Let's talk about Nvidia's RTX 3080 CTD Issues - Are SPCAPS and MLCCs to blame?

[WYP]

These Crash To Desktop issues are a lot more complicated than you think.



Read more about Nvidia's RTX 3080 stability issues and their complexities.

 

[AlienALX]

No I don't think so. And thanks Mark for starting this thread. I just got done posting on another forum, and here are my thoughts.

It wasn't about the capacitors in the first place. If the filtering and noise was becoming an issue (where as it wasn't on Turing etc) then it is because the core is not stable. Once again people jumped the gun in order to claim a world's first, when the real issue was nothing to do with it.

For some reason the driver itself, without any software, is boosting the cards faster than they are rated at. Quite probably to make them look better in reviews, IE let them automatically overclock and then you don't have to worry about reviewers marking them down by 10% and then adding it in with overclocking to make people think "Well I don't overclock so to me it's only going to be 20% faster". The problem though, just like with the 5600XT after launch driver is Nvidia have written cheques they can't cash. IE, not every single 3080 die out there is going to be able to do what the drivers are telling them to do without running into stability issues. And that was your problem, not capacitors or filters or anything else.

The fact is that some cores are going to need more power. 10w more it turns out, and lower boosts (they have been lowered to 1930 IIRC from 1970 odd). Quite why they are doing this when the box rated is 1710 I don't know. However what I do know is that 1900+ mhz is 10% more than 1700+ mhz. Especially in your benchmark scores

It's why I have had my cards under water ever since the air coolers simply couldn't cope any more. I hated leaving that sort of performance and more wasted blowing around in a haze of hot air.

The TGP was already absolutely terrible. However, it would have sounded even more terrible with another 10w slapped onto it.

 

[CaTcHmG]

Sorry but undervolting and underclocking should not be THE fix, it clearly states BOOST, so why pointing to reference speeds on the box when it clearly states cards can boost???

Another factor concerning cards that are going to be doing the rounds in the next few months that are going to be in the market as STAY AWAY maybe give the 3080 series a miss

 

[WYP]

Sorry but undervolting and underclocking should not be THE fix, it clearly states BOOST, so why pointing to reference speeds on the box when it clearly states cards can boost???

Another factor concerning cards that are going to be doing the rounds in the next few months that are going to be in the market as STAY AWAY maybe give the 3080 series a miss

Underclocking is not a fix, but it should allow affected users to get their cards working in the meantime. It may take a while for AIBs to be ready to replace cards or for Nvidia to fix things with drivers etc.

Companies still need to react to this, and that takes time. In the meantime, RTX 3080 owners can underclock their GPUs.

No I don't think so. And thanks Mark for starting this thread. I just got done posting on another forum, and here are my thoughts.

It wasn't about the capacitors in the first place. If the filtering and noise was becoming an issue (where as it wasn't on Turing etc) then it is because the core is not stable. Once again people jumped the gun in order to claim a world's first, when the real issue was nothing to do with it.

For some reason the driver itself, without any software, is boosting the cards faster than they are rated at. Quite probably to make them look better in reviews, IE let them automatically overclock and then you don't have to worry about reviewers marking them down by 10% and then adding it in with overclocking to make people think "Well I don't overclock so to me it's only going to be 20% faster". The problem though, just like with the 5600XT after launch driver is Nvidia have written cheques they can't cash. IE, not every single 3080 die out there is going to be able to do what the drivers are telling them to do without running into stability issues. And that was your problem, not capacitors or filters or anything else.

The fact is that some cores are going to need more power. 10w more it turns out, and lower boosts (they have been lowered to 1930 IIRC from 1970 odd). Quite why they are doing this when the box rated is 1710 I don't know. However what I do know is that 1900+ mhz is 10% more than 1700+ mhz. Especially in your benchmark scores

It's why I have had my cards under water ever since the air coolers simply couldn't cope any more. I hated leaving that sort of performance and more wasted blowing around in a haze of hot air.

The TGP was already absolutely terrible. However, it would have sounded even more terrible with another 10w slapped onto it.

Nvidia's last couple of GPU generations have had GPUs go far past their rated boost clocks. That's just how GPU boost 2.0 and newer work.

