Capacitor issues are causing RTX 3080/3090 crashes

[midian182]

Facepalm: The RTX 30-series hasn't had the smoothest of launches, from limited stock to bots snapping up what few cards were out there, but at least they don't have any hardware issues—do they? In the case of factory-overclocked aftermarket products, you might find your RTX 3080 and 3090 suffering from crashes and stability problems.

Follow up: Get your driver fix: graphics card makers respond to RTX 30-series capacitor controversy

The issue appears related to the type of capacitors found on the rear of the PCB underneath the GPU. As explained by Igor's Lab, AIB partners are supposed to use two types of capacitors for filtering voltages: large-area POSCAPs (Conductive Polymer Tantalum Solid Capacitors) or the smaller and cheaper MLCCs (Multilayer Ceramic Chip Capacitor). As per Nvidia's specifications, manufacturers' designs can use all-POSCAP, all-MLCC, or a combination of both.

MLCCs are smaller, cheaper, and have high current, voltage, and temperature ratings. They also operate better at higher frequencies. But they're prone to cracking and piezo effects and have poor temperature characteristics. The more expensive POSCAPs, meanwhile, have lower voltage ratings and aren't as good at high frequencies, but are not prone to cracking or the piezo effect, and operate better at high temperatures.

Igors Lab images of Nvidia PG132 PCB, Zotac Trinity, and Asus TUF. POSCAP are colored red, MLCC are green

The problems are occurring when certain cards that feature all-POSCAP designs reach/exceed 2.0 GHz. Nvidia's Founders Edition 3080/3090s, which haven't seen many reports of issues, don't feature as many POSCAPs. While unconfirmed, there are claims that manually downclocking the impacted cards by around 100 MHz can prevent the crashes.

EVGA has confirmed that the problems are related to the capacitors. "During our mass production QC testing we discovered a full 6 POSCAPs solution cannot pass the real world applications testing. It took almost a week of R&D effort to find the cause and reduce the POSCAPs to 4 and add 20 MLCC caps prior to shipping production boards, this is why the EVGA GeForce RTX 3080 FTW3 series was delayed at launch. There were no 6 POSCAP production EVGA GeForce RTX 3080 FTW3 boards shipped," writes the company.

"But, due to the time crunch, some of the reviewers were sent a pre-production version with 6 POSCAP's, we are working with those reviewers directly to replace their boards with production versions. EVGA GeForce RTX 3080 XC3 series with 5 POSCAPs + 10 MLCC solution is matched with the XC3 spec without issues."

Interestingly, Buildzoid claims that none of the RTX 30-series use POSCAPs. These capacitors are called SP-Caps.

Nvidia subreddit monitor Nestledrink has compiled a list of manufacturers and cards, noting which use MLCCs, POSCAPs, or a mixture of both. It's important to remember, though, that just because a card uses all POSCAPs doesn't automatically mean it will have issues.

**EVGA FTW3 - Early config with 6 POSCAPs is confirmed to be pre-production board per EVGA Jacob

Permalink to story.

https://www.techspot.com/news/86900-capacitor-issues-causing-rtx-30803090-crashes.html

 

[donnieD]

When you let bean counters interfere with engineering....

 

[Alfatawi Mendel]

Given that there are reports of the crashes happening on Nvidia FE cards too - A bad/faulty reference board design, is more likely than the choice of capacitors.
@Igors Lab and @Jay2cents have been blowing smoke about this, without knowing the REAL problem and as such Nvidia have already washed their hands of the problem.
Now imagine the hulabaloo if this was AMD, given the stramash about poor drivers on the 5x00 series of cards.

 

[QuantumPhysics]

Does this affect the 3090 kingpin addition?

I have the 3080 EVGA XC3 and a 3090 FE coming. I was going to sell the FE to someone and use the extra money to buy a Kingpin because I want the AIO built in.

 

[Endymio]

Given that there are reports of the crashes happening on Nvidia FE cards too - A bad/faulty reference board design, is more likely than the choice of capacitors.

Well, EVGA has already confirmed the capacitor issue. It's not a binary choice either: there could be multiple problems. It's also a possibility that those few FE crashes were related to PSU or other issues, rather than with the card itself.

