS AVG. Before, the 5080 was tied with the 7900 XTX and within error. Now, the 7900 XTX outperforms the 5080 by 12%. That is a huge swing and makes the 7900 XTX significantly better value. Sure, the partners might help you replace a defective model; however, that’d require noticing it.
The gap over the not-ROPs-deficient 5070 Ti is also reduced to nothing. This particular title and the way we test it is highly reactive to this defect.
Dragon’s Dogma 2 - 4K
In Dragon’s Dogma 2, the 5080 Special Edition landed between the stock model and the 7900 XTX, cutting the gap in half. This significantly harms the value of the RTX 5080. The lead is reduced from 10% to 5%. Literally halved. The lead over the 5070 Ti is also cut, now 9.4% from 15%.
Dragon’s Dogma 2 - 1080p
1080p shows the 5080 Regression of Performance edition at 157.1 FPS AVG from 165, which reduces it to equal the 4080 (watch our review). Before, they were functionally equal. Now, they’re literally equal. The 5080’s lead over the 5070 Ti was 9%. Now it’s 3.8%. It was cut into a third of the benefit, basically. The 7900 XTX now is nearly within run-to-run variance of the 5080 defect.
Dying Light 2 - 4K
We’ll just look at one more. In Dying Light 2 at 4K, the RTX 5080 normal card ran at 81 FPS AVG, with the 5080 ROPs defect card at 74.5 FPS AVG. The 5080 was 11.7% ahead of the 7900 XTX, but is now only 2.8% ahead.
Conclusion
Visit our
Patreon page to contribute a few dollars toward this website's operation (or consider a
direct donation or buying something from our
GN Store!) Additionally, when you purchase through links to retailers on our site, we may earn a small affiliate commission.
If you had bought the 5080 instead of the 7900 XTX because of the expectation and ended up with a defect, this is a big problem because now the ranking shuffles. If you don’t notice and you keep using the defective device, you’re going to get screwed. You’ll be stuck with something worse than you thought.
Checking for this is really easy. When you buy a 50 series card, the first thing you should do is check for all of the ROPs to make sure that your card is not affected.
You have to do a clean install of the drivers. If you check GPU-Z without the drivers installed, it will reference a look-up table and tell you the correct amount even if they’re not present. So you need to install the drivers first and then install the latest version of GPU-Z and look for ROPs.
Then look at the image above to see how many ROPs should be present.
If what you have differs from what it should be, you absolutely need to seek a refund or replacement, but we’d encourage a refund as it’s the fastest path.
There is absolutely a performance impact and NVIDIA’s approach of “just reach out and we’ll make it right” is completely unacceptable. They also left out the 5080. The company deserves to get raked over the coals for this. Most users will not notice this.
NVIDIA originally said the problem was with 1 ROP instead of 8 ROPs.
We did take apart the card and observed no physical difference to the die with the text looking the same. It also has the same branding.
This puts a cap on what has been an utter disaster of a launch for NVIDIA. One or two mistakes is understandable, but the totality of these mistakes is insane, especially for the prices they are going for.
Connexion