As far as the capacitors go, its a multitude of factors. If Nvidia weren't as aggressive with their clocks, this wouldn't be a problem. If Nvidia had given their AIB partners more time to test their designs and gave them full driver access earlier, this wouldn't be a problem. Stricter component guidelines may also have helped matters.

There are a lot of ways that Nvidia could have avoided this, and a lot of them involve Nvidia taking their time with this launch. Ampere feels rushed.

Like lots of Intel's recent offerings, Nvidia has pushed hard on the Voltage-Frequency curve. There's a reason why Ampere's real world performance/watt improvements over Turing aren't that substantial. To me, it sounds like Nvidia feels threatened by RDNA 2.

 

[Greenback]

Sorry but undervolting and underclocking should not be THE fix, it clearly states BOOST, so why pointing to reference speeds on the box when it clearly states cards can boost???

Another factor concerning cards that are going to be doing the rounds in the next few months that are going to be in the market as STAY AWAY maybe give the 3080 series a miss

If it's on the box as 1710 and goes up to 1710.5 every 4 hours it can boost, can is a very grey area

 

[Warchild]

I still think this was all known before launch, and part of the reason why stock is so bad. Once you realise the issue, and you already had XX amount of stock on its way to sellers, you arent going to keep making more, if you think Nvidia can resolve it through drivers or not.

If Nvidia fix it via drivers then AIB can continue to use cheaper components. Otherwise a small redesign on the pcb is needed.

regardless of it all, a driver fix is just a workaround solution and no way on Earth should any owner of a 3080/3090 have to settle for this without entitlement to an RMA. It does not matter what is on the box. If you are allowed to overclock it then you should be able to push it to its limits without hitting issues because of component instability. This is not the same as silicon lottery.

 

[AlienALX]

Underclocking is not a fix, but it should allow affected users to get their cards working in the meantime. It may take a while for AIBs to be ready to replace cards or for Nvidia to fix things with drivers etc.

Companies still need to react to this, and that takes time. In the meantime, RTX 3080 owners can underclock their GPUs.



Nvidia's last couple of GPU generations have had GPUs go far past their rated boost clocks. That's just how GPU boost 2.0 and newer work.

As far as the capacitors go, its a multitude of factors. If Nvidia weren't as aggressive with their clocks, this wouldn't be a problem. If Nvidia had given their AIB partners more time to test their designs and gave them full driver access earlier, this wouldn't be a problem. Stricter component guidelines may also have helped matters.

There are a lot of ways that Nvidia could have avoided this, and a lot of them involve Nvidia taking their time with this launch. Ampere feels rushed.

Like lots of Intel's recent offerings, Nvidia has pushed hard on the Voltage-Frequency curve. There's a reason why Ampere's real world performance/watt improvements over Turing aren't that substantial. To me, it sounds like Nvidia feels threatened by RDNA 2.

Yeah it definitely smells of fear to me. Hare and tortoise come to mind.

 

[AlienALX]

I still think this was all known before launch, and part of the reason why stock is so bad. Once you realise the issue, and you already had XX amount of stock on its way to sellers, you arent going to keep making more, if you think Nvidia can resolve it through drivers or not.

If Nvidia fix it via drivers then AIB can continue to use cheaper components. Otherwise a small redesign on the pcb is needed.

regardless of it all, a driver fix is just a workaround solution and no way on Earth should any owner of a 3080/3090 have to settle for this without entitlement to an RMA. It does not matter what is on the box. If you are allowed to overclock it then you should be able to push it to its limits without hitting issues because of component instability. This is not the same as silicon lottery.

The issue is not cheaper components. At all. That is what certain gun jumping Youtubers said.

Nvidia FE uses all cheap filters. They are cheaper, they filter better but they are prone to cracking as they are ceramic. Where as the Tantalum ones (more expensive, not as good at filtering but last longer) when mixed with the cheaper ones causes supposed noise which means your card crashes.

MLCC caps are cheaper, but the 3080 needs loads. So AIBs used better tantalum ones and mixed them, which apparently was the issue that Jay discovered via Igor.

It wasn't. The issue is that certain cards (ALL cards with a poorer die than the FE) can not do the 2ghz+ being asked of them by the boost algorithm in the drivers (note drivers, not an app overclocking) and thus they crashed to desktop just like your GPU would when you reach the OC limit.

