This article's now wrong on several counts:

1) It's not a problem with "overclocking", it was Nvidia's boost mechanism.
2) It's not about the "quality" of the capacitors.
3) It's not the AIB's fault.

I don't think anyone can tell what MLCCs used here just by looking at it. It is a joke addressing them the "high quality" or "more expensive" capacitor just because they are MLCC.

Quote from: _MT_ on September 28, 2020, 10:36:09

Quote from: vertigo on September 27, 2020, 19:55:54
It's more like just adding some more cylinders to an engine to make a car faster, at the expense of a significant decrease in gas mileage, vs fine-tuning the engine and attaching a turbocharger to get a similar speed improvement while keeping the gas mileage much closer to before. Yes, it's faster, but at the cost of much higher electricity usage and turning your computer into even more of a space heater.

Actually, a turbocharged engine is going to be less efficient than naturally aspirated engine at higher loads. The reason they can get better fuel economy is that they can be more efficient at lighter loads. Like calm driving. That's down to throttling being very inefficient compared to controlling boost of a turbocharger. At higher loads, the big problem is exhaust gas temperature (in gasoline burning engines). It's so high it can destroy a turbine. And one way to combat it is to enrich the mixture which ruins fuel economy. It's a complex topic as there are quite a few things you can do to an engine to improve efficiency (and many revolve around throttling losses). So, when you put on a turbocharger, you typically improve low load efficiency at the expense of high load efficiency. This can be very easily demonstrated in aviation engines. The designs are old and primitive, but the laws of physics haven't changed.

I was told that turbochargers actually increase fuel economy, at least compared to a N/A engine of similar power. And they typically don't even work very well at lower loads, only activating at higher RPMs. Car and Driver did some fairly extensive testing and found the turbocharged cars generally performed better than their rated mileage while N/A cars actually tended to perform worse, and AFAIK two similar cars with N/A vs turbocharge, the turbocharged car is rated with better mileage. So that would certainly seem to indicate that it does indeed offer better mileage for similar power. Then again, I'm far, far from an expert, so I very well could be mistaken in this understanding, in which case I'd appreciate if you could point me to something showing, preferably with actual testing, that it's not the case.

Either way, while maybe it wasn't the best analogy in terms of accuracy, hopefully it gets the point across, that Ampere's main reason for its improved speeds seems to be simply cranking up the wattage vs actual improvements to the card.

Quote from: vertigo on September 27, 2020, 19:55:54
It's more like just adding some more cylinders to an engine to make a car faster, at the expense of a significant decrease in gas mileage, vs fine-tuning the engine and attaching a turbocharger to get a similar speed improvement while keeping the gas mileage much closer to before. Yes, it's faster, but at the cost of much higher electricity usage and turning your computer into even more of a space heater.

Actually, a turbocharged engine is going to be less efficient than naturally aspirated engine at higher loads. The reason they can get better fuel economy is that they can be more efficient at lighter loads. Like calm driving. That's down to throttling being very inefficient compared to controlling boost of a turbocharger. At higher loads, the big problem is exhaust gas temperature (in gasoline burning engines). It's so high it can destroy a turbine. And one way to combat it is to enrich the mixture which ruins fuel economy. It's a complex topic as there are quite a few things you can do to an engine to improve efficiency (and many revolve around throttling losses). So, when you put on a turbocharger, you typically improve low load efficiency at the expense of high load efficiency. This can be very easily demonstrated in aviation engines. The designs are old and primitive, but the laws of physics haven't changed.

Now we know what brands to avoid.

Quote from: Wambo on September 27, 2020, 23:00:13
In the article there is written that the MSI Trio 3080 has 5 POSCAPS and 1 MLCC, but on the MSI Website both the 3080 Trio X and the 3090 Trio X have 2 MLCC on the image.
Which one is correct?
Also, is it the case that the MLCCs are 100% better than POSCAPS or are there scenarios in which the POSCAPS are better than MLCCs?

