rtx 3080 crashing possibly due to capacitor choice

[Bombastinator]

5 minutes ago, Kisai said:

Let the scalpers and miners shoot themselves in the foot.

Only the foot?  

[Vishera]

They crash because the manufacturer cheapened out on components.

Anyway,the Founder's Edition cards seemed more high quality than the AIB cards since the day they were released.

[Bombastinator]

4 minutes ago, Vishera said:

They crash because the manufacturer cheapened out on components.

Anyway,the Founder's Edition cards seemed more high quality than the AIB cards since the day they were released.

FE cards have always been that way. They’re also famous for being an absolute pita to take apart which seems to also continue.

[Delicieuxz]

4 hours ago, sof006 said:

Looking at that table it would suggest a RTX 3080 Gigabyte Eagle card is fine? If so i'm good because thats the card i've pre-ordered.

Jay took the backplate off a Gigabyte 3080 and showed that it has 6 "POSCAP"s. Dunno if all the cards are going to be like that, but if not wanting to take a chance maybe change your order. Currently, the Asus seems to be the only MSRP-priced AIB card that's all in the clear.

 

10:09

 

[Moonzy]

22 hours ago, GoodBytes said:

So the affectedtd AIBs are going BELLOW specs from Nvidia.

i doubt it's below spec

it's probably nvidia not giving out proper design criteria, and manufacturer just goes with whatever is cheapest

 

22 hours ago, GoodBytes said:

The affected manufacture will release a vBIOS which will gimp the GPU abilities to 'solve' the problem.

"gimp", sure if you compare it with current state

but it's still boosting above spec, as long as it's above official spec it's not a gimp but a fix

 

but as a consumer, your other points are very valid, return those cards ASAP, wait for official statement before deciding on what to buy

this incident highlights how AIB cards can be mis-advertised (showing pics with different config than actual product) and even varies from individual cards (some models have different config between itself it seems?)

it's a real shit show this time around.

[spartaman64]

1 hour ago, Moonzy said:

i doubt it's below spec

it's probably nvidia not giving out proper design criteria, and manufacturer just goes with whatever is cheapest

 

"gimp", sure if you compare it with current state

but it's still boosting above spec, as long as it's above official spec it's not a gimp but a fix

 

but as a consumer, your other points are very valid, return those cards ASAP, wait for official statement before deciding on what to buy

this incident highlights how AIB cards can be mis-advertised (showing pics with different config than actual product) and even varies from individual cards (some models have different config between itself it seems?)

it's a real shit show this time around.

whats crazy about this is its only $0.0058 cheaper per card at most if the spcaps cost nothing like wtf

[Deli]

4 minutes ago, spartaman64 said:

whats crazy about this is its only $0.0058 cheaper per card at most if the spcaps cost nothing like wtf

Same as Intel using crappy TIM vs soldered TIM to save 0.05 cent.

[Bombastinator]

1 minute ago, spartaman64 said:

whats crazy about this is its only $0.0058 cheaper per card at most if the spcaps cost nothing like wtf

Apparently it’s the mounting. The devices go in spools, the spools go in machines that can only hold so many kinds of devices, and the machines have to be programmed. Those multilayer ceramic things require 6 for every 1 Spcap.  Things get weird though when dealing with that kind of thing.  It’s not impossible mounting 2 as non spcaps cost as much as doing all 6.  I don’t know.  Savings for doing 6 of all the same chip might have been big.  May come down to the particulars of a given machine, availability of chips in august, etc..

[Bombastinator]

2 minutes ago, Deli said:

Same as Intel using crappy TIM vs soldered TIM to save 0.05 cent.

My personal tinfoil hat theory was that one was actually about slowing the chips. Got an extra generation out of em.  One of the reasons pure capitalism got beat by communism around the turn of the 20th century is it does not reward material innovation while rewarding financial “innovation”

[spartaman64]

 

3 minutes ago, Bombastinator said:

Apparently it’s the mounting. The devices go in spools, the spools go in machines that can only hold so many kinds of devices, and the machines have to be programmed. Those multilayer ceramic things require 6 for every 1 Spcap.  Things get weird though when dealing with that kind of thing.  It’s not impossible mounting 2 as non spcaps cost as much as doing all 6.  I don’t know.  Savings for doing 6 of all the same chip might have been big.  May come down to the particulars of a given machine, availability of chips in august, etc..

it will take them like 1 second longer to produce the card. probably not even a second

[Bombastinator]

4 minutes ago, spartaman64 said:

 

it will take them like 1 second longer to produce the card. probably not even a second

But the machine has to be programmed and a machine can only hold so many spools.  Programming costs money and more machines cost more money.  Also if they couldn’t get parts in august they’re boned.  You can’t get blood from a turnip.