For some reason the drivers are boosting the cards to and in some cases beyond their limits. It has bugger all to do with capacitors, as when you drop the clocks by 50mhz and increase the TGP by 10w the cards are all of a sudden stable, yet warmer and slower.

The fact is that these cards should have been launched at the 1710mhz stated by Nvidia on the box. They should have been reviewed at those speeds because that is the ONLY speed Nvidia are guaranteeing you. Then, after that reviewers could have overclocked the cards and showed the gains.

Instead Nvidia wanted every one to see ALL THE GAINS ! (insert that meme here) to make their cards look as fast as possible and it has basically blown up in their face like when AMD did that 5600XT bait and switch and companies like MSI had to come out and say "Look, not all cards can run at these speeds and it's crappy of AMD to put the onus on us".

That is what Nvidia have done. Allowed the cards to clock to the limit and sadly for some beyond.

So their fix is more watts and lower clocks.

That is why they "overclock" so poorly. 6% was it? because yeah basically they have been pre overclocked by Nvidia to look as good as possible.

Because they are worried. And they cheaped out. Those are the two main reasons for this.

 

[Blane267]

Gotta love the double standard for you Reviewers if AMD had done something similar you'd have been up their arses about it.it's Nvidia so its fine.

 

[WYP]

Gotta love the double standard for you Reviewers if AMD had done something similar you'd have been up their arses about it.it's Nvidia so its fine.

Double standard? I don't know what you'd want me to do to here, but if you don't think this paint Nvidia in a negative light, you are sorely mistaken. If I had a pro-Nvidia bias, I'd not have reported on these issues at all.

Let's not mince words here, these issues are Nvidia's fault. They should have given their partners more QA time with proper drivers to uncover these issues, and they should have had their driver stability fixes in place in time for launch.

Had this been an AMD or Radeon issue, it would have been reported in a similar way. If you want me to scream hatred at Nvidia and over sensationalise things, you can go elsewhere. There are other folks who do that kind of reporting.

This article is about the complexities of this issue and how saying "it's the SPCAPS" does not tell the whole story.

 

[Blane267]

Had it been an AMD issue there would be a video showing how they messed up,yet nothing from you guys,Jayztwocents,Bitwit,Paulshardware,Gamersnexus,Linustechtips etc they all would have had videos calling AMD out.blah blah no stock so we can't test it.I'm sure they would have found a way to show AMD's f up.

 

[NeverBackDown]

Had it been an AMD issue there would be a video showing how they messed up,yet nothing from you guys,Jayztwocents,Bitwit,Paulshardware,Gamersnexus,Linustechtips etc they all would have had videos calling AMD out.blah blah no stock so we can't test it.I'm sure they would have found a way to show AMD's f up.

You're argument is entirely hypothetical based off assumptions of people you don't know. You must be fun at parties

 

[Tolemac]

Had it been an AMD issue there would be a video showing how they messed up,yet nothing from you guys,Jayztwocents,Bitwit,Paulshardware,Gamersnexus,Linustechtips etc they all would have had videos calling AMD out.blah blah no stock so we can't test it.I'm sure they would have found a way to show AMD's f up.

Any evidence to back your comments up? any links?.

 

[WYP]

Had it been an AMD issue there would be a video showing how they messed up,yet nothing from you guys,Jayztwocents,Bitwit,Paulshardware,Gamersnexus,Linustechtips etc they all would have had videos calling AMD out.blah blah no stock so we can't test it.I'm sure they would have found a way to show AMD's f up.

I don't know what you expected from us, but we don't typically do reactionary videos about issues before we have all of the relevant information on hand. I try my best not to spread misinformation. I am also not a YouTuber.

I think you will also find that there are several videos from other reviewers discussing the issue on YouTube. I believe JayzTwoCents and Pauls Hardware have videos on this topic.