 

[Uncle Al]

Well, they can use the Microsoft approach and claim it's an "undocumented feature" but I'm more inclined to just rename the product "Crash & Burn" and go with that ....... LOL

 

[Theinsanegamer]

Given that there are reports of the crashes happening on Nvidia FE cards too - A bad/faulty reference board design, is more likely than the choice of capacitors.
@Igors Lab and @Jay2cents have been blowing smoke about this, without knowing the REAL problem and as such Nvidia have already washed their hands of the problem.
Now imagine the hulabaloo if this was AMD, given the stramash about poor drivers on the 5x00 series of cards.

Nvidia FE boards also use a single MLCC capacitor. If the asus boards with 6 MLCC cpas dont have any issues then that would indicate that even nvidia's FE design is subpar.

 

[Maxiking]

Why not to actually contact someone from the industry with the knowledge or professional instead of a youtube kiddo who is wrong half of the time. Oh, that wouldn't be free.

 

[Araeven]

"MLCCs are smaller, cheaper" & "The more expensive POSCAPs"
Those statements are reversed. If you check the igors lab article it specifically says:
"somewhat more expensive MLCCs " at the end of paragraph 5.

 

[Irata]

Maybe people are just holding it wrong ? Has happened before....

 

[grumblguts]

Well samsung 8nm being such a greedy silicon has caught them out .

 

[Lionvibez]

If this can happen to nv it can happen to anyone for the brand loyalist out there.

I'm hoping the extra delay in the RDNA2 launch is to prevent mishaps like this.

 

[madboyv1]

"MLCCs are smaller, cheaper" & "The more expensive POSCAPs"
Those statements are reversed. If you check the igors lab article it specifically says:
"somewhat more expensive MLCCs " at the end of paragraph 5.

I was gonna mention the same thing. From how I've seen it explained, MLCCs are slightly more expensive, but you're putting 10 of them in the same space as one POSCAP. It complicates the tooling for the card a little bit too, again, because your doing almost 10x the "work" to put them on the PCB compared to a single POSCAP.

 

[Irata]

If this can happen to nv it can happen to anyone for the brand loyalist out there.

I'm hoping the extra delay in the RDNA2 launch is to prevent mishaps like this.

I would prefer them to not rush things and have a proper launch, as well.

 

[Avro Arrow]

I'm really confused by all this. Here, nVidia has the awesome RTX 3080 (and it really is awesome) but it's very obvious that they rushed the launch so much that problems were inevitable. Why didn't they just take another month to make sure that they had all of their ducks in a row (and stock on the shelves)? It's not like people would have died if the launch was delayed by a month. They also could've avoided the issues with stock and bots along with customers not having crashy cards.

It's not just nVidia that does this either and it's not just the tech industry. All industries pull this stupid crap and it makes me wonder when they'll finally learn to get the damn product right before asking people to pay for it. It really shows how little regard that companies have for the customers that keep them solvent.

 

[Nobina]

Maybe people are just holding it wrong ? Has happened before....

If you're holding a card by it's capacitors then yeah, you're probably doing it wrong.

 

[Avro Arrow]

If you're holding a card by it's capacitors then yeah, you're probably doing it wrong.

I trieded that once but it didn'ted worked neither.

 

[richcz3]

Well, its hard to believe that during testing they didn't OC their boards to 2+Ghz. They would have readily seen the crash to desktop. That Founders Editions are prone to the same issue given nVidia has preferred bins says a lot.

Given the increased power draw and extensive cooling required, this series isn't anywhere as OC friendly as previous generations. If you're good to use factory clock settings, your not going to have these issues.

 

[grumblguts]

Since when was a resistor a CAP
These resistors on back of the board clean up the power going into the silicon they are resistors.
NOT capacitors

 

[p51d007]

Reminds me of the PC and other things that had bad caps, back in the early-mid 2000's. Changed a bunch of leaky caps back then.

 

[cliffordcooley]

Since when was a resistor a CAP

What makes you think they are resistors? Surface mount components are hard to tell, if you are not looking at labels.

 

[VariableSpike]

Since when was a resistor a CAP
These resistors on back of the board clean up the power going into the silicon they are resistors.
NOT capacitors

Since when do you smooth out power delivery with resistors? These are being used as bypass caps to smooth out the power coming in from the PCI-E slot and the connector, as any noise gets shorted to ground using them - I will agree that SMD's are a pain to identify, but common sense would dictate that these aren't resistors, especially as resistors usually just have component marking numbers on the top which can be used with tables to identify the resistance among other specs.