Edited September 26, 2020 by Bombastinator

[spartaman64]

1 minute ago, Bombastinator said:

But the machine has to be programmed and a machine can only hold so many spoils. Programming costs money and more machines cost more money.  Also if they couldn’t get parts in august they’re boned.  You can’t get blood from a turnip.

from what i see programming isnt really an issue it will take a day at most but probably like an hour. and if you dont have enough slots then why not go all MLCC like asus since the price per card will be at most 1.78 cents more with all mlcc again assuming spcaps cost nothing. also the majority of manufacturers manage to have a mix of mlcc and spcaps just fine so I dont really buy that argument.

[Bombastinator]

Just now, spartaman64 said:

from what i see programming isnt really an issue it will take a day at most but probably like an hour. and if you dont have enough slots then why not go all MLCC like asus since the price per card will be at most 1.78 cents more with all mlcc again assuming spcaps cost nothing. also the majority of manufacturers manage to have a mix of mlcc and spcaps just fine so I dont really buy that argument.

If you can’t get MLCCS for that factory it doesn’t matter what they cost.   I’m just presenting possibilities. 

[GoodBytes]

6 minutes ago, Bombastinator said:

If you can’t get MLCCS for that factory it doesn’t matter what they cost.   I’m just presenting possibilities. 

I don't think that was the issue.

I think the AIB went what they are used to do it. I won't be surprised if the similar series 2080 is similar. It wasn't an issue before, most likely (but that is just my guess), is that Nvidia (or AMD) always have a nice head room on chips to really over built things in terms of reliability. This is where we typically come in and be able to nicely overclock our GPUs (or CPUs). Some can't do much, others can really push things. 

 

This time around, the AIBs, I think, got the surprise, that there is no headroom compared to before. The chips are being pushed to the near max (hence why no one seems to be able to do any substantial overclocks). 

 

[GoodBytes]

29 minutes ago, spartaman64 said:

whats crazy about this is its only $0.0058 cheaper per card at most if the spcaps cost nothing like wtf

Well, the reality of things is cost. While Nvidia (especially them) have large profits, AIBs don't.

They tend to make most of their money on the premium products. Due to Nvidia rushing, sadly, premium cards are coming later, much later then the release. So they focused on releasing their entry models just to be on the scene and not miss the boat, as if they don't, their competitor will.. especially following the hype.

 

So it might seem like fraction of pennies, but in the day of the day, like car manufacture that puts the oldest and crapiest SoC they can get in their car infotainment, it is millions of dollars at the end gain as profit... ok well in the case of AIB, maybe half a million dollar at most. But that is a lot of money for them.

 

It just sucks for the customer who wanted specifically the Megabytes GeForce 3080 Oreo's RGB UltraDurable Fatal1ty Xtreme Championship Edition Apple Pie, who saw benchmarks of a certain performance, and won't be able to get it... some other card will now top the chart for the same price as he paid.

[Bombastinator]

3 minutes ago, GoodBytes said:

Well, the reality of things is cost. While Nvidia (especially them) have large profits, AIBs don't.

They tend to make most of their money on the premium products. Due to Nvidia rushing, sadly, premium cards are coming later, much later then the release. So they focused on releasing their entry models just to be on the scene and not miss the boat, as if they don't, their competitor will.. especially following the hype.

 

So it might seem like fraction of pennies, but in the day of the day, like car manufacture that puts the oldest and crapiest SoC they can get in their car infotainment, it is millions of dollars at the end gain as profit... ok well in the case of AIB, maybe half a million dollar at most. But that is a lot of money for them.

 

It just sucks for the customer who wanted specifically the Megabytes GeForce 3080 Oreo's RGB UltraDurable Fatal1ty Xtreme Championship Edition Apple Pie, who saw benchmarks of a certain performance, and won't be able to get it... some other card will now top the chart for the same price as he paid.

There’s another one that I remember Apple ram into.  Local engineers would think they knew better than Apple engineers about what was needed for a product.  Occasionally they were even right.  Sometimes though there wound up being reliability problems.  That’s what happened with the Apple magnetic wall chargers that lawsuit was about.  Apple engineers specced cast silicone cables.  Local engineers thought they could do it with a double layer of plastic hosing because some asshat there didn’t understand how strain arresters work.  Didn’t see fit to inform Apple. The Apple chargers all broke at the strainarrestor point because duh, and apple caught the stinky end of a multi million dollar lawsuit.  They’ve got language now that absolutely requires manufacturers to not alter the spec.