 

[grec]

If your argument is that more people would be doing deeper technical analysis of the issue were it AMD then I'm afraid I think you're quite mistaken, if even AIB partners knew what the root issue was (That affects all cards) then I think we'd all know by now, and without knowing that there's little point spending days analysing/simulating/ect power setups

 

[AlienALX]

Had it been an AMD issue there would be a video showing how they messed up,yet nothing from you guys,Jayztwocents,Bitwit,Paulshardware,Gamersnexus,Linustechtips etc they all would have had videos calling AMD out.blah blah no stock so we can't test it.I'm sure they would have found a way to show AMD's f up.

Not true.

Jay is an asshole. The more I watch Not An Apple Fan the more I think so. I also agree with him about Linus. I think once it was about the passion for Linus but his only passion now seems to be paying his staff and growing his company. Bitwit and Paul are vapid eejits who just copy every one else and make a living from it.

They are boils on the ass of the PC gaming industry. They are not helping anything but sales.

However even though I can't stand his messy persona I do like Steve @ GN. He did what none of those other greedy gits had the guts to do, and showed the 3090 for what it is. Totally over priced stupidity, helped mainly by the scarcity of the 3080.

People think I don't like Nvidia just because. And that is not true. I don't like them mainly for their practices, and to a real person like me living in the real world even when they have a fantastic product they have to go and ruin it all by chatting s**t.

I don't like people, or companies, or any one else who embellish the truth. None of us do !! if your friends or my friends or wife etc spent the whole time lying to you or I we would ditch them. No one likes liars, apart from when they are selling you something.

Yeah see I don't follow that philosophy. I buy products based on their merits, and when they are encrusted in BS it just makes me want to avoid them. Which in turn makes me feel sad, because I can't see the good product past the BS.

I also get accused a lot of being an Nvidia hater and an AMD lover but believe you me I think Vega and Fury were two of the poorest products I have ever seen. Mostly because not only were they crap, but they too were surrounded with bulls**t. So not only do you end up having to admit it's a crap product but you also end up having to accept all of the crap they spoke when they talked about it before launch etc.

The thing is? Nvidia have grown "good?" at talking s**t. What I mean is, their loyal fanbase absolutely love it. Mostly because they are braggart waynes who just love to go around posing. AMD on the other hand? are not good at talking s**t. Nor should they even try.

Influencers. That is what those guys are.

Tom could not influence people if he tried. Firstly he knows we are not stupid enough and secondly he doesn't have the "customer base". IE about two million more subs, an American accent and the ability to turn a ten second sentence into a 30 minute video.

 

[AlienALX]

https://videocardz.com/newz/msi-qui...-gaming-x-trio-design-amid-stability-concerns

MSI redesign their PCBs.

https://videocardz.com/newz/asus-al...eforce-rtx-3080-tuf-and-rog-strix-pcb-designs

And so do Asus.

 

[KingNosser]

I'm not an electrical engineer so I can't really comment much on the caps issue, but as a normal guy i expect it's normal to cost cut in lots of ways to make a bit more margin.

Does that mean they are not upto spec or able to maintain clocks higher than rated, this is a hard thing for me to judge really it's a new product lots of things have issues at launch, like the 2k series had ram issues if i remember so this isn't out of the normal not really.

It could be partly drivers if it is it'll improve over time sure enough.

Thou last week I was in a zone where I still wasn't decided between 3080/rdna2.

This week thou RDNA2 Regardless!!

I'm not a fanboy, i've had as many amd cards as i have had nvidia ones over the years i'm old. I can see thou given I heard full production started in August, that this is a rushed product launch, they are scared of amd and rdna2 and they have every reason to be them cards are going to be really strong, people have got so used too nvidia being ahead they seem to think they always will be so automatically asume amd will fall behind.

Well amd have caught up now, nvidia needed the product out to get sales build hype to sell more, and they will despite some of the issues regardless of what they are they will still sell. AMD on the other hand do like to wait and see what nvidia do before they step in it's been normal for them now for a long time, they simply can't take the massive risk of huge production if they can't be on level footing in some way.

Going RDNA2 for me is for two main reasons,

Nvidia over promised the product expectations they stated things in the reveal that do not turn out to be 100% correct.

2nd I hate their marketing last few years RTX sorry no it's DXR and it's made by microsoft I don't care if you call it something else to me that is plain theft in my view it's no different to taking a story and then pasting it on another site without giving credit to the person that wrote it and i really really dislike that.

