Rtx 3080 lower quality capacitor Issue

azazl187 · 26 Sep 2020 at 06:16

This is a huge issue and basically means that the aibs who have used the cheaper part will not sell their cards

Rroff · 26 Sep 2020 at 06:19

Aretak said:
An AIB coming out and directly addressing and admitting the problem (and what it took to rectify the situation) is 100% cast iron proof of the situation, without any need for further speculation. Of course, there's probably still going to be some card to card variance based on the silicon lottery and manufacturing tolerances. It probably is the case that some individual cards work fine with the dodgy layout. But the issue is clearly inherent and widespread when using that capacitor layout, or there wouldn't have been such a panic about it behind the scenes and expensive, last-minute scrambling to replace it.

I'm not really convinced this is all there is to the story - I'm by no means an expert but I've enough experience to design and build a GPU power supply stage in a pinch from prototyping headphone amplifiers and know my way around things like voltage regulators and bypass caps. If you are so close to the edge of stability this is causing such a reproducible failure of this kind then something has gone very wrong.

Normally you could rip half these caps off, overclock a card to the max and maybe see sporadic instability if you were unlucky.

Grim5 · 26 Sep 2020 at 06:24

Does anyone here who has a crashing 3080 have an ASUS TUF model?

Grim5 · 26 Sep 2020 at 06:26

esoteric said:
On the flip side the AIB chose not to follow nvidias example and cheaped out on some materials AND charge a higher price than MRSP. Wether nvidia signed off on the design or not nvidia didn't make them do this.

Yep, blaming Nvidia is a cop out. These AIB's employee a lot of engineers, so suggesting that none of those engineers could spot a simple fault in power delivery and just assumed its fine if Nvidia approved is hilarious.

esoteric · 26 Sep 2020 at 06:27

The cards that are clocking over 2000MHz are they self boosting that high? I guess if ZOTAC cut their card down they are.

Rroff · 26 Sep 2020 at 06:44

Grim5 said:
Yep, blaming Nvidia is a cop out. These AIB's employee a lot of engineers, so suggesting that none of those engineers could spot a simple fault in power delivery and just assumed its fine if Nvidia approved is hilarious.

Be interesting to know how much the coronavirus situation has had an impact there - could be working from home has resulted in things being rushed and/or not as effectively done fully or partially remotely, etc.

esoteric said:
The cards that are clocking over 2000MHz are they self boosting that high? I guess if ZOTAC cut their card down they are.

Haven't looked into it in detail but looks like the regular models are self boosting to ~1840-1915MHz and the OC models ~1960-2040MHz in operation.

KompuKare · 26 Sep 2020 at 06:58

Grim5 said:
Yep, blaming Nvidia is a cop out. These AIB's employee a lot of engineers, so suggesting that none of those engineers could spot a simple fault in power delivery and just assumed its fine if Nvidia approved is hilarious.

Yes, but what if it isn't a simple fault?
Intermittent maybe only manifesting at 2+ GHz when certain other things are true?
The super secret driver and AIBs only having that Nvidia power load tool to work with might have meant that they couldn't test things properly.
Since there are six capacitor groups behind the chip, does this mean there are six sets of power lines? And the power load tool maybe didn't stress all six in a realistic way?

dazzlaa · 26 Sep 2020 at 07:22

Highlander71 said:
I have 25+ years experience in electronics manufacturing and I've never heard of tantalum capacitors being called 'lower quality' or 'cheaper' than multi-layer ceramic caps. They are simply different types. In cost terms multilayer ceramic caps are probably more than 25x cheaper than tantalum caps. In turn tantalum caps are larger and offer much bigger capacitance so you can put less on the board to achieve the same capacitance. The boards like the Asus TUF with all ceramics caps will likely be cheaper to build for that section even considering the extra costs for the SMT pick and place of 60 ceramic caps vs 6 tantalum caps.

They do have different characteristics like impedance, ESR etc which may account for some boards being better than others. If the capacitors do turn of to be an issue it's not going to be down to component quality or cost cutting, it's purely down to design and presumably if Nvidia have a design spec that gives the OEM board makers a choice of what to fit then the fault lies with Nvidia.

interesting, thanks for sharing that information.

Kaapstad · 26 Sep 2020 at 07:22

Grim5 said:
Yep, blaming Nvidia is a cop out. These AIB's employee a lot of engineers, so suggesting that none of those engineers could spot a simple fault in power delivery and just assumed its fine if Nvidia approved is hilarious.

NVidia also have a lot of engineers and should have taken more care keeping an eye on what the AIB partners were up to with their product on launch day.

NVidia have been developing Ampere for a long time and will have built up a huge amount of information about any pitfalls and possible problems that can occur, they should have used this knowledge to be more supportive of their partners.

I also think NVidia already know that Ampere has a bag load of potential problems because of the way their FE cards are massively over engineered with their massive fancy coolers etc.

Minstadave · 26 Sep 2020 at 07:24

Kaapstad said:
I also think NVidia already know that Ampere has a bag load of potential problems because of the way their FE cards are massively over engineered with their massive fancy coolers etc.