 

[TheBigT42]

Reminds me of the PC and other things that had bad caps, back in the early-mid 2000's. Changed a bunch of leaky caps back then.

We had that problem with a batch of 120 + Dell 970 Mother Boards and PSUs

Dell ended up sending us a bunch of Mother Boards, PSUs and a field tech to replace the parts in our entire order. After we did some screaming first.

 

[Irata]

Reminds me of the PC and other things that had bad caps, back in the early-mid 2000's. Changed a bunch of leaky caps back then.

One of my favorite builds sadly suffered that fate - a 1.13 Ghz ULV Tualatin Celeron system. When the first batch of Caps went bad a friend replaced the with hq Japanese ones. Sadly, the rest went bad after he moved, so that was it....

 

[Evernessince]

Given that there are reports of the crashes happening on Nvidia FE cards too - A bad/faulty reference board design, is more likely than the choice of capacitors.
@Igors Lab and @Jay2cents have been blowing smoke about this, without knowing the REAL problem and as such Nvidia have already washed their hands of the problem.
Now imagine the hulabaloo if this was AMD, given the stramash about poor drivers on the 5x00 series of cards.

I watched Jayz video on it and yeah, he is somehow trying to frame it as not Nvidia's fault. It is 100% Nvidia's fault, they set the spec that AIBs follow. There are people with the FE having the issue too.

He also pointed out that everyone likes to pile on when a card has issues, including people that don't even have the product. That made me chuckle a bit. It's true and yet he didn't seem to care when the same happened to Navi. Jayz is as much as an Nvidia homer as he's ever been.

Does this affect the 3090 kingpin addition?

I have the 3080 EVGA XC3 and a 3090 FE coming. I was going to sell the FE to someone and use the extra money to buy a Kingpin because I want the AIO built in.

The 3090 kingpin will likely have the more expensive capacitor array so it may not have the issue. That said there have been numerous reports of cards having issues even with those more expensive arrays. The only common point that seems to be causing issues is clock speeds exceeding or close to 2000. It was demonstrated at those clocks speeds how even small amounts of EMI can cause those cards to crash. For example, from your other electronics, microphones, ect.

I'm really confused by all this. Here, nVidia has the awesome RTX 3080 (and it really is awesome) but it's very obvious that they rushed the launch so much that problems were inevitable. Why didn't they just take another month to make sure that they had all of their ducks in a row (and stock on the shelves)? It's not like people would have died if the launch was delayed by a month. They also could've avoided the issues with stock and bots along with customers not having crashy cards.

They wanted to beat AMD to the punch. Why? If Nvidia launches first, their cards get compared to the overpriced and underperforming turning cards. The $1,200 price tag of the 2080 Ti and the 28% performance improvement it brought makes the 3080 look like good value, essentially giving Nvidia free positive reviews. Nvidia was praised for lowering prices, even though it's the one that inflated prices in the first place. Had AMD launched first, the script completely flips. Nvidia instead looks greedy (which it very much is) and the AMD cards look very good in comparison to a very mediocre turning. On top of that, Nvidia's Ampere would instead be compared to RDNA2 cards, which would be much more unfavorable than Turing.

Nvidia was very likely aware of what AMD could bring to the market thus why it priced the 3080 at $700 while giving it the big die and why it rushed to the market despite the possible issues.

The 3080 is good, not awesome. 20% average performance boost over the 2080 Ti for those running at 1440p and inconsistent performance. Another increase in power consumption as well. I wouldn't expect much from a 3080 Ti either if there even is one. There's only a 10% gap between the 3080 and 3090. For all intents and purposes the 3080 is the 3080 Ti this time around.

Well, its hard to believe that during testing they didn't OC their boards to 2+Ghz. They would have readily seen the crash to desktop. That Founders Editions are prone to the same issue given nVidia has preferred bins says a lot.

Given the increased power draw and extensive cooling required, this series isn't anywhere as OC friendly as previous generations. If you're good to use factory clock settings, your not going to have these issues.

It's not just at OC either, it happens if you leave the card at stock. It just becomes immediately obvious if you OC.