I don't care if it's drivers or hardware it'll get solved, I'm not even saying they are terrible cards even thou the power useage is huge but not priced way above the norm.

I just can't buy into them as a company, the need to make something sound better than it is, really is not needed, they are just telling porkies to build hype, i'm not blind i can see the benchmarks, but these recent crashing issues regardless of how they are happening or how they will get fixed is a huge turn off from the whole series.

over promised, over hype'd, terrible marketing, crashing.

Why would I want one, i'll take my chances with AMD.

 

[AlienALX]

I read an article the other day about RT and so on. It was saying how Nvidia didn't want to keep making "gaming" GPUs so found something their industry cards could do (RT and DLSS) and thus sell tanks back to gamers whilst using the money to push on in the industry.

Those sorts of GPUs are never good for gamers. Ever. Seems like when Pascal was over though that was the direction they were going in regardless, so made sure to call them RTX cards to sort of "pee" on the RT naming so they got in first. Someone made a good point about it, but now AMD are kinda behind and people will just say "But does it do RTX?". Talk about how to take a common name and make it part of your branding.

I don't dislike that though.

On the counter side of that? AMD have slashed CCNs, so the clock speeds should soar. Which is a bit of a risk, but with so few big RT games out and only one really big one coming that could be the best move they've made in years. Nvidia go back to tanks, AMD go back to lighter dies with huge clock speeds.

But yeah the whole launch has been a mess. From paying people to tell porkies to stretching and bending about every bit of truth they could and why? the 3080 would have been an absolutely killer card if they had just reigned in the bull and focussed on the card itself. They didn't need to make it sound twice as fast as a 2080 it was fast enough. They didn't need to talk crap on Turing owners and make them feel like their £1300 Ti was about to get smashed by a card costing less than £500 (which I can't see happening at all) but yeah, their whole "treat em mean keep em keen" is just outdated.

If people really are daft enough to let some nob make them feel inferior then I guess the world really has gone to hell.

The thing is it was just all so unnecessary. So many people totally skipped Turing. It's not even funny how many forum users have told me they are coming from like, a 980Ti and wanted a 3080. Even people with the 1060 seem to think it's a great deal. And then they go and wreck it all by just talking poo.

I've never understood that American mentality. Never. Like, I once lived in a pretty poor part of the country (USA) and I used to walk into like, 7/11 and some poor haggard looking woman was busting her ass working there for peanuts. Her life looked miserable and minimum wage, yet she was expected to smile and say "have a nice day". I used to say to them "Seriously, you don't need to do that. It's OK". To me it was like a slap in the face. Even the poor hard workers are forced into this positive mindset when their lives are total s**t and all for the sake of making someone else rich.

The failures after launch are not what you would expect from a so called world class leading company. The launch was childish and stupid. BFGPU? come on now. That was funny when I was like 13. It's a tacky thing to call your world beating £1500 GPU really. One that costs nearly twice as much as it should because of the extra VRAM.

3090, fail. 3080 marketing, fail. 3080 stability, fail. 3080 power consumption, epic fail. There is only so much you can take negatively before it spoils all of the good things and all of it is their fault. No one else's. They could have seriously made me look like a grumpy old fool but grumpy as I may be I am wise and I am not stupid.

If the cards had been in stock and were out there now? I'd have probably thrown a U-ey and friggin bought one. But all of the BS and all of this other crap? I will probably skip the next gen completely now.

Oddly it literally took this whole filter thing to make loads of people cancel their orders. I guess they are feeling let down and a bit paranoid now.

 

[Kleptobot]

Good write up, i appreciate the emphasis you placed upon the complexity of the issue and that putting it down to just component selection is ignoring other factors.

Just to put in my 2c.

Power delivery is arguably the most fundamental part of PCB design. But designing the power supply goes beyond setting up the rails and selecting the power delivery components. Power supply design has to factor in proximity to data traces, where to put large areas of ground plane and the different power planes, what layers the different power planes primarily reside on etc.

I have done a reasonable amount of pcb design, nothing as complex as a gpu, but reasonably complex embedded systems running real time mission critical stuff. It has been incredibly frustrating to what people latch onto the capacitors as the smoking gun. If it was just the bypass capacitor configuration that was the problem there would be far fewer reports of instability.