That looks more related to using Samsung 8nm. They had to completely blow the power budget to get the performance they wanted.

Kaapstad · 26 Sep 2020 at 07:27

Minstadave said:
That looks more related to using Samsung 8nm. They had to completely blow the power budget to get the performance they wanted.

Power budget, Power delivery, Capacitors....

Jam3s · 26 Sep 2020 at 07:36

Good morning all,

I've been lurking these forums for a while now.. I'm admittedly a new convert to PC gaming and decided recently to build my first PC to get into it, predominantly (at this stage) for VR.

Anyway...

I borrowed my brothers old GPU in anticipation of the 3080 release. I pre-ordered the gigabyte eagle and it has been concerning to see the developments over the past few days regarding the issues plaguing many 3080 models.

My question:

Do you think these high end issues would affect someone like me who may not necessarily be missing out by running at factory stated Hz since I have no other high end gaming experience to benchmark from?

TIA

Jam3s

Kaapstad · 26 Sep 2020 at 07:39

Jam3s said:
Good morning all,

I've been lurking these forums for a while now.. I'm admittedly a new convert to PC gaming and decided recently to build my first PC to get into it, predominantly (at this stage) for VR.

Anyway...

I borrowed my brothers old GPU in anticipation of the 3080 release. I pre-ordered the gigabyte eagle and it has been concerning to see the developments over the past few days regarding the issues plaguing many 3080 models.

My question:

Do you think these high end issues would affect someone like me who may not necessarily be missing out by running at factory stated Hz since I have no other high end gaming experience to benchmark from?

TIA

Jam3s

Even if you do get a faulty card the good news is this is something that will be fixed quickly and any faulty cards replaced.

Jam3s · 26 Sep 2020 at 07:45

Kaapstad said:
Even if you do get a faulty card the good news is this is something that will be fixed quickly and any faulty cards replaced.

Hi Kaapstad, thanks for the lighting reply and advice.

I think I'll hold of from switching cards for now on that basis. Thanks again!

Jam3s

Troezar · 26 Sep 2020 at 07:46

My bet is there are a number of moving parts in all of this. If it was just capacitor choice all cards of that configuration would be affected. However, if it's a combination of that, GPU quality, cooling, boost behaviour etc. Only those cards unlucky enough to have a particular configuration will exhibit instability. Much harder to find the pattern though if it's more than one factor.

The biggest question for me is the obvious lack of QC before the cards went out. At this price point I'd expect more. It's not like they were shipping millions of cards to satisfy global demand and a few crept in, there's only a handful of people that even have a card.

Smrtka · 26 Sep 2020 at 07:46

Rroff said:
I'm not really convinced this is all there is to the story - I'm by no means an expert but I've enough experience to design and build a GPU power supply stage in a pinch from prototyping headphone amplifiers and know my way around things like voltage regulators and bypass caps. If you are so close to the edge of stability this is causing such a reproducible failure of this kind then something has gone very wrong.

Normally you could rip half these caps off, overclock a card to the max and maybe see sporadic instability if you were unlucky.

I am no expert by any means, but strangely enough no one is mentioning power supply as potential casue, if the psu is pushing "dirty power" to the card could this not be the case?

Troezar · 26 Sep 2020 at 07:48

Kaapstad said:
Even if you do get a faulty card the good news is this is something that will be fixed quickly and any faulty cards replaced.

I'm not so sure about that. Where are these cards coming from with such huge backlogs of orders? I've already seen petite on this forum having to push to get to the front of the queue after being told their RMA would put them at the back.

0h_no_not_again · 26 Sep 2020 at 08:01

Holy ****, this whole "issue" is pure speculation.

There's a massive assumption that it is power deliver, in which case there are several causes:

Poor power plane layout, noting that the reference cards are HDI boards and some partner boards are not HDI. HDI are better for ultra high frequency stuff but quite a bit more expensive.
High frequency current issues, where the cards have 0201, 0402 and 0603 caps for this purpose.
Medium frequency current issues, which is where those big bulk caps in question come in.
Low frequency current issues, which are dealt with by the VRMs and their response to load steps.

If you were taking a stab, the bulk caps don't seem a likely cause, more the layout or VRM design. Note that VRMs that have huge current capability often have crap frequency response. More phases of lower capacity are what you want for frequency response.

sHo0sH · 26 Sep 2020 at 08:04

EVGA statement on the matter:

'Recently there has been some discussion about the EVGA GeForce RTX 3080 series.

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.

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.

Also note that we have updated the product pictures at EVGA.com to reflect the production components that shipped to gamers and enthusiasts since day 1 of product launch.
Once you receive the card you can compare for yourself, EVGA stands behind its products!

Thanks
EVGA'

SDK^ · 26 Sep 2020 at 08:05

Kaapstad said:
Even if you do get a faulty card the good news is this is something that will be fixed quickly and any faulty cards replaced.

How are the cards faulty? They still run stable at the required base speed.

Worst case an updated driver is issued to change the OC function to run a bit slower - could be just 15Mhz clock slower.