Let us be clear. Let us be candid. There is no overclocking of 5090s beyond setting them to ‘600 watt’ mode and maybe tweaking the memory frequencies. That is because they all are using only one power connector and it is maxed – arguably over-maxed – out already. So instead of overclocking let’s look at over-safety-flying it. After all, there are many people, including ourselves, who do have legitimate concerns with the new 16-pin standard. Things have gotten much better since the 12-pin days and unlike the 12-pin or even 16-pin v1 days, it is a relatively safe standard now.
More importantly, the chances of things going wrong with the MSI NVIDIA GeForce RTX 5090 Suprim Liquid SOC are about as likely as saying that tongue-twister of a name ten times fast. In row. At 2 am. In the middle of a pub crawl. Put simply, MSI has mitigated 95% of our concerns over the 16-pin handling 600 watts of power.
With that said, some will still have legitimate concerns. Many of which stem from the fact that a new standard should improve upon an older one it “replaces”. Otherwise… why replace it? While the older PCIe 8-pin was not perfect it however worked, was safe, and was pretty much ID10T proof. With the older 8-pin, you plugged it in until it clicked, you tidied up the cables with little thought over bend radii… and you went about your life. You could do that because it pretty much was a belt and suspenders type standard with impressive amounts of overkill baked right into it. It was however (relatively) large, and yet relatively low power. Thus, when trying to provide 50Amp’s worth of power a lot of them were required. So many that it was starting to impact heatsink design… and the sanity of PCB design teams. Thus a new standard was born, and it was not without its host of controversy.
So let’s break down what these concerns are, and what can be done to alleviate them. First, a bit of background is required. When PCI-SIG designed the older 8-pin power connector they did so around 18AWG, but were extremely conservative in their final rating of 150watts. Which is factually incorrect when stated in watts and where the confusion starts. No standard really gives a fig about ‘watts’. It’s all about the amps baby. Thus the term ‘ampacity’ or the amount of amperage that it can safely, consistently handle is the name of the game. So to be more precise the 8-pin standard is built around three 12v power pathways (3Pos+3Neg+2 sense pins to provide backward compatibility with the 6-pin standard). Each rated for an oddball 4.1666A. At a perfect 12.00000v, that means 150 watts is the rated maximum power draw… and this is where when comparing and contrasting the old with the new standard experienced builders (and insurance actuaries) start to get an eye twitch.
“Maximum” rated power draw is not the same as saying what it was safely built to handle. Instead, it is what it can handle under less-than-ideal conditions with as much safety factor baked in as is realistically possible. The Holy Bible for North American electricians is the NEC code book. It lays out what a power standard should do, should act like, and should be. The National Electrical Code (NEC) calls for a 125% safety factor with a maximum continuous current of 80% or below said standard’s nominal maximum power draw. In layman’s terms, a design team should start with a number in their head on what they want a standard to pull. Then they add in ~20% (though since you do it in reverse it’s a bit more than twenty percent). Then they add in another 25 percent on top of that. Which is a lot of math. So the old-school engineers would start with an insanely overkill assumption and work down from it.
In this instance, the design team took a flat 10A (@12v) as their starting point. Which for 18AWG is extremely conservative (the book rates it for 14A @90c). But 10A with three power circuits per header at 12V means 360 watts. Then they downrated it to 8A to give their new standard an official 125% safety factor on top of the 140% safety factor’s starting point. Thus giving the weird “288 Watt” safe maximum number the standard calls for. Then they downrated it again to 6.25A (~225watts) to be safe even if a muppet of a manager in a Power Supply manufacturer would want to make a 6+8 ‘adapter’ (aka pigtail) cable. Which they all did… and yet this standard can safely handle it. Then, on top of all that they downrated it again by a third. To the actual 4.1666A, it can be expected to consistently handle. Put another the design teams baked in safety factor on top of safety on top of safety… with an always round down mentality for added security. As the spirit of NEC codes calls for.
None of that was done with the 12 and 16-pin standard. While there are true ‘16’ pins, in this new standard it is just an extension of the 12pin with an added four pins added for sense pins. Which are null from a power transfer point of view. So instead of it being an upgrade from 3 to 8 power circuits per header, it is only a doubling from 3 to 6. Yet the allowable safe maximum power draw has not doubled from 24A to 48A. It has increased to 57A. That is because PCI-SIG increased the max amperage from 8A to 9.5A (at 12v) per channel/path/pin/whatever you want to call it.
