News Nvidia (initially Zotac) is shipping GeForce RTX 5090 with missing ROPs, up to 4.5% lower performance

Kaching999

Kaching999

Galvanizer
Zotac is at it again with their abhorrent QC. Member of the TechPowerUP forums has discovered that a ZOTAC RTX 5090 graphics card has missing ROPs (Raster Operating Units or Raster Operating Pipelines). The ZOTAC SOLID card, which should have 176 ROPs, is reportedly shipping with only 168 ROPs enabled.

In the original review, TechPowerUP noticed that the ZOTAC 5090 SOLID card underperformed slightly and was falling behind the RTX 5090 Founders Edition. They admitted they didn’t initially notice that their card had the same issue, and only after a discovery by TechPowerUP forum member “Wuxi Gamer” did they confirm that their card had the same problem with missing ROPs.

In the first test, “Elden Ring” at 4K UHD with maxed-out settings and native resolution (no DLSS), the Zotac RTX 5090 Solid falls behind every other RTX 5090 tested, including the NVIDIA Founders Edition, which serves as a reference design establishing a performance baseline for the RTX 5090. The Zotac card is 5.6% slower than the Founders Edition and 8.4% slower than the ASUS ROG Astral RTX 5090 OC, the fastest custom design card for this test. Officially, the Solid is clocked at a 2407 MHz rated boost frequency, matching the Founders Edition clocks—it shouldn’t be significantly slower in real-life scenarios. The interesting thing is that the loss of performance is not visible when monitoring the clock frequencies, as they are as high as expected—there are just fewer units available to handle the rendering workload.
Click to expand...

1740140380081.png
1740143206657.png

TechPowerUP has confirmed that only the ZOTAC SOLID RTX 5090 has been seen with this error; none of the other samples, including those from ASUS, Gigabyte, MSI, Palit, and NVIDIA, have shown the same problem. Perhaps this is just a simple BIOS mishap, or perhaps it only affects initial cards, but the damage has already been done. Even if this is fixable with a new BIOS, it will still require a BIOS update on a $1999+ graphics card, good luck with that.

ZOTAC has not issued a statement on the issue, but TechPowerUP will work with ZOTAC and NVIDIA to send the card back and find a solution. It is very likely that these ROPs can be re-enabled, but if it is not possible, then there is a major problem.

Computerbase has confirmed that their unit has no such issues.

https://twitter.com/x/status/1892898298106618186

Source: TECHPOWERUP VIDEOCARDZ
 
Last edited:
  • Wow
  • Like
  • Haha
Reactions: Rezep, Crazy_Eddy, Tracer_Bullet and 3 others
Seems like the root cause is actually some defective GB202 chips.

1740201924305.jpeg


This means that anyone with this particular model is stuck with the lower performance until and unless Zotac or Nvidia decide to offer replacements. Imagine paying 3.2L which is over 1L over the supposed MSRP and getting this trash.

Source: WCCFTECH
 
  • Like
  • Sad
Reactions: PunkX 75 and MostlyHarmless

How to Extend the Longevity of Your Graphics Card​

Choosing the right graphics card from the start is crucial. Avoid cheaper variants of your selected model, as they often come with compromises in build quality and cooling.

Key Factors:​

  • Larger PCB and Surface Area
    A bigger PCB provides more space for heat to spread and dissipate, apart from the heat sink, helping to extend the lifespan of the card and its components.
  • More Components for Load Sharing
    In most cases, having more active and passive components is better because they share the electrical and thermal load. Manufacturers often reduce component count to cut costs, which increases stress on individual parts, leading to higher temperatures and shorter lifespan.
  • Avoid Overclocking
    GPUs are binned, meaning even the same model with same level of overclock can have different thermal and stability limits. Overclocking can push components beyond their designed tolerances, increasing heat and wear.
  • Stable Power Supply with Good Regulation
    A clean and stable power supply reduces the stress on the card's filtering components, minimizing heat generation. Electrolytic capacitors, in particular, are prone to failure over time due to charge and discharge cycles.
  • Temperature Management
    The primary goal is to lower peak temperatures. Every degree of reduction in sustained workloads contributes to a longer lifespan for your GPU.
 
  • Like
Reactions: Uvula, Tracer_Bullet, DigitalDude and 1 other person
Kaching999 said:
Seems like the root cause is actually some defective GB202 chips.

This means that anyone with this particular model is stuck with the lower performance until and unless Zotac or Nvidia decide to offer replacements. Imagine paying 3.2L which is over 1L over the supposed MSRP and getting this trash.
Click to expand...
It's not just Zotac. More partners cards have surfaced with this specific issue. nVidia have said they will honour warranty and users with this defect should contact manufacturer for replacements, issue being the ability of the manufacturer to provide a replacement given the paucity of chips and cards.

