Well, Gigabyte is simply saying that it's a driver related issue. And they are also summarizing their overall design as a "we cut no quality corners" in the manufacture of our RTX 3080. This no doubt sets well with them, but those who are experiencing the problem may not feel that way. 🙁

According to Der8auer the caps only made a 20mhz difference in clocks and the issue seems to be the way the card boosts either from vbios boost table or a driver fault: Problem seems to be fixed with latest driver. Some were boosting to 2.1ghz then crash, these cards were not made to go that far at stock. Video removed, it's been re-posted too much already.

Thunk_It:

Well, Gigabyte is simply saying that it's a driver related issue. And they are also summarizing their overall design as a "we cut no quality corners" in the manufacture of our RTX 3080. This no doubt sets well with them, but those who are experiencing the problem may not feel that way. 🙁

It seems like they are right - the driver seems to have resolved the issue. All these manufacturers are going to have higher binned cards with better power delivery to sell at higher price points, the lower ones were never meant to go as high as they were.

Denial:

the lower ones were never meant to go as high as they were.

That's like some lawyer speech. If they weren't meant to go that high, they wouldn't have gone that high. They weren't clocking so high by accident. It was all defined in the bios/driver. Someone coded it all there. It's no accident. The manufacturers obviously wanted the cards to look good in reviews, no worse than the competitors' offerings. Perhaps you could say they ended up burning their fingertips, but that also means they wrote down the RTX 3080 launch in history as one with a problem that needed a fix. That's just how it goes.

Kaarme:

That's like some lawyer speech. If they weren't meant to go that high, they wouldn't have gone that high. They weren't clocking so high by accident. It was all defined in the bios/driver. Someone coded it all there. It's no accident. The manufacturers obviously wanted the cards to look good in reviews, no worse than the competitors' offerings. Perhaps you could say they ended up burning their fingertips, but that also means they wrote down the RTX 3080 launch in history as one with a problem that needed a fix. That's just how it goes.

But as stated before on this forum, the manufacturers couldn't benchmark the cards in actual gaming workloads - so if they are getting a "pass" on Nvidia's testing software at a given clockspeed - then what are they supposed to do when they actually launch the cards and they are boosting momentarily to unattainable clocks? Further from what I read, it seems like these "unstable" boosts only occur in a few workloads, for like a microsecond before the card actually stabilizes at a much lower frequency for the remainder of the game. I doubt this is effecting benchmark scores to any real metric. Reddit responses post-driver mostly say scores aren't affected. I'd like to see an independent review, obviously - but it doesn't seem like this was intentional.

Denial:

But as stated before on this forum, the manufacturers couldn't benchmark the cards in actual gaming workloads - so if they are getting a "pass" on Nvidia's testing software at a given clockspeed - then what are they supposed to do when they actually launch the cards and they are boosting momentarily to unattainable clocks? Further from what I read, it seems like these "unstable" boosts only occur in a few workloads, for like a microsecond before the card actually stabilizes at a much lower frequency for the remainder of the game. I doubt this is effecting benchmark scores to any real metric. Reddit responses post-driver mostly say scores aren't affected. I'd like to see an independent review, obviously - but it doesn't seem like this was intentional.

One question: shouldn´t Nvidia´s GPU/turbo boost take care of such situations and only boost as long as the cards can remain stable??? Or Nvidia set the boost too high and the new driver corrects this? Or maybe Samsung´s process ir worse than expected and the chips can´t boost that high? Strange stuff that should have been caught in tests but unfortunately we consumers have become the new testers and we even pay for the privilege...:|

Denial:

It seems like they are right - the driver seems to have resolved the issue. All these manufacturers are going to have higher binned cards with better power delivery to sell at higher price points, the lower ones were never meant to go as high as they were.

And yet we had EVGA stating that 6 SP-CAPS weren't stable enough and that they needed 4 SP-CAPS + 20 MLCCs to get the FTW3 cards working well. And 5 SP-CAPS + 10 MLCCs for their XC3 models. So the caps do matter as well as the drivers/BIOS. Like Der8auer said - it's a bit of both.

