Graphics card makers respond to RTX 30-series capacitor controversy (updated)

midian182

Posts: 6,075   +50
Staff member
TL;DR: There's been reports of GeForce RTX 3080/3090 crashing issues, rumored to be related to the type of capacitors used on the cards. However, several vendors have given statements regarding the situation and finally Nvidia came to the rescue with a driver update that should solve all problems. Basically Nvidia had pushed all cards to the edge of their capabilities and factory OC models were pushing them a little too hard. The latest GeForce driver lowers boost clocks by 1 to 1.5% which sees no real change in performance, but graphics cards are now stable, and that's all that matters.

For those who don't know, we've seen numerous reports of factory overclocked aftermarket RTX 3080 and RTX 3090 cards suffering from crashes and stability issues when they reach or exceed 2.0 GHz. It was thought that the problem was due to the capacitors found on the rear of the PCB underneath the GPU, but this has now been rebutted by most manufacturers.

The cards can use two types of capacitors, or a mix of both: the large-area POSCAPs (Conductive Polymer Tantalum Solid Capacitors), which operate better at higher temperatures but aren't as good at high frequencies; and MLCCs (Multilayer Ceramic Chip Capacitor), which have high current, voltage, and temperature ratings, operate better at higher frequencies, but are prone to cracking. EVGA confirmed that problems arose during QC testing when it used an all-POSCAP design, prompting a change to a mixed solution.

Several card makers and Nvidia, have responded to reports of the RTX 30-series crashes. "NVIDIA posted a driver this morning that improves stability. Regarding partner board designs, our partners regularly customize their designs and we work closely with them in the process. The appropriate number of POSCAP vs. MLCC groupings can vary depending on the design and is not necessarily indicative of quality."

The driver does fix the issues, and does so by slightly limiting the maximum GPU Boost clock speed. The driver lowers the boost clocks by a very minor 1 to 1.5% which sees basically no change in performance. It is also possible (and apparent) that Nvidia has made slight tweaks to voltages, but all in all, this improves stability without affecting performance.

Inno3D is confident that its products don't suffer from any issues: "To all current users and prospective buyers, Please be rest assured, and we hereby declare that INNO3D/ICHILL GeForce RTX 30 Series products do not have any instability problems."

For a while though, some makers were sending mixed signals. Zotac said it is working on a solution: "What you said is recognized by both us and the head office. We have been informed that we are aware of the current situation and are working with NVIDIA on a solution. If a countermeasure is available, we will notify you through a notice immediately."

Asus posted on Instagram "All retail ROG Strix and TUF Gaming GeForce RTX 3080 and 3090 graphics cards use only MLCC capacitors for decoupling close to the GPU. During development, we discovered the improvement this makes to RTX 3090 and 3080 overclocking headroom, so we made specification changes before we started shipping cards to reviewers and customers. Please note that some of the product images used on etail sites and our product pages were from early development samples, so are not final. All images will be updated soon. Please bear with us!"

MSI released a statement as follows: "MSI became aware of reports from customers, reviewers, and system integrators that there may be instability when GeForce RTX 30 Series graphics cards core clocks exceeded a certain amount. The latest GeForce driver (456.55) includes fixes for the issue. As such, MSI recommends owners of GeForce RTX 30 Series graphics cards update to the latest driver release which can be downloaded from the NVIDIA GeForce website. MSI stands behind its design decisions for its GeForce RTX 30 Series graphics cards catalog which consists of GAMING models and VENTUS models. MSI utilizes a mixed capacitor grouping in its designs to benefit from the strengths of both SP-Caps and MLCCs."

Permalink to story.

 

brucek

Posts: 529   +614
TechSpot Elite
In the linked PC World article, the author reported that the prior driver was running OK at 2010 MHz, and crashing at 2025 MHz; and with the new driver, runs at 1980-1995. This resulted in a drop from 129 fps to 127-128 fps in the game tested.

The article was subsequently updated: "An Nvidia representative reached out to say the company hasn’t “done anything to lock the GPU to sub-2 GHz operation with the new driver.”
 
  • Like
Reactions: MWR01

p51d007

Posts: 2,518   +1,804
And that`s why I never preorder...
I never like to be the "gen 1.0" of anything. Software updates, new phones, new computer. I like to give them a couple weeks/months to air out the gen1 bugs.
I got burned 30 years ago on software updates. We had a bunch of stand alone fax machines that had to have the firmware updated. That required replacing 10 e-proms on a board for each machine. I received a bunch of tubes of e-proms, went out and replaced them in 10 machines.
2 days later, they recalled them due to an unknown bug, requiring me to go back and pull them all out. Ever since then, I just wait a week or two before updating, or, buying anything "new".
 

Squid Surprise

Posts: 3,453   +2,348
What we need to know is if you have a card without defects (like the Inno one), will the updated drivers still decrease the speed?

Glad I’m not upgrading this generation....
 
