NBA 2K27 launched last week, officially bringing NVIDIA’s neural rendering technology to the masses in the form of DLSS 5, but the new feature could cost more than performance hits. DLSS 5 has already been criticized for how it can drastically change images. In some instances, these alterations can be considered improvements, while in others, not so much. In either case, DLSS 5 has already been documented to be a compute-workload-intensive process, which can also halve FPS in games, but lower frames may be the least of your worries as your GPU works overtime.
VideoCardz has reported yet another circumstance in which an RTX 5090 experienced a melted power adapter. This time around, it was said to involve an MSI GeForce RTX 5090 Gaming Trio OC, which was being used to test NBA 2K27 with DLSS5. The user, who goes by erek, reported that the card stopped working during testing, and then they smelled something unusual. GPU-Z was monitoring the card and showed it reaching past 600W to 613.5W total board power. Now, this doesn’t mean that the full load was going through the 12V-2×6 cable, since up to 75W can be delivered via PCIe, but the total is still beyond the max limit for an RTX 5090, and far past the 575W MSI states for this card. The user also reported that this was not a one-off brief spike, as the card had sustained drawing over 600W for a time before it failed.
Tom’s Hardware did its own testing with DLSS 5 and confirmed that not only does the new feature cause increased power draw, but its performance gains are also constrained by power limits. Testing showed an RTX 5090 Founders Edition drew an extra 69W, going from 482W to 551W, when DLSS 5 was enabled. An MSI RTX 5090 Lightning Z, which has 2 of the 12V-2×6 power connectors, along with a factory vBIOS 1000W power limit, was also used in testing. Power draw on it rose by 25% from 580W to 723W, a 143W increase. Tom’s also noted how the Lightning Z saw a slightly smaller performance hit at 39% vs 42%, as witnessed with the Founders Edition, but it is largely debatable if that 3% gain is worth the 170W difference between the two.
While at the moment it cannot be truly proven that DLSS 5 is the culprit for our latest melted power connector, early testing has shown concerning power draw, and given the many, many, many stories regarding the 12VHPWR connector, things are not looking good.
Discussion (19 replies)
Join Discussion →I mean I am not surprised the writing was on fhe walls. Nvidia needs to create drivers with real limits on power consumption that a user can override instead of the limit being the override.
This effin plug is so dumb. Ok fine less wires, make em a good gauge, a fat connector.
Whats the issue with that?
Whats the mental cancer taking over engineers. Just what is it really?
I think, it was a drive toward rushing the standard, to be able to deliver what NVIDIA wanted at the time. It was pushed out too fast to fit a demand, and wasn't ready. Perhaps wasn't even the only design on the table, but just the first one they went with, given a time constraint.
Just my opinion and speculation, of course.
Good ... fast ... cheap. Pick two. They went with fast and cheap, bypassing the good.
Fast ... as in quick to market.
Cheap ... as in lower cost per unit.
It's like it was designed to fail. If they simply rewrote the specifications for PCIEX connector to handle the same power I think we'd see zero issues. All because nvidia wanted a sleek connector on their FE card. At the least they should not have mandated partners to use this crap.
Thats the thing though, theres nothing to be ready, its all already done, plenty of fat good connectors to choose from or derive from.
All the really difficult stuff is well made, the traces, the mosfets of whatever idk, i know this is not an incandecent lightbulb, it is a GPU getting precise power, precisely distributed... And yet, the easiest part, get a fat connector and over spec wires... No, we need to do a dainty little connector, with some little wires ( i mean the wires are ok I guess given proper distribution, but they could be just 2 of good gauge)... Because? Mental rot? What is the point of acting like these nvidia monstrosities dont chug power, they do, act accordingly. Absurd.
crazy stuff. & we have to live with this until the super or 60 series releases !!!
HUB comment — don't do victim shaming
[embedded media]
That idiot Moppel Mat knows the Wireview was created because of the problem right? If the 12VHPWR connector wasn't an issue there would be little to no use for the Wireview hardware.
"after buying a $5090 graphics card, it's your job to babysit it like a toddler with a fork in its mouth"
Has to be a troll, no way someone is this dented.
Well the people that had the failure did catch it and take action. But they also noted it happened.
But there is also some ownus on hardware testers with proper monitoring tools to actually use them.
These connectors were rated for 30 insertions when launched. People running test benches should be swapping to new cables and destroying the old after 30 swaps to stay in spec. I have very real doubt that happens because were all engineers and engineers are by a rule lazy about annoying shit like that even though we know better.
So was the comment bad? No not really. Its not like they went to user a and said.. bad on you for not checking the cable thats been plugged in and in use for the past six months... only ever plugged in once..
They went to people with the tools and the knowledge and presumably resources to properly monitor and use in spec connectors. Part of this failure condition is potentially on them.
I'm pretty sure even the old PCIe connector had a similar limit on insertions. As well as many other connectors.
Right but because testers ignored them then.. doesnt mean you can ignore them now.
Not saying its right. Clearly small connectors distributing a lot of power need to be over built.. but clearly they are NOT. Hence following the spec.
In enterprise compute you always over follow spec. Max seatings on a connector are x. Replace at 2/3rds of x.
Retail consumer hardware needs to follow the spec guidelines for high end hardware.
Thats like running a 5090 and high end system with multiple drives on a 800 watt chinesium power supply that is 80 plus bronze but nowhere to be found on the registry.. it might work... or you might need the fire department.
I mean I don't disagree, but I'd be willing to bet the issues isn't because they're going over that specific spec.