RavenMaster:

And yet we had EVGA stating that 6 SP-CAPS weren't stable enough and that they needed 4 SP-CAPS + 20 MLCCs to get the FTW3 cards working well. And 5 SP-CAPS + 10 MLCCs for their XC3 models. So the caps do matter as well as the drivers/BIOS. Like Der8auer said - it's a bit of both.

Unless again, all the cards are just being accidently boosted beyond where they are supposed to be by the driver, right? These cards have a range of operating frequencies they are designed to be stable at. Higher end cards like FTW3 are going to need better power delivery to hit those numbers and still be stable. If a driver is causing all cards, including the FTW3 to hit clocks way outside it's range - then obviously shoving more power delivery is going to suddenly fix the problem. Perhaps with the new driver, the 6 SP-CAPS would have been stable enough on the FTW3 and all of this would have been avoided. To me this just like a clear example of the boost frequencies being way too aggressive. Obviously there exists some component combination to make it stable but they shouldn't have ever been that aggressive to begin with. It's Nvidia's fuck up afaic.

RavenMaster:

And yet we had EVGA stating that 6 SP-CAPS weren't stable enough and that they needed 4 SP-CAPS + 20 MLCCs to get the FTW3 cards working well. And 5 SP-CAPS + 10 MLCCs for their XC3 models. So the caps do matter as well as the drivers/BIOS. Like Der8auer said - it's a bit of both.

Caps matter as part of the overall power delivery circuit. It's not as simple as x is bad, y is good. You have to take the whole system into consideration, not just the bit you can see exposed on the PCB.

I'm really hating the trends with boost clocks lately. For Nvidia it's like "our cherry-picked models given to reviewers will achieve these speeds, but nothing else will" For AMD it's like "theoretically, you could achieve these speeds, for a fraction of a second" For Intel it's like "we can't release anything better so let's just keep bumping up the clock speed and throw practicality out the window"

southamptonfc:

Caps matter as part of the overall power delivery circuit. It's not as simple as x is bad, y is good. You have to take the whole system into consideration, not just the bit you can see exposed on the PCB.

Indeed. 4 SP-CAPS + 20 MLCCs seems to be optimum for stability. Makes sense since that's what Nvidia themselves use on their the FE cards. Although, this issue does provide one with an interesting speculation. If Nvidia had gone with TSMC's 7nm node and paid a bit extra, would we be seeing the same instability problems that are present now? I wonder if they will switch to TSMC 7nm for the 3080Ti/Super refresh...

Guys, due to Der8auer's test, just 2 sets of MLCC conbination make 30MHz difference. What if we replace all 6 SP-CAPs? The hardware is been set when it off the streamline, the only thing to make it work properly is to tweak the driver software. If there's a defect in component aspect, software can do noting to make hardware better, the only thing software can do is, adapt its defect. Also, Gigabyte said 'SP- CAPs are not cheaper than MLCCs', that's really interesting. Your engineer chose SP-CAPs with the same cost only to get much worse result. I guess some people will lose their job.

https://cdn.discordapp.com/attachments/594661727057215510/760671295649087578/Screenshot_2020-09-30-11-14-23.png https://cdn.discordapp.com/attachments/594661727057215510/760671296764772444/Screenshot_2020-09-30-11-14-36.png Can everyone say "lip service"?

then what are they supposed to do when they actually launch the cards and they are boosting momentarily to unattainable clocks?

They could start by employing qualified engineers that can actually run numbers and simulate for noise floor concerns as a variable.

Denial:

Unless again, all the cards are just being accidently boosted beyond where they are supposed to be by the driver, right?

You can't boost beyond any point accidentally, all boost steps are programmed into the bios with accompanying voltage step, if its in the cards bios, there is no accident - only a device which cannot operate at the peaks of its programmed capability, ie - Defective.

H83:

Or Nvidia set the boost too high and the new driver corrects this?

Nothing regarding boost apart from a small ns delay in switching between steps is done driver side, the steps are all on the cards vbios. The difference between Gigabytes design and EVGA's, is EVGA's can operate just fine with K-boost enabled, while Gigabytes would fall over itself if it were even compatible with K-boost.

