Rtx 3080 lower quality capacitor Issue

[Rogueywon]

I haven't read anything about the 3090 cards yet...

I've got a Palit GamePro non-OC 3090. Seems Palit did not use the 6xPOSCAPs for their 3090s (rather the 4/2 hybrid arrangement). No crashes at all during gaming so far at stock speeds (I've not tried an overclock). The boost regularly takes it to around 2010MHz during gameplay and I've seen a couple of spikes as high as 2030MHz (Serious Sam 4 in 4k/ultra with an uncapped framerate), but so far rock-solid stable in game.

Still having the weird issue where the graphics driver sometimes crashes and restarts while playing videos in MPC/MPV. Seems to happen if I've jumped around the video a bit, or opened a few videos in rapid succession. Left it to play a few hours of video this morning with no interference while I did other stuff and that was fine.

 

[gbgt2]

The advertised speeds are the manufacturer’s commitment/obligation to the customer. They don’t “owe” you anything more than that. Whatever you get from the GPU boost algorithm is icing on the cake. This is a luxury product and a completely optional purchase. You (as a consumer) decide whether you want to pay the extra money for what has been promised. If you don’t like it don’t buy it. If enough people do this then the manufacturer knows that they need to do better next time.

GPU Boost runs automatically! If the card crashes to desktop because of GPU Boost then obviously that's a problem and the product doesn't work as intended.

Since some cards work fine at over 2Ghz and others crash then it seems some combination of GPU Boost / silicon lottery / caps is causing this that we don't have a root cause for at the moment.

 

[killari]

GPU Boost runs automatically! If the card crashes to desktop because of GPU Boost then obviously that's a problem and the product doesn't work as intended.

Since some cards work fine at over 2Ghz and others crash then it seems some combination of GPU Boost / silicon lottery / caps is causing this that we don't have a root cause for at the moment.

Everyone is rushing to argue that the hardware is at fault. A more reasonable one is that the boost algorithm is the issue and needs to be dialled back. The partner boards have to be approved by nVidia before they go live after all.

 

[Curlyriff]

What I don't get is how some people are accepting that 2000mhz is some sort of magic barrier that cards can not run at.

There are plenty of older cards that have no problem going above 2000mhz.

This need a hardware fix, not excuses.

Why does it? Why would a new card have to go over anything specific. You say about accepting a barrier at 2000Mhz, why would it have to be the other way just because an older card does. The boost clock and overclock headroom changes every generation. If you are getting an OC over the boost of 1710Mhz that what the box says then surely your return varies as per all over launhces were some cards are better than others?

For instance my 980 is a really bad overclocker and wouldn't get anything over the standard boost EVGA sold it with yet others were getting another 300Mhz. How it goes, how it has always been.

 

[gbgt2]

Everyone is rushing to argue that the hardware is at fault. A more reasonable one is that the boost algorithm is the issue and needs to be dialled back. The partner boards have to be approved by nVidia before they go live after all.

Completely agree. You can't blame the AIB manufacturers for following the specs that Nvidia gave them and said would work.

 

[killari]

Why does it? Why would a new card have to go over anything specific. You say about accepting a barrier at 2000Mhz, why would it have to be the other way just because an older card does. The boost clock and overclock headroom changes every generation. If you are getting an OC over the boost of 1710Mhz that what the box says then surely your return varies as per all over launhces were some cards are better than others?

For instance my 980 is a really bad overclocker and wouldn't get anything over the standard boost EVGA sold it with yet others were getting another 300Mhz. How it goes, how it has always been.

Agreed. Apples and oranges. Different architecture, different node. You can’t expect a directly comparable outcome in clock speed.

 

[SeeNoWeevil]

I feel like people are really oversimplifying this issue and the complex interaction between hardware and software. If the software is asking the hardware for something and the hardware can't provide it in a stable way and crashes. Is that a hardware or software issue? Is the hardware lacking or is there a bug in the software asking for too much? Ultimately, it doesn't matter. What if for example, there's a slight error in the voltage tables of the cards that means the software is asking the hardware to push slightly more voltage than it should do and correcting that resolved the instability with NO performance drop at all (we're already seeing evidence of tiny undervolting solving crashes). You could no doubt fix this with a hardware solution (e.g what EVGA did with different caps). But why would you when you can fix it in software with no performance drop? If you're producing an ECU for a car and your code causes a brief unintentional spike in boost pressure that causes turbos to start failing. You don't recall and replace all turbochargers with beefier models to compensate for that unintended spike. You fix the damn bug in your code.