  • Like
Reactions: Avro Arrow

Evernessince

Posts: 5,418   +6,010
So all 3080's get crippled because of a few AIB's cheaped out?
You mean because Nvidia set the minimum spec too low. AIBs were following Nvidia guidance. There are always going to be budget models regardless of card. Not everyone needs ln2 cards. It was up to Nvidia to set the floor and they messed up.

So what cards are going to be affected by this driver performance reduction and what solution is being provided to owners of cards that are effected (a clock reduction is not a solution)?

I believe we will also need a retest with the updated driver as well to see the impact on performance. If they reduce clocks by 5%, that would be significant.
 
Last edited:

texasrattler

Posts: 988   +433
So if reading this right, evga, msi and inno dont have any issues to begin with? So they should be able to be overclocked back to their original speeds, correct? Evga and msi made changes before production went out which is why I included them.
 

Shadowboxer

Posts: 933   +553
Nvidia's new driver addresses the problems, but does so by limiting clock speeds

Kind of sounds like Apple admitting they were slowing down older phones by doing nothing until caught.
In my opinion it was bad but they did it for good reasons, old batteries drop voltage to the point where phones power down. I used to own an LG G3 which suffered from this, it was fixed when I replaced the battery but I would have preferred to just drop the clocks by a couple of hundred MHz.
 
So all 3080's get crippled because of a few AIB's cheaped out?
Nope, you're wrong there. Changing the Boost 3.0 table to lower the possible boost still allows the cards to run at advertised base and boost speeds. It limits the upper range in a minor way, or so it appears. If you buy a card and manually over clock you take the chance of running into the limit yourself. And you still have the choice to manually overclock if you so desire, so no effect there at all. I would say that 80 - 90% of the users out there will not even notice a difference, other than the card stops hitting a hard limit that caused it to crash.

Der8auer released a great video today looking at the capacitor configuration too. The whole thing has been a typical internet maelstrom of garbage.

I would bet that many content makers are hard at work doing bench marks and we will find that the change has had a minimal effect on performance. I would hazard to guess it will be margin of error even.
 

Ludak021

Posts: 279   +209
So all 3080's get crippled because of a few AIB's cheaped out?
well, if they had 1 MLCC and 5 POPs they were fine, they were within the nVidia base spec, if they had more MLCC (FE had 2) they were fine. Now the AIBs that cheped out were the ones that put no MLCCs whatsoever. I will avoid them forever, no matter AMD/nV this that, that's a mentality, should go broke if you ask me.
 
  • Like
Reactions: eforce
In my opinion it was bad but they did it for good reasons, old batteries drop voltage to the point where phones power down. I used to own an LG G3 which suffered from this, it was fixed when I replaced the battery but I would have preferred to just drop the clocks by a couple of hundred MHz.
And in my opinion they created the problem in the first place by grossly underspeccing their batteries. iPhones have absurdly small battery capacities compared to equivalent Android phones and peak power delivery of a lithium rechargeable cell is directly proportional to capacity. Android phones can afford to lose far more capacity than Apple phones before power instability even becomes a problem. IMHO they were technically correct that the throttling was the correct move but customers still have every right to be infuriated both because they hid this from them and because it was entirely preventable by actually putting a decently sized battery in the damn thing in the first place.
 
  • Like
Reactions: Ludak021

Julio Franco

Posts: 8,675   +1,549
Staff member
Story updated...

TL;DR: There's been reports of GeForce RTX 3080/3090 crashing issues, rumored to be related to the type of capacitors used on the cards. However, several vendors have given statements regarding the situation and finally Nvidia came to the rescue with a driver update that should solve all problems. Basically Nvidia had pushed all cards to the edge of their capabilities and factory OC models were pushing them a little too hard. The latest GeForce driver lowers boost clocks by 1 to 1.5% which sees no real change in performance, but graphics cards are now stable, and that's all that matters.
 
  • Like
Reactions: MWR01 and treetops

Sherwoodnt

Posts: 23   +39
Would having a better PSU (platinum rated or whatever rating is best) have any effect on the GPU stability? Would that provide cleaner power to these capacitors and cause them to be "less necessary"? Or does it not work like that?
 
  • Like
Reactions: Mr WW

Avro Arrow

Posts: 211   +203
I never like to be the "gen 1.0" of anything. Software updates, new phones, new computer. I like to give them a couple weeks/months to air out the gen1 bugs.
I got burned 30 years ago on software updates. We had a bunch of stand alone fax machines that had to have the firmware updated. That required replacing 10 e-proms on a board for each machine. I received a bunch of tubes of e-proms, went out and replaced them in 10 machines.
2 days later, they recalled them due to an unknown bug, requiring me to go back and pull them all out. Ever since then, I just wait a week or two before updating, or, buying anything "new".
I wait at least three months. A week or two is enough for them to find the problem but three months is usually enough time for them to fix any problem they found.
 

Avro Arrow

Posts: 211   +203
What we need to know is if you have a card without defects (like the Inno one), will the updated drivers still decrease the speed?