In electrical engineering, this question's answer depends on the device's application. Each type of capacitor has its peak effectiveness over a small range of frequencies. Use it outside that frequency, the quality of the design diminishes, without change in the quality of the parts. Here it is pretty clear that the frequency range is likely outside of what polymer caps can do and ceramic bypasses are needed. Not having any ceramic caps very close to the GPU is asking for stability issues, whereas further away from the GPU and closer to the PWM power delivery circuit, the polymer caps are much better. Sourcing 2 different capacitors add extra cost, and I can imagine that some AIBs just decided to use one type everywhere, which is a very common phenomenon in the electronic manufacturing industry. Same kind of substitution is usually done for OPAMPS, diodes, transistors, and just about anything that has a cheaper equivalent or single-part-does-all ability.

POSCAPS are good, they are expensive, but here, they are not the best, although if the AIB used them in the PWM and didn't want to source ceramics, then POSCAPS all through-out.

Quote from: Wambo on September 27, 2020, 23:00:13
Also, is it the case that the MLCCs are 100% better than POSCAPS or are there scenarios in which the POSCAPS are better than MLCCs?

From Tom's Hardware:

"Both types of capacitors have their respective advantages, with MLCC caps being smaller, cheaper, and operate well at high frequencies - but they have temperature limitations and crack more often. Conversely, POSCAPS are more expensive, bigger, have a higher temperature rating, and don't crack as often, but aren't as good at higher frequencies."

In the article there is written that the MSI Trio 3080 has 5 POSCAPS and 1 MLCC, but on the MSI Website both the 3080 Trio X and the 3090 Trio X have 2 MLCC on the image.
Which one is correct?
Also, is it the case that the MLCCs are 100% better than POSCAPS or are there scenarios in which the POSCAPS are better than MLCCs?

I do have the Asus tuf 3080 with the good capacitors and it crashes!!!

Quote from: Spunjji on September 27, 2020, 17:46:13
Been enjoying the way this has been spun in the press.

Whenever someone has a stability issue with an AMD card, it's AMD's fault.

But when a bunch of brand-new Nvidia GPUs on boards designed to Nvidia's own specifications have stability issues? Must be the OEM's fault.

Couldn't possibly be that Nvidia produced a design that used more power than intended in order to hit the clock speeds they wanted, and therefore pushes the circuitry to the limits when overclocked, just like with Fermi... No sir...

Not defending Nvidia, and AMD definitely doesn't seem to get the same treatment as them and Intel a lot of the time, but as I mentioned in my other/previous post and originally brought up by @JohnIL, it's not an issue with stock speeds, only with overclocking. Nvidia's design works fine at stock, but when the OEMs take those design specs and build their own versions, they make their own design decisions which affect, among other things, how well the card overclocks, and some did better than others. This really has nothing to do with Nvidia, and lies solely on the OEMs, some of whom used (presumably) better components and some who used (presumably) lower-quality ones. They all do what they're rated for, but some were better designed by the OEMs and OC better.

I don't follow GPUs much since they're not really that important to me (I'm still using a 770), and frankly I'd rather have one that's a bit less powerful and smaller and more power-efficient (I'd prefer a replacement that's twice as powerful but 2/3 the size and uses 2/3 the power than one that's four times as powerful but is the same size and uses 50% more power), but my takeaway so far from the 3000 series is that while it's a lot faster than the 2000 series, and while a fair amount of that extra power comes from better architecture, it seems a lot of it comes from simply using a lot more power, i.e. it gets its processing capabilities largely through brute force. So for all its benchmarks and records, I'm not really that impressed by it. It's more like just adding some more cylinders to an engine to make a car faster, at the expense of a significant decrease in gas mileage, vs fine-tuning the engine and attaching a turbocharger to get a similar speed improvement while keeping the gas mileage much closer to before. Yes, it's faster, but at the cost of much higher electricity usage and turning your computer into even more of a space heater. It will be really interesting, and perhaps revealing, to see performanc-per-watt figures.