The reason it surfaced more in Zotac cards than others is simply because Zotac was the only brand with decent stock (in the US) at launch. Not all users will know they have this issue, 80% of users don't even check tech news and can be saddled with this until their 12HPWR cable melts.
 
  • Like
Reactions: Kaching999
Heisen said:

How to Extend the Longevity of Your Graphics Card​

Choosing the right graphics card from the start is crucial. Avoid cheaper variants of your selected model, as they often come with compromises in build quality and cooling.

Key Factors:​

  • Larger PCB and Surface Area
    A bigger PCB provides more space for heat to spread and dissipate, apart from the heat sink, helping to extend the lifespan of the card and its components.
  • More Components for Load Sharing
    In most cases, having more active and passive components is better because they share the electrical and thermal load. Manufacturers often reduce component count to cut costs, which increases stress on individual parts, leading to higher temperatures and shorter lifespan.
  • Avoid Overclocking
    GPUs are binned, meaning even the same model with same level of overclock can have different thermal and stability limits. Overclocking can push components beyond their designed tolerances, increasing heat and wear.
  • Stable Power Supply with Good Regulation
    A clean and stable power supply reduces the stress on the card's filtering components, minimizing heat generation. Electrolytic capacitors, in particular, are prone to failure over time due to charge and discharge cycles.
  • Temperature Management
    The primary goal is to lower peak temperatures. Every degree of reduction in sustained workloads contributes to a longer lifespan for your GPU.
Click to expand...
gpu reviews should include quality of internals too + checking for red flags. Noobies like me wont understand much otherwise.
Only now people are realizing that 4090/5090 and to smaller extent 4080(5080?) are burning up because nvidia cheapened out and removed power balancing.
This was maybe hard to find, but for other stuff - anybody who reviews this ?
 
Power balancing is not possible on a 500W + power draw off of a single cable. Each wire+connector on the cable is too close to the limit, so where is that excess power gonna go? None of the lines have a 'reserve' capacity to handle additional load. ROG Astrals have 6 separate sensors but that's all they do, sense and warn the user so you can stop loading and shut it off. The problem is the connector design, it's flimsy, weak and pathetic, like the child that best advertises contraceptives.

The derBaeur case (and the Reddit one that preceded it) was caused by poor contact on some of the receptacle sets. If you have both ends basically paralleled and still have uneven current draw, it's the connector at one or both ends. The 12HPWR has a max insertion number of 30, and after 30 it doubles from 15 to 30mOhm contact resistance. You can expect a YT influencer to use a cable more than 30 times. The Reddit user was reusing a cable from a 4090, and that was no slouch when it came to burning cables with excess power draw.

It is possible to get 50A in that connector profile, but not with those pins. It needs spade receptacles that can carry over 60A each at not much bigger size (and smaller than the EPS/PCIe connectors). Or a 24V supply standard, which would be better and remove issues with the 12HPWR because current draw is essentially halved.
 
  • Like
Reactions: Tracer_Bullet and Kaching999
cranky said:
Power balancing is not possible on a 500W + power draw off of a single cable. Each wire+connector on the cable is too close to the limit, so where is that excess power gonna go? None of the lines have a 'reserve' capacity to handle additional load. ROG Astrals have 6 separate sensors but that's all they do, sense and warn the user so you can stop loading and shut it off. The problem is the connector design, it's flimsy, weak and pathetic, like the child that best advertises contraceptives.

The derBaeur case (and the Reddit one that preceded it) was caused by poor contact on some of the receptacle sets. If you have both ends basically paralleled and still have uneven current draw, it's the connector at one or both ends. The 12HPWR has a max insertion number of 30, and after 30 it doubles from 15 to 30mOhm contact resistance. You can expect a YT influencer to use a cable more than 30 times. The Reddit user was reusing a cable from a 4090, and that was no slouch when it came to burning cables with excess power draw.

It is possible to get 50A in that connector profile, but not with those pins. It needs spade receptacles that can carry over 60A each at not much bigger size (and smaller than the EPS/PCIe connectors). Or a 24V supply standard, which would be better and remove issues with the 12HPWR because current draw is essentially halved.
Click to expand...
Yeah I mean imagine trying to power something that can draw as much power as something that could be used to heat an decent sized room, like literally with 50A passing through each connector, it was kind of destined for failure form the start but it has been accelerated due to the such high power usage levels. It's really confusing as to why they would stick to a single connector when it had multiple issues in the past and not just opt for multiple connectors of try for a 24V standard as you suggested which would be the best, maybe reserve it for the high power draw halo cards, cutting the current draw in half would in fact greatly reduce the melting risk.
 
