I've been pretty happy with the P2 Pro camera I bought a few months ago - so many
Another option is to use a GPU that doesn't send a nonsensically high current through a badly-designed connector
Nvidia drivers 576.02 (better roll back!)
- Thread starter Dragsham
- Start date
Another option is to use a GPU that doesn't send a nonsensically high current through a badly-designed connector
Not "water" but "water mist". Seriously, google it - they are a thingOk a water extinguisher for an electrical fire is about as sensible as removing the hard shoulder from motorways.
If you have a better way of getting 5090 performance I'm all ears
And nope, sorry, I know of no way to get 5090 performance without taking your life in your hands with that godawful connector. It's never gonna be a problem for me personally, because (a) I don't want to spend that much on a card and (b) I don't want a space heater in my bedroom. The way I see it, I am already pushing the boat out pretty far with a 220 W card (which has the same connector btw, but with the lower power, the worst-case heating in the cable is down by more than factor of 7 wrt the 600 W case, so I probably wouldn't melt anything even if one wire took all of the current).
There are two rows of pins, one for ground, one for +12V and failures are almost always due to major current imbalances between terminals on the +12V input (the card has a lot of ground connections through the slot so failures at the ground pins, which carry much less current in practice, are very rare). These current imbalances can in turn be traced back to defects/excess wear at the terminals, the crimp that attaches the terminals to the wires, or user error (too much stress placed on the connector, or failure to insert it fully). Use a known good cable, measure the current through each +12V wire to make sure it's relatively balanced, then leave the power cable/connector alone, and the odds of issues are negligible. A basic clamp ammeter is not an excessive expense for a bit of peace of mind when dealing with something as expensive as a 4090 or 5090.
If one really has an issue with the connector, it's not that hard to replace or supplement, as long as one isn't concerned with warranties. Gigabyte's PCBs have a second set of through holes for another connector (which they use on some of their eGPU models to make them fit in narrow enclosures) that attach to the same power and ground planes as the standard connector. One could easily add a second 12V-2x6 connector here and cut the current load per-pin in half; removing a 12VHPW/12V-2x6 connector is a pain (the PCB is a very good heatsink that makes it hard to reflow the solder without preheating the entire board), but adding a second one is much easier. One could also just replace the row of +12V pins with a single bus bar and then use a much heavier gauge single wire for the +12V input, which would avoid any potential for current imbalance at the card. One could attach a single binding post and run one 10AWG solid copper wire to it, which would safely handle 600W+ by itself at 12V.
Obviously replacing the connector is pretty extreme for most people, but not relative to automatic fire extinguishers or live feeds from IR cameras inside people's cases.
Unless there is real new feature that matters to me or an important safety fix is introduced - I never update drivers, because why? I follow the rule: never change a running, secure and well performing system.
My only disagreement is that clamp meters are just rubbish. I've got several of them, they're are incredibly inconsistent. I've asked others what I was doing wrong and the feedback I get it not to waste my time with them unless I'm using them on high voltage applications where the error is lower. I can get a 150% ( guestimate! ) disparity just by the angle at which the wires go through them. I rarely bother with them now and rely on good old multi meters.
Otherwise, I like your reasoning.
100%. IMO the amount of work that comes with remediating update problems is not worth the effort. If it wasn't for the new card I was very happy on 566.
All this for an overpriced video card
Nvidia should never have created this mess. Most users have no idea what they are doing. This is potentially a dangerous situation that could have been avoided completely.
Note: The advice to measure current.. Or to start soldering on your overpriced card... Wow just wow
Nvidia should never have created this mess. Most users have no idea what they are doing. This is potentially a dangerous situation that could have been avoided completely.
Note: The advice to measure current.. Or to start soldering on your overpriced card... Wow just wow
Edit:
Oops wrong quote
A not updated video card might be very NOT secure...
Urgent New Nvidia Security Warning For 200 Million Linux And Windows Gamers
Gamers love Nvidia graphics cards, and hackers love to compromise systems. Updating your Nvidia security immediately is recommended, as high-severity flaws are confirmed.
www.forbes.com
Last edited:
An accuracy of +/- a few amps is sufficient to see if any wires have major imbalances, or if anything is significantly exceeding specs. I can reliably get this out of a clamp.
Trying to guess current from temperature has huge errors and there is no good way to measure resistance or voltage drop across individual wires in situ with an multimeter when they all connect to the same power plane, without some kind of in-line sensor.
An inline measurement is surely more accurate, but unless it's permanent, it's not very useful for detecting issues that would result from incorrect insertion/mating cycles, as the interface is going to be disturbed when one disconnects the sensor and reconnects the power to the device. And even permanent ones are another pair of interfaces, and thus potential points of failure, themselves.
I'd argue that all non-secondhand video cards are overpriced in the current market.
A high-end video card that does what I want it to do is still a better value than an arbitrarily less expensive card that cannot. No matter how much I've spent, I've spent the least amount practical to meet my particular goals...goals which may be very different from someone else's goals.
Measuring current is fairly non-invasive.
Adding a power connector that will literally cut peak current loads at the connector in half, for less than 1% of the price of the card, is a cheap and effective modification for someone with soldering experience.
None of this should ever be required and neither of these mitigations can reasonably be expected from most end-users, but they're certainly available to me. While I think the issue is rather overblown (probably less than 1% of these cards are going to fail due to issues with power connectors), it costs me nothing to take some measurements. I also quickly void the warranties on most of my parts anyway; spending a trivial extra bit of effort to mitigate connector issues is not far out of my way and is further insurance against failure.
The alternative is to not have a GPU, or refrain from settings, that can pull so much power, or rely on questionable warranty service and spotty supply to fix any issues that may crop up. I want that extra performance and I'm willing to take matters into my own hands.
I follow the same general rule of thumb, which, in practice, leads me to updating to every new version of GPU driver that comes out, at least for my newest GPUs. There are always bugs and security flaws in need of patching, and new features relevant to my uses are almost as common.
For my 3000 and 4000 series parts, the 576.15 hotfix driver is pretty good. Fixes prior instabilities of the 575 branch, and the 575 branch has meaningful performance advantages (even on these generations) when it comes to DLSS. However, these drivers are not bug free (they never are)...in this case, it would appear the efforts to reduce idle power consumption has resulted in the return of an interrupt latency spike during power state transistions. NVIDIA has fixed and broken this every couple of years or so for the last twenty years. It's not a big deal (it mostly just obfuscates other measurments), and doesn't seem to be high priority, but it's annoying enough for me to be ready to try whatever comes out next, on the off chance they have a fix.
Last edited:
Valid point - thankfully I see that particular alert was prior to 566.03 so I'll probably stick with 566.36 for now.A not updated video card might be very NOT secure...
Urgent New Nvidia Security Warning For 200 Million Linux And Windows Gamers
Gamers love Nvidia graphics cards, and hackers love to compromise systems. Updating your Nvidia security immediately is recommended, as high-severity flaws are confirmed.
One other thing I'll add (and the reason I decided to revert to an older driver so quickly) ... I recently had to send my new PC off to be repaired (after stupidly tipping a glass of water into it). Anyway, the damage turned out to be localised to a small USB card which was replaced but, somewhat unsurprisingly, when the PC came back various drivers (including those for the nvidia GPU) had been updated. Anyway, I started noticing very occasional flickering across the top of the screen which I'd never seen before so I was already looking for potential problems when this forum thread popped up. For what it's worth, since rolling back the drivers, I haven't seen the flickering again.
