We have witnessed a very complicated start of the NVIDIA RTX 3000. To the availability problems have been added problems in the frequencies, where the cards hung the PC for many users. The culprits have been established as POSCAP vs MLCC , where each manufacturer has used each of these types of capacitors with different results. Is it safe to buy an RTX 3000 today? What should we consider?
A dance of statements from each manufacturer has only caused panic, when some like Zotac called for a review of their models put on the street, EVGA delayed its KINGPIN and FTW models for detecting problems at speeds above 2 GHz and ASUS was ahead of the game, spotting these problems and switching capacitors before launching its GPUs on the market.
Others have chosen to mix these capacitors from the start and claim to report no problems. Faced with this chaos, can we buy any model or are there preferences?
POSCAP vs MLCC, a war to nowhere
The first thing that we must be clear about is that we are talking about two different concepts when addressing this problem: frequency and current. The frequency as a value in these RTX 3000 begins to be a problem after 2 GHz, where the second factor comes into play, the so-called current.
Current as a basic concept is nothing more than a sum of electrons per second, and as we know it is measured in amperes, but GPUs do not work with amperes, but with volts, so here everything gets complicated and we have to make a brief reflection to get a little deeper into the POSCAP vs MLCC war.
The PSU converts our 240 volts from home into amps, so that then the VRMs on the motherboards or GPUs in this case, transform those amps into volts. This is easy to understand, what is not so much is the fact that both the PSU and the VRM are, compared to the rest of the components of a GPU, very slow in the change of situation.
That is, they take a long time to correct any issues versus, for example, the time it takes for a GPU to switch frequencies on its SKU. To alleviate this time GAP, capacitors are used throughout the card, but especially after the PCB where the soldered GPU goes, since a lot of regulation is needed at that particular point.
Time differences and how AIBs fix it
With the current consumptions NVIDIA has had to work hard to ensure that the voltage fluctuations occur on time. Depending on the capacitors / capacitors that each custom model has, more or less voltage variations per second can be made and this is where the dispute between POSCAP vs MLCC comes in.
The problem on the RTX 3080 and RTX 3090 is the time those capacitors give them when the frequency goes up. At a higher frequency, a longer time is needed for the voltage not to drop, because the interactions and changes of this frequency are carried out in nanoseconds. These frequency jumps take much less time than the VRMs or the PSU itself can react, so voltage corrections will come late if you don’t get a fix on all of this.
In the absence of capacitors / capacitors, a voltage drop due to a very rapid change in frequency would result in the PSU detecting said drop late and sending more amps to the VRMs, which would transform them into volts, putting a “kick” of extra voltage to the GPU when you don’t need it. This used to happen more often and therefore many GPUs ended up burning.
The solution was found by adding extra capacitors and capacitors the closer to the GPU the better, in this case of the RTX 3080 there are two versions of these, POSCAP vs MLCC, and hence the rest of the debate.
Which are best suited for Ampere GPUs?
Reduce the problem to the so-called POSCAP vs MLCC is to simplify a lot and we explain ourselves. The problem is not that there is one more correct choice than another, the problem is that there are almost 1 million MLCC models on the market and more than 10,000 in POSCAP.
Seeing these numbers and understanding that the differences between the two simply refer to the working thresholds and how they respond to frequency and voltage jumps in nanoseconds, the answer is impossible to give. That is to say, the options are enormous, there are manufacturers that have chosen to mix POSCAP and MLCC, others opt for the NVIDIA reference design in this regard and others have said publicly that they have noticed improvement with MLCC exclusively.
NOBODY has the answer to this “problem”, because it requires a study, trials and errors with equipment that can perfectly cost a home and also, we need a team of electronic engineers to monitor for weeks all possible cases and falls that could occur .
There is no solution?
We have not said that either, but it is difficult to be blunt on such a complex issue. What we are going to do is use common sense, which is often the least of the senses … We have all witnessed the fact that manufacturers have taken the issue very seriously, because reporting sales below expectations is not something that nobody can afford.
But it is true that there have been many silent changes in this POSCAP vs MLCC, where some AIBs have changed from the first to the second, even updating their official images so that users can see these changes reflected on the official websites of each model .
So, when you see that many manufacturers jump straight to MLCC having previously made designs with POSCAP or mixed with both, it means that the internal tests they have carried out in a dizzying way give a presumed winner. And we must emphasize the assumption, since there is no way to check it if it is not in the way we have said above.
By the way, NVIDIA came to the fore with a new driver that calmed the waters, providing, in the company’s words, “greater stability.” As we saw, what the new driver does is to be less aggressive in the frequency / voltage curve, so the energy requirements change and it is possible that the times have been lengthened in the capacitors, allowing them to react in time and avoiding hangup. from the PC.
Do we lose performance or security with it?
What is shown indicates that the loss of performance is less than 1% on average, specifically we are talking about 0.7% and in some cases 0.6%, a GAP so small that it falls within the error of any standard measurement.
In terms of safety, it has undoubtedly won, especially and seen what has been seen after the change in the designs of the PCBs of the manufacturers, where more and more models choose either MLCC or a mixed between POSCAP and these. So is it safe to buy an RTX 3000 today?
Yes, it is safe, but it is because of the modification of the driver curve provided by NVIDIA and that implies more obstacles to the GPU to reach or exceed 2 GHz. This means that if the GPU you are going to buy has a design ” old “or based on POSCAP, you may have to forget to tighten it under overclock.
If this is not a problem for you, we can make the purchase with total security, since it is not ruled out that manufacturers launch BIOS Upgrade with new Boost tables to definitively solve any possible problem, even with overclocking.
It must be clarified that this does not seem to happen to the RTX 3070, mainly due to the difference in consumption between the three main NVIDIA graphics cards, since the higher the consumption, the greater the relative problems we will suffer on paper.