To sum up the 5090:
1. Intentional artificial scarcity
2. Melting connectors (AGAIN)
3. Driver bricks
4. Missing rops due to bad QC control (which they knew about in all likelihood, just kept shut hoping nobody would catch them)
5. Fires

They need lawsuits to hit them, at this stage. ASUS was severely flamed for months due to their warranty policy scandal, not too long ago. What Nvidia is doing is WAY worse and getting away with it. At this stage, Nvidia is really pushing their luck with all this BS.
 
Last edited:
  • Like
Reactions: Tracer_Bullet
PunkX 75 said:
To sum up the 5090:
1. Intention artificial scarcity
2. Melting connectors (AGAIN)
3. Driver bricks
4. Missing rops due to bad QC control (which they knew about in all likelihood, just kept shut hoping nobody would catch them).

They need lawsuits to hit them, at this stage. ASUS was severely flamed for months due to their warranty policy scandal, not too long ago. At this stage, Nvidia is really pushing their luck with all this BS.
Click to expand...
Add the fires to that.


Nvidia has confirmed that 0.5% of RTX 5090 and 5070 Ti were shipped with fewer ROPs

NVIDIA has just issued a statement on something that we reported yesterday. Only after a report from TechPowerUP, who discovered an anomaly in product specs, did the company decide to confirm that there was indeed a production issue.

We have identified a rare issue affecting less than 0.5% (half a percent) of GeForce RTX 5090 / 5090D and 5070 Ti GPUs which have one fewer ROP than specified. The average graphical performance impact is 4%, with no impact on AI and Compute workloads. Affected consumers can contact the board manufacturer for a replacement. The production anomaly has been corrected.

— NVIDIA
Click to expand...
The statement confirms that there are also RTX 5070 Ti cards affected by this problem and not just the 5090s. The 0.5% number seems very accurate to report, so perhaps NVIDIA knew about this for some time. According to NVIDIA, the defect will cause 4% lower performance, also confirmed by TPU yesterday, which is something completely unacceptable on graphics cards this expensive.
The statement doesn’t provide a clear path for affected users on how to proceed. They are left without much help here and are asked to seek a replacement from board partners and in the case of the 5090 series, a replacement may be extremely hard to get.

Source: Videocardz
 
Last edited:
  • Haha
Reactions: PunkX 75
Kaching999 said:
Add the fires to that.


Nvidia has confirmed that 0.5% of RTX 5090 and 5070 Ti were shipped with fewer ROPs

NVIDIA has just issued a statement on something that we reported yesterday. Only after a report from TechPowerUP, who discovered an anomaly in product specs, did the company decide to confirm that there was indeed a production issue.


The statement confirms that there are also RTX 5070 Ti cards affected by this problem and not just the 5090s. The 0.5% number seems very accurate to report, so perhaps NVIDIA knew about this for some time. According to NVIDIA, the defect will cause 4% lower performance, also confirmed by TPU yesterday, which is something completely unacceptable on graphics cards this expensive.
The statement doesn’t provide a clear path for affected users on how to proceed. They are left without much help here and are asked to seek a replacement from board partners and in the case of the 5090 series, a replacement may be extremely hard to get.

Source: Videocardz
Click to expand...
LOL, what? What are they smoking?
0.5% of the 10 cards they actually made and shipped out.

Does not affect AI workloads? Yeah, that consoles gamers. Save that statement for AI customers, who already got their stock in bulk.

4% performance loss? More like 8-10%.

The production "anomaly" was corrected pretty fast. Seems like they indeed knew about it beforehand and just kept shut until they were caught.

What are they going to replace it with lol? They don't even have any cards to sell for $3000, let alone replace one for free.

Nobody would give a shit if this were an entry-mid-tier level GPU. When you've already been scammed by buying the thing for the price of a new car, only to be treated like dogshit with bad QC, doesn't bode well.
 
Last edited:
PunkX 75 said:
To sum up the 5090:
1. Intention artificial scarcity
2. Melting connectors (AGAIN)
3. Driver bricks
4. Missing rops due to bad QC control (which they knew about in all likelihood, just kept shut hoping nobody would catch them).

They need lawsuits to hit them, at this stage. ASUS was severely flamed for months due to their warranty policy scandal, not too long ago. What Nvidia is doing is WAY worse and getting away with it. At this stage, Nvidia is really pushing their luck with all this BS.
Click to expand...
Only positive thing out of this gen so far seems to be DLSS4 upscaling.
 
  • Like
Reactions: Kaching999 and PunkX 75
Top Bottom