[Vishera]

42 minutes ago, Bombastinator said:

Apparently it’s the mounting. The devices go in spools, the spools go in machines that can only hold so many kinds of devices, and the machines have to be programmed. Those multilayer ceramic things require 6 for every 1 Spcap.  Things get weird though when dealing with that kind of thing.  It’s not impossible mounting 2 as non spcaps cost as much as doing all 6.  I don’t know.  Savings for doing 6 of all the same chip might have been big.  May come down to the particulars of a given machine, availability of chips in august, etc..

EVGA,Asus and Nvidia managed to deliver their cards Spcaps.

It means that it's doable and still profitable,which also means that the rest of the manufacturers are cheap-asses you should consider staying away from.

The AIB partners have shown their true colors with this launch.

[Bombastinator]

2 minutes ago, Vishera said:

EVGA,Asus and Nvidia managed to deliver their cards Spcaps.

It means that it's doable and still profitable,which also means that the rest of the manufacturers are cheap-asses you should consider staying away from.

The AIB partners have shown their true colors with this launch.

Every factory is in a different place, has different supplier lists, different inventory and different manufacturing equipment. There may have also been some cheaping out.  If it turns out that’s what it was I suspect heads will roll. 

[Energycore]

2 hours ago, Delicieuxz said:

showed that it has 6 "POSCAP"s

Just remember that not all SMD Caps are the same. Those could be 220 Farad ones, or 360Farad or 470Farad. Higher is better obviously.

 

6 caps with 470 Farads is significantly better than 6 220s according to buildzoid's video. IIRC the Gigabyte has 6 470s

 

I could be slightly wrong about the numbers

[Delicieuxz]

32 minutes ago, Vishera said:

EVGA,Asus and Nvidia managed to deliver their cards Spcaps.

It means that it's doable and still profitable,which also means that the rest of the manufacturers are cheap-asses you should consider staying away from.

The AIB partners have shown their true colors with this launch.

 

I think you mean they delivered their cards with MLCC caps. The Nvidia has 4 SP-caps and 20 MLCC, the EVGA has 5 SP-caps and 10 MLCC on the lower-tier models and 4 SP-caps and 20 MLCC on the FTW3, and Asus has no SP-caps and 60 MLCC.

 

SP-caps are the larger caps used for the GPU filtering that people were calling POSCAPs. But Buildzoid then pointed-out that POSCAPs are, I think, a specific brand of SP-caps, and not the brand that GPU manufacturers were using.

 

https://industrial.panasonic.com/ww/products/capacitors/polymer-capacitors/sp-cap/lineup

 

21 hours ago, rcmaehl said:

Buildzoid's opinion:

 

[spartaman64]

2 minutes ago, Delicieuxz said:

 

I think you mean they delivered their cards with MLCC caps. The Nvidia has 4 SP-caps and 12 MLCC, the EVGA has 5 SP-caps and 6 MLCC on the lower-tier models and 4 SP-caps and 12 MLCC on the FTW3, and Asus has no SP-caps and 36 MLCC.

 

SP-caps are the larger caps used for the GPU filtering that people were calling POSCAPs. But Buildzoid then pointed-out that POSCAPs are, I think, a specific brand of SP-caps, and not the brand that GPU manufacturers were using.

 

https://industrial.panasonic.com/ww/products/capacitors/polymer-capacitors/sp-cap/lineup

 

minor correction its 10 mlcc per 1 spcap

[Delicieuxz]

2 minutes ago, spartaman64 said:

minor correction its 10 mlcc per 1 spcap

Thanks. Corrected it now in my previous post.

[Bombastinator]

20 minutes ago, Energycore said:

Just remember that not all SMD Caps are the same. Those could be 220 Farad ones, or 360Farad or 470Farad. Higher is better obviously.

 

6 caps with 470 Farads is significantly better than 6 220s according to buildzoid's video. IIRC the Gigabyte has 6 470s

 

I could be slightly wrong about the numbers

Clearly at this point someone needs to name some small pet “farad” just to further confuse the issue.  

[Energycore]

14 minutes ago, Bombastinator said:

Clearly at this point someone needs to name some small pet “farad” just to further confuse the issue.  

My case is a Silverstone Fara R1 doppleganger, that count?

[RejZoR]

I wonder if crashing to desktop is a part of GPU recovery trying to recover a GPU crash or is it something else. Should be easy to check in Event Viewer. I'm wondering if such hardware level fault is still being recovered by GPU recovery (resulting in dropping you on desktop over full on BSOD) or would it fail unconditionally to a point whole system would be gone instead. What I'm essentially wondering if this might be a driver level borkup where GPU shoots up its clock but forgets to properly follow the voltage curve at that point and is not actually really related to the capacitors on the back. Either this or someone sticks an oscilloscope to the back of those capacitors...