Quote from: JohnIL on September 27, 2020, 11:17:37
Not defending shady quality but over clocking is not something many manufactures guarantee to work properly. If its stable at advertised specifications then the manufacture has provided a acceptable product.. if its proven this power stability issue affects even default settings then I would consider it a defect. Otherwise it's always been assumed you can over clock but it's not guaranteed.

True, but this isn't really saying they're poor quality, though it can certainly come across that way and could be made clearer in that respect. Regardless of what's officially supported and guaranteed, many people, especially the type that are going to spend hundreds to thousands of dollars on video cards to get the very best performance possible, buy the cards with the intent to overclock, in order to squeeze out every last FPS they can, and manufacturers know this. Therefore, it's in the manufacturers' best interest to do what they can, within reason, to ensure their cards overclock well, otherwise users will buy cards from their competition.

In any case, the article isn't saying the cards are poorly designed/built, rather that the 3000 series cards have been found to generally not overclock well, and this may be why. There's nothing wrong with the cards that don't overclock as well as others, and the supposed lower-quality caps aren't necessarily a bad thing and aren't indicative of a low-quality product, but rather (presumably, since it still seems unproven if they are indeed the cause) the use of (again, supposedly) lower-quality component makes the cards slightly less "quality," which is to say they just don't overclock as well, and are therefore less attractive to people intending to overclock, versus saying that they are actually poor quality cards. Like how a Chevy/Ford/Dodge is lower quality than a BMW/Mercedes/Audi, but that doesn't mean they're poor quality, just not as good.

Been enjoying the way this has been spun in the press.

Whenever someone has a stability issue with an AMD card, it's AMD's fault.

But when a bunch of brand-new Nvidia GPUs on boards designed to Nvidia's own specifications have stability issues? Must be the OEM's fault.

Couldn't possibly be that Nvidia produced a design that used more power than intended in order to hit the clock speeds they wanted, and therefore pushes the circuitry to the limits when overclocked, just like with Fermi... No sir...

It is not the quality per se. POSCAPS or other types of polymer capacitors are also expensive and suitable for high frequency applications. In this application, it might be their higher ESL due to the size of their packaging relative to the smaller MLCCs that is causing an issue at high frequency. That is just a guess until this can be measured, but the idea that the capacitors are responsible is confirmed by EVGA, who says all polymer caps in the six positions is not stable.
www.pcgamer.com/evga-confirms-nvidia-rtx-3080-card-capacitors-caused-crashes/

Usually the polymer caps are bypassed with high quality MlCCs, like you would see on motherboards, so I would expect to see both for high frequency applications.

Ceramic capacitors might be very popular among engineers, but AFAIK tantalum capacitors are perfectly appropriate for this application. I don't know what is the current situation, but a couple of years ago, there was a shortage of ceramic capacitors and it was predicted to not improve for years to come. So, designers were encouraged to use alternatives like tantalum more. Being less popular doesn't mean it's not suitable.

If my memory serves me right, this isn't just a question of construction, but also of size. Frequency response of a capacitor is a function of its size. Notice that Nvidia uses two different sizes of tantalums in addition to small ceramics. Zotac uses only one and it's not small. Asus, on the other hand, uses only small capacitors. You can hardly blame this on poor quality of capacitors. It's not like the card died or worked just fine before and not anymore. More like poor engineering. Unless the capacitors are actually failing, I struggle to see it as a problem of quality of capacitors. You could get into the same situation using the best components on the market.

Not to mention that if cards fail at advertised frequencies, what does it say about their testing (during development) and quality assurance? Even if guidance from Nvidia wasn't ideal, they should have discovered the issue. It does happen that datasheets are wrong and things don't work as they're supposed to. That's why it's important to have engineering support.

Not defending shady quality but over clocking is not something many manufactures guarantee to work properly. If its stable at advertised specifications then the manufacture has provided a acceptable product.. if its proven this power stability issue affects even default settings then I would consider it a defect. Otherwise it's always been assumed you can over clock but it's not guaranteed.