Interesting that Gigabyte comments on the "high-quality, low-ESR 470uF SP-CAP capacitors" on their RTX 3080 Gaming OC and Eagle OC cards but dont mention whats used in their top Aorus cards. They mention the latters better cooling performance, but thats it.

"An Nvidia representative reached out to say the company hasn’t 'done anything to lock the GPU to sub-2 GHz operation with the new driver."

Astyanax:

"An Nvidia representative reached out to say the company hasn’t 'done anything to lock the GPU to sub-2 GHz operation with the new driver."

Did they give any tidbits on what was done to make the cards more stable.

The release notes only say a few titles were optimized for 3000 series stability with the new driver. My assumption is that a few titles now have more stringent boost behavior so they can boost above 2 Ghz and NVIDIA hasn't locked it but it requires more. Similar to AMD boosting the 5700's aggressively with the non-Adrenaline 2020 drivers and then the average went down a bit in later releases but with a few more conditions met the GPU's can still boost just as before. Different systems behind these boost algorithms but similar in results but yeah I don't know any of the specifics here either. EDIT: Comparisons would be needed though with the 456.20 drivers and the 456.55 drivers through GPU-Z or similar maybe use GeForce Experience to up the wattage and power draw and see if that lifts the boost restriction but the tweaks could be more than simply thermals or power. Could even be game specific if there's a particular set of titles that are more finicky and prone to crashing than others and here NVIDIA could have done some other stuff that might not be about boost at all, hard to say without further details on what this actually is or someone more skilled being able to see what's been changed. Guessing it's not as simple as profile comparing via NVIDIA Inspector for some of these games. (The new Call of Duty perhaps?) EDIT:

Nothing regarding boost apart from a small ns delay in switching between steps is done driver side, the steps are all on the cards vbios.

Ah so NVIDIA's is almost entirely bios side, interesting. 🙂 Hmm wonder what was tweaked then, well there's probably ways to handle this in the driver to resolve stability in various ways.

Appreciate the time it took you to type all that.

Yeah but it's kinda invalidated by what Astyanax posted if it's handled deeper from the GPU bios rather than the drivers controlling the behavior so it's something else that's been done here. But without any details on as to what's been changed and there's a lot of stuff some of the display driver overrides can be capable of even for DirectX 12 and Vulkan and the resulting fixes, optimizations or tweaks for stability or game engine issues and more. Hopefully though whatever NVIDIA did tweak does at least improve the stability situation with the current GPU models that do have problems here until newer bios updates can alter the algorithm making it a bit less unstable even if it means stricter criteria for how the GPU boosts and as a result a lower overall clock speed in some systems. 🙂 AMD did that with their AGESA code updates for the Ryzen CPU's as a bios example I am aware of (Mainly the temperature thresholds lowered by 5 - 10 degrees or so Celsius.) though GPU wise it's often been handled by Wattman and Overdrive and their power play API in software which would differ from NVIDIA if theirs is primarily from the bios data on the GPU. (For AMD there were Bios updates to set a higher speed for the 5600 GPU's memory though that's a little bit different.) Some hooks do exist though like what can be done with Afterburner or the new GeForce Experience but NVIDIA directly stated they aren't capping the clock speeds and that probably also means no tweaks to power draw to limit GPU max boosts either. EDIT: Hmm that or NVIDIA or the manufacturer for cards that do have confirmed issues open up for RMA's for refreshed models with hardware fixes but that's costlier than a bios update.

ATi_Radeon:

Guys, due to Der8auer's test, just 2 sets of MLCC conbination make 30MHz difference. What if we replace all 6 SP-CAPs? The hardware is been set when it off the streamline, the only thing to make it work properly is to tweak the driver software. If there's a defect in component aspect, software can do noting to make hardware better, the only thing software can do is, adapt its defect. Also, Gigabyte said 'SP- CAPs are not cheaper than MLCCs', that's really interesting. Your engineer chose SP-CAPs with the same cost only to get much worse result. I guess some people will lose their job.

YOU have asus TUF with 6 mlcc's , do they boost higher ? do they overclock better ? Should be quite simple to test right now.