GeForce RTX 3080 CTD Issues likely due to POSCAP and MLCC configuration

WCCFTECH as a reference.....LOL

Seriously man.

 

Since its software limited, It's pretty safe to say that there will be several work arounds in the near future.

Would you rather they limited it with hardware?

 

My point was 2 fold
He knew nothing and had basic reasoning such that even a simple study is better than nothing.
He could have researched himself but as he hadnt I gave him a small leg up.
The article is scarce on detail about how the driver/bios contribute but does cover the situation of card hardware stability, what he didnt seem to realise.
Its a good start for him

 

This whole situation is ridiculous, cards are guaranteed to run at certain clock speeds, so why include a 'boost' that sends the clocks higher automatically without the user wanting / or even knowing that the card is being run ouside of it's warranty specification.
How long has this boost idea been around where 1 clock speed is stated on the box, with the component actually running a much higher speed ?

 

->

It is interesting how people can see the same events or facts in different ways, therefore I am replying to it, it is not meant as a provocation at all, this is just my way to see it.

First Gen of RTX cards had no issues at all, besides the fact that people would have liked bigger differences in performance to the Pascal generation, I can say this as I owned the 2080Ti the first day it came out.
Of course the price was a bit tough but the performance of the 2080Ti was great and is still great if you also consider the overclocking potential of these cards, this brings you to near +-15% to the standard 3080 and is not freezing.. ( joke ! )
So the 2080ti custom or watercooled card has another 2 glorious years to come.

Then again the 3080 or 3090 have mostly too high boosts at the moment , far higher than guaranteed, this can be rectified by a new bios version or driver update and is not a hardware failure or big design issue.
If the boosts were not as high as they are, nobody would even have noticed such a possible issue at this time.
The 3080 and especially the 3090 are really great cards with features AMD will maybe reach when the next gen comes in 2 years.
I have an AMD 3900x proc. so I do like AMD too but for graphics, I think not yet.

I always say buy a new generation, maybe not the first day, but very early ! Why ?

Waiting for prices to drop is bullshit, the longer you wait the shorter the time till the next gen comes out and all starts again. Also the prices do not drop in a way it would change a lot and in the end you are paying for the time you use the card as high end card. If you wait a year till you buy a new generation it is in my opinion, too late, if there is not a major reason.
Also the 2080S, in my opinion an unnecessary card in between the 2000 and 3000 generations.

When you think like the post I am replying too you are one of the guys which are mostly waiting for better prices, better hardware, better games, better.. whatsoever, while waiting you are just getting older !

The best cards, at this time, are in my opinion the ASUS 3080 + 3090 Tuf and ROG ( which have 6 x MLCC´s at the backside ) and the MSI X Trio versions and the FE versions of course.

Of course I do not like the fact that the cards are not in stock for so long, here again lucky people who stick with their 2080Ti ( for high end) at the moment.

Cheers.

 

Cards boost as high as they can while remaining stable, this is different for each level of silicon quality and temperature it will run at.
The algorithm takes account of temperature and power delivery (not exceeding maximum allowed) to determine if its safe to boost higher.
The problem occurred because mfrs werent allowed to test with games, they only had a representation of a gaming load given to them by NVidia before launch.
So the problems only started to come to light when the first driver was released for reviewers. Reviews were delayed for reviewers to receive improved cards.
The day after the reviews were published, sales began and it became clear the problem was larger than anticipated when users had problems too.

Boost clocks going as high as they can safely has been happening for probably 10 years+.
The algorithms werent hardcore until Turing, allowing owners to overclock and get substantial overclocks where silicon quality (binning) allowed.
These latest cards are worse because they are pushed very hard to achieve highest performance possible, note the huge power used, such that there is very little room left to overclock.
But without the ability of mfrs to properly test the cards with games, the algorithms are boosting too far.
The realistic expectation now is cards will be improved where possible and clocks lowered by the driver to make sure all cards are stable.
Because all cards are being pushed near to their max boost already, the overclocking experience isnt going to be very good.
They are however quick anyway its just some of the hardware tweaking fun has been removed.