That is because they started with 12A as their starting assumption point instead of 14A. Then they baked in a 125% safety factor and rounded down. Giving a 9.5A maximum safe max power draw. So far. So good. Conservatism is always good and going above the mandatory minimum 25% is always welcome. Sadly, this is where the standard went off the rails. Most likely because suddenly NVIDIA’s future core designs were calling for major increases in power consumption… or you know… NVIDIA only wanted one header on their high-end cards.
Either way, the standard threw out previous PCI-SIG best practices. So instead of then downgrading (and rounding down) from “684” watts (6x12x9.5A) to the “actual” 7.5A (~540w) as the rated safe maximum power draw they decided that a flat 600Watts was fine… and worked backward to land at 8.33A
Which is not great. But still within most people’s tolerances if they had then applied the standard 80% rule. Sadly, they called it a day here and did not round down again to give a proper 80% continuous use power draw rating of 6.66A. Instead, they simply rounded down to a flat 8A (“576” watts) giving it a fudge factor of a whopping 1 watt to land at the 575 continuous power draw that the x90 cores are allowed in factory stock configuration to pull. On a standard built around 9.5A as the absolute maximum safe power draw. Yeah. Then to add insult to injury they allowed manufacturers to let end-users pull the full 600W. Turning an already redlined ~118.75% safety factor into ~112% one. With no 80 percent rule anywhere in sight. Instead, they are relying on the PCIe slot to give the safety-factory wiggle room via its ’75 watts’ of power.
That is the crux of the issue with this standard. To be precise, it is still safe. Even if the amount a card pulls from the PCIe slot is variable and should never be relied upon… 18AWG is rated for 14A at 90c (aka well below good cable sheath melting thresholds). It just is not as safe or as robust a standard as the one it replaces. A lot of safety steps were sacrificed on the altar of expediency and miniaturization. Even PCI-SIG tacitly admits they cut corners and that is why they added in a mandatory minimum standoff of 35mm from the header before any bending can occur to this new standard. Which on paper is a helpful addition… but not in the real world where ‘cable tidying’ means bending and zip-tying cables as close and ‘neatly’ as possible. Thus melting of headers in the real world can and does happen from time to time… as people are stupid and most should not be playing with electricity with reckless abandon. Especially high amperage DC… which just plumb wants to kill you, your family, your friends, your pets, their pets… and the entire world if given the chance.
So how can we mitigate all of this? First and foremost. MSI has stepped up to give you a go/no-go gauge baked directly into their 16-pin adapter. Use it. This visual indication will easily tell you if it is plugged in safely… or not. That is the second step. The first is don’t buy a video card that does not have a similar safety feature baked into its cable… and make sure it is a cable that uses four PCIe 8-pins (and tells you to not daisy chain them!). Not three. These are all pass/fail equations. The Suprim Liquid OC passes all of them with flying colors.
Next. Let’s put a bit of wiggle room back into the standard. To adhere to the “80 percent rule” of a 8.33A standard one needs to use a 6.66A hard limit. Which is “480” watts. However, there is the PCIe slot and its 75 watts of power to account for. Since it is impossible to know how much is being pulled from that slot… 80 percent of it is “60” watts. For a grand total of “540” watts (7.5A) on a “688” watt (9.5A) standard. So worst case scenario we have a 126% safety factory and at best case 143% safety factory. Still not ‘great’ but a lot safer than what PCI-SIG allowed to be certified.
So what is the easiest way to accomplish this safety factor with the RTX 5090 Suprim Liquid SOC? First slide the BIOS switch to ‘Gaming’. Yes. This will put it in “600W” mode. AKA self-overclocking mode. It however will allow for more aggressive cooling. Which in turn will mitigate some of the performance loss – as the core will overclock higher as it will be running cooler. Next load up MSI Afterburner and slide the power slider down till it says 540 watts (or if you really want to be 100 percent certain… 480W).
Hit save and load on startup.
Boom. Done.
So how much does this additional safety cost you? In real-world gaming scenarios, 540W mode is a rounding error compared to stock “575” watts mode. One maybe two percent. That is all doing NVIDIA’s job for them costs you. We sincerely hope MSI either adds in a third BIOS option… or makes their ‘Silent’ BIOS option be synonymous with ‘Safe’ and makes their 5090s the safest on the market on both the installation and electrical side of the equations. Until then it is not a hard tweak. Yet it is a tweak that will make you safer and give you peace of mind. Which is priceless. Please at least consider doing what we routinely do. I.E. Buy a safety-orientated model and make sure it is set to a safe power limit.