Glad I’m not upgrading this generation....
Man, I haven't seen an Inno3D card in years. It makes me think of a lot of the other AIB brands I haven't seen in years like BFG (BFG is dead), Diamond, Albatron, VisionTek, Sparkle, Club3D and HIS. :D
 
Last edited:
  • Like
Reactions: Charles Olson

Avro Arrow

Posts: 211   +203
Would having a better PSU (platinum rated or whatever rating is best) have any effect on the GPU stability? Would that provide cleaner power to these capacitors and cause them to be "less necessary"? Or does it not work like that?
I don't think so. There's like a 3% difference between Gold and Titanium which is really splitting hairs at that point. You need top-tier components just to reach Gold. Having said that, I wouldn't get a PSU that isn't at least rated at 80+Gold but I also wouldn't pay a lot more for any rating above that either.
 
Last edited:
  • Like
Reactions: Sherwoodnt

hahahanoobs

Posts: 3,022   +1,195
I thought the point of the update was to say there is no
Nvidia's new driver addresses the problems, but does so by limiting clock speeds

Kind of sounds like Apple admitting they were slowing down older phones by doing nothing until caught.
Less like Apple, more like AMD...

Basically Nvidia had pushed all cards to the edge of their capabilities and factory OC models were pushing them a little too hard.
 

hahahanoobs

Posts: 3,022   +1,195
Sooooo the AIB's and caps are not to blame after all?

Update: I get it now. Not the caps. Could be driver.
 
Last edited:

Adhmuz

Posts: 2,060   +854
In my opinion it was bad but they did it for good reasons, old batteries drop voltage to the point where phones power down. I used to own an LG G3 which suffered from this, it was fixed when I replaced the battery but I would have preferred to just drop the clocks by a couple of hundred MHz.
I never had any such issues with my G3, used it for 3 years before upgrading to the G6, gave the G3 to a family member who used it another year before the battery wouldn't hold a charge anymore. Put a new battery in it and still works great to this day, gotta love user replaceable batteries.

Sooooo the AIB's and caps are not to blame after all?
I'm suspicious either way, will need larger sample size of cards experiencing the crashes. If enough cards with the 6 large cap design are crashing when exceeding 2GHz core clock there's clearly an issue with the design.

What I would also want to know is if cards that use no large caps, instead opting for arrays of the smaller MLCC caps, are genuinely more stable at higher clocks or do not experience any crashing. These models would be what I'd be interested in buying over any other design. Looks like Asus for sure has gone this route, which is good, because they would have been my first choice of card anyways.
 

Avro Arrow

Posts: 211   +203
I never had any such issues with my G3, used it for 3 years before upgrading to the G6, gave the G3 to a family member who used it another year before the battery wouldn't hold a charge anymore. Put a new battery in it and still works great to this day, gotta love user replaceable batteries.


I'm suspicious either way, will need larger sample size of cards experiencing the crashes. If enough cards with the 6 large cap design are crashing when exceeding 2GHz core clock there's clearly an issue with the design.

What I would also want to know is if cards that use no large caps, instead opting for arrays of the smaller MLCC caps, are genuinely more stable at higher clocks or do not experience any crashing. These models would be what I'd be interested in buying over any other design. Looks like Asus for sure has gone this route, which is good, because they would have been my first choice of card anyways.
Well, one thing is certain, the nVidia reference spec has five POSCAPs and one MLCC array. Therefore, the cards with POSCAPs are BELOW nVidia's spec and that IS the AIB partner's fault, unless nVidia was directly involved.

Here you see the EVGA RTX 3080 FTW using the reference spec of 5 POSCAPS and one MLCC array. Below it is an RTX 3080 FE which is actually built above reference spec with four POSCAPs and two MLCC arrays. I know that Zotac, Colorful and Gigabyte used the below-spec six POSCAP setup and if there are problems with it, it is THEIR fault because they didn't build the cards to reference spec or above. The reference spec is the MINIMUM spec for the card. I'm sure that nVidia is going to have words with Zotac, Colorful and Gigabyte over this because the other major manufacturers did their cards correctly:
 
Last edited:
  • Like
Reactions: MWR01
Well, one thing is certain, the nVidia reference spec has five POSCAPs and one MLCC array. Therefore, the cards with five or six POSCAPs are BELOW nVidia's spec and that IS the AIB partner's fault.
Unless you have NVidia's specs at hand, you're just guessing. There is nothing that says NVidia didn't build their own cards beyond the minimum spec they say should be used.

Der8auer posted a great video discussing the capacitor issue, and he modded a card that had 6 SP-CAPs by removing 2 and adding MLCCs in their place. He found that it affected his overclock by about 30 MHz.

You can bet that the AIBs are following the engineering instructions, but yes, they will cut corners where they can. That is why you find cheaper cards, and more expensive cards. From the videos by Der8auer and Buildzoid, the MLCCs are better high frequency capacitors, and the SP-CAPS are better at capacity. Having a mix will net you a card that can run more stable at higher frequencies, it appears. Having 6 SP-CAPS is likely that they will just stay within advertised boost numbers.