 

Overclocking is still overclocking its not 100% stable thats with everything really in the pc busines.
As long its fixed with drivers then the problem is solved.

Game A might like a slight OC, while Game B is crashing your pc.

 

I do not think this makes any sense. There are unlocked product and locked product. This is unlocked up to a certain point, above which, instability is demostrated.
You can also probably write a software to override the limit and do your overclocking.

 

No my understand is not closed at all.
And that articled does not saying anything extra that hasn't been said here.

The card was launched few days ago with a wrong boost clock, and now is fixed.
Free boost clock and at will overclock is something that people made up, is not a product characteristic.

If the product launched with a 1900 max boost clock, this would have not happened, and people would have discovered with testing that there isn't much space for tweaking in this chip. Everyone would have been happy about the card results, because the card is still plenty fast compared to the 2080ti and lot cheaper.

Later in time will pop up the lightning editions, the kingpin with better electronics and ln2 mode, and maybe slightly higher clock results in boost.
Probably this launch with no drivers for the partners won't happen again.

 

I'm going to let you live the fantasy, I dont do corrective medicine.

 

I've got the card and i have a 1/5 configuration and not 2/4 like on the product pictures meaning there is already a second revision out (or they just shopped the pictures to dodge the flames from the mob, but I doubt that).

 

Don't pretend to be on another level, because you are not. Here we are talking about a reduced boost clock of a 5% at max and not a 5% performance loss.
Room of overclock is just something you loose when there is more competition on the market.
And i repeat it, easy overclock is something that people are wrongly expecting.

 

You havent repeated anything, you are saying different things.

 

It doesn't seem like there is a loss of performance but I don't care for the argument that overclocking is wrongfully expected. The cards were automatically boosting to these numbers - had it had an impact on performance it would be super misleading to have reviews feature unattainable clocks only to have them backed down later for stability purposes. The phone manufacturers pulled this crap with special frequency boost profiles for reviewers/benchmarks - it caused all kinds of problems with actually knowing the performance of what you're buying. We don't need that here.

In this particular case I don't think Nvidia was trying to pull a fast one with the reviews. Some users actually reporting better performance after the patch because even though the top frequency is less the GPUs are spending more time at higher frequencies then pre-patch. But I do think whenever you run into a situation like this it should be independently checked by reviewers before it enters a stage where Nvidia/AMD are releasing cards that perform 10-15% faster at review time then they do when you buy them. They shouldn't get to hide behind "well technically you're only guaranteed xxxx performance"

 

When I say "fixed in hardware" I mean two things: AIB's testing their cards better before committing to a certain boost clock, and secondly that whole SPCAP and MLCC thing in terms of making sure that's nailed down so they have greater stability at higher clocks...those two points are related, but what I don't like to see is a one-size-fits-all driver "gimping", that's just a band aid to the causes of the problem that I've outlined.

 

Hi,
These are the facts:
yesterday my new card MSI GeForce RTX 3080 Gaming X Trio work perfectly in with 456.38 nVidia driver, no CTD in CoD MW,
today with the new driver 456.55, after 30 sec, in CoD MW, at 60Celsius, CTD pop in, after restarting the game CTD again, after 2 min!
Any thoughts!?

 

I think you're looking at it the wrong way.

Say the driver told the cards to boost to 2.5ghz. Would it still be hardware? What if it was 3ghz? At what point would you say "okay the driver is just boosting too high and the hardware requirements cannot be met" ?

The AIBs perspective is that it already crossed that threshold. Also you're ignoring the fact that the AIBs couldn't test the cards prior to the initial launch driver. So for all we know none of them experienced this issue with Nvidia's test kit software.

 

Well I don't want the new driver to limit overclocking, perhaps I don't know enough about the new driver implementation.....it's not placing an absolute ceiling on the frequency is it? I mean if you overclock with an offset is it artificially limiting it to a max of say 2100Mhz no matter what offset you put in? If the driver is not implementing a ceiling then I suppose I don't mind.....like with the new driver perhaps you just need an extra 13Mhz offset or something to hit the same boost frequencies as the previous driver for example?