 

[Hakanese]

I feel like people are really oversimplifying this issue and the complex interaction between hardware and software. If the software is asking the hardware for something and the hardware can't provide it in a stable way and crashes. Is that a hardware or software issue? Is the hardware lacking or is there a bug in the software asking for too much? Ultimately, it doesn't matter. What if for example, there's a slight error in the voltage tables of the cards that means the software is asking the hardware to push slightly more voltage than it should do and correcting that resolved the instability with NO performance drop at all (we're already seeing evidence of tiny undervolting solving crashes). You could no doubt fix this with a hardware solution (e.g what EVGA did with different caps). But why would you when you can fix it in software with no performance drop? If you're producing an ECU for a car and your code causes a brief unintentional spike in boost pressure that causes turbos to start failing. You don't recall and replace all turbochargers with beefier models to compensate for that unintended spike. You fix the damn bug in your code.

Indeed

 

[Gamer_3983]

You guys should watch frame chasers video because he explains about this issue properly.

He is the video link: https://m.youtube.com/watch?v=U5pBDEtH5Jk

 

[Hakanese]

You guys should watch frame chasers video because he explains about this issue properly.

He is the video link: https://m.youtube.com/watch?v=U5pBDEtH5Jk

https://youtu.be/--YvuxE0xl4
Latest update from him

 

[VincentHanna]

So you can voltage lock the card to stop the issue occurring and the clock speed stays higher.

 

[Joxeon]

So you can voltage lock the card to stop the issue occurring and the clock speed stays higher.

Kind of although mine would still crash around 2070mhz but this maybe just the silicon limit.

 

[WatchTower]

Looks like Asus TUF it is then. I Noticed there is a regular one and I overclocked one. I assume I may as well get the regular one and push it to the OC speeds with MSI Afterburner and save a little money in the process.

 

[Hakanese]

what may work for one may not work for another, it just seems there's more then just badly configured parts at play. The rushed launch, lack of proper testing, boosted to the limit settings to get the max performance, along with the lack of stock and scalpers, just led up to a "perfect storm". Even if it doesn't boost past 2ghz, I'll still take an 3080 or a big Navi. I'm just glad to have dodged the usual early adopter madness

 

[jokerguv]

Seems like a driver fix should resolve the issue. Not surprising given new hardware.

Though AMD would've been slaughtered for this by now if it were them .

 

[Kaapstad]

Why does it? Why would a new card have to go over anything specific. You say about accepting a barrier at 2000Mhz, why would it have to be the other way just because an older card does. The boost clock and overclock headroom changes every generation. If you are getting an OC over the boost of 1710Mhz that what the box says then surely your return varies as per all over launhces were some cards are better than others?

For instance my 980 is a really bad overclocker and wouldn't get anything over the standard boost EVGA sold it with yet others were getting another 300Mhz. How it goes, how it has always been.

2000mhz is actually a pretty bad show for Ampere, it is on 8nm and also uses physically smaller chips than big Turing or Volta.

 

[mat690]

Looks like Asus TUF it is then. I Noticed there is a regular one and I overclocked one. I assume I may as well get the regular one and push it to the OC speeds with MSI Afterburner and save a little money in the process.

Based on pictures from ASUS's website. The regular TUF card has the problematic capacitors and the OC version doesn't.

 

[Joxeon]

2000mhz is actually a pretty bad show for Ampere, it is on 8nm and also uses physically smaller chips than big Turing or Volta.

The difference between 2000 and 2100 is only like 2fps so it's not a major issue, it's also probably why nvidia went for more cuda as the performance gains from clock speed is limited and also this node don't clock well.

 

[Kaapstad]

The difference between 2000 and 2100 is only like 2fps so it's not a major issue, it's also probably why nvidia went for more cuda as the performance gains from clock speed is limited and also this node don't clock well.

If NVidia had not been greedy they would not have had to use Samsung 8nm and the cards would have been capable of more.

 

[p1on]

The difference between 2000 and 2100 is only like 2fps so it's not a major issue, it's also probably why nvidia went for more cuda as the performance gains from clock speed is limited and also this node don't clock well.

If we're looking at 20fps vs 22 it might seem a bit more