That’s not how that works. The card will be power limited at 600W, it doesn’t go above it safe for the millisecond power spikes. There are absolutely no reports of furmark destroying 5090ies and the individual components stay far below their maximum rated temps.
As for the cable even if there’s 20A going through one pin it wouldn’t immediately melt but I would see it in the ASUS software and could act. The point was to check whether the pins will draw stable amps and are within limits. If I had three pins at 11A and then three at just below 7A I would have tried to reseat the connector to fix the contacts. This might not be obvious even at 400-500W as resistance changes with increasing temperature which could increase what might be a slight imbalance at 450W to a problematic one at 600W. The hole point is to stress the connector making sure it’s safe.
I’m using a Black Panther Disney Infinity figure to support my GPU. I wanted to use Spiderman, but his classic spider pose made it too low. I looked at using Sam Flynn, but he was too tall. Black Panther was perfect. I just happened to have some extra figures that I was collecting. Anyway, I initially worried about it melting, but I am pleasantly surprised by how well the figure is working out. Just sharing…
How do you feel about Linpack? It’s pretty much the same thing for processors… I agree with Derby’s report of how he uses Furmark, and I only ever really use it to see if a card needs to be repasted, or that my paste job is good enough. Which is about a 5 minute run to get it to stable max and that’s it (obviously if the paste is not good the run gets aborted )
But I’m not dealing with any cards using 12VHPWR or 12V2X6 if that’s the specific use case you’re referring to?
Yep it’s a test to see whether everything is as it should. Of course there are people running it to hunt high scores but to me that’s pretty pointless. I want top notch VR performance in DCS and I need to know that everything is safe with the connectors. How else do I get the card to its power limit? No game will get it there consistently and any game I have actually stays well below the cards power limit.
Run the benchmark until temps stabilise and make sure the Amps through each pin are stable and within limits and don’t increase or decrease over time which would point to a serious problem with the connection, that’s it. Done. It certainly doesn’t hurt the GPU long term otherwise you’d already have reports about it the way some people OC and bench their cards xD
You’re seeking the answer to a question I would never ask. Goes right with “how many needles can I stick in my eye without affecting my vision?”
And no, I haven’t OC’d a part of my PC since about 2005, maybe farther back than that.
Oh come on. That’s a ridiculous comparison and you know it (If that was meant as a joke then it didn’t come across as one). If you think making sure a potentially bad cable connection is not ruining your expensive hardware is stupid then maybe just don’t comment on it.
If you want a good analogy it’s like getting on a trainer while running an ECG to spot any potential heart problems you might have. In this case getting on a trainer is running 600W through the cable and the ECG is me monitoring the amps through each pin. It’s an entirely reasonable thing to do with a connector like this which is known to melt on occasion when used close to its absolute maximum power which the 5090 definitely does (it’s apparently 675W with the cable rated for 600W which doesn’t leave any reasonable headroom).
This stuff is close enough to my actual job that I know what I am talking about, it’s pretty basic stuff when it comes to electronics. The problem here has nothing to do with over clocking or stressing your GPU beyond reason. It’s a problematic design choice by PCIe Sig and Nvidia insisting on using a single cable. If it was rated for 300-400W and Nvidia used two cables instead of one it wouldn’t be a problem.
My point is that is NOT what I feel Furmark does. I would say instead it’s like going to a doctor when you have a known history of heart problems in your family and he says “ok, run on this trainer with the ECG on while you have this 100lb backpack on and I crank the heat up to 95F and 80% humidity…I’ll be back in 2 hours.”
I do not disagree with making sure it’s working properly using controlled testing. I instead disagree in the strongest possible terms with using Furmark for any purpose whatsoever because I do NOT classify it as “controlled.” There have literally been viruses written to do the same thing in an attempt to destroy the hardware it infects by overwhelming the cooling and voltage regulators.
I do not and never will trust it.
Emphasis mine. Furmark doesn’t do what you think it does. By default you literally can’t run the GPU hard enough to break it, the driver won’t allow it to go past 600W because that is the cables maximum power rating and the power stages are completely overspec’d on these monster GPUs. You’d need to disable the power monitoring for the 5090 Astral to do the stuff you think furmark is capable of on its own and you can’t do that without physically altering the PCB.
The memory doesn’t go past 80°C even after running furmark for 20min and the GPU did get to 72°C peak before settling around 70°C with half the fans limited to 50% RPM. That’s so far below the GPU’s and memory’s temperature limits it’s not even funny. This thing is so overbuilt you’d need to bridge the power monitoring chips and probably run closer to 1000W than 600W through it to get into dangerous territory (if the connector could handle that of course, which it can’t)
The full GB202 die (which the 5090 is a cut down version of) that you find in something like an RTX Pro 6000 is a 875W chip that’s designed to run in huge data centers 24/7 at very intensive workloads. That’s far more demanding than running a synthetic benchmark for 20min every few years.
I managed to get my VR settings dialed in. There’s still some jitter when going low over big cities but otherwise it’s silky smooth 90fps. I went back to a pixel density of 1 because 1.2 would lead to much more jitter in frametimes than I liked and the visual difference is minimal. DLSS preset is set to J and there’s no noticeable ghosting. It’s incredibly good.
It sounds like you’re arguing that the very thing you were testing to prevent happening is not even possible in the first place? If they’re all designed to 130% or so of normal, then I don’t get what the point is of testing 100%?
People’s systems fry just when they’re running them “normally” without stressing them, just because there’s a defect somewhere in the PSU or one of the boards.
Totally healthy people, even athletes, will suddenly drop dead because there was an unknown defect in a valve or something and after years of no symptoms it just fails.
Or to put TLDR: you think the risk is acceptable to push these components to what some chart says should be fine. I do not, because I would rather run at 85% and have a lower failure chance. I don’t want to pay to replace a part, ever, and if you think these companies will honor their warranty because you say you didn’t go too far but the part LOOKS like it was pushed too far…$$ down the drain. The web is littered with stories of people who got their claims denied because it looks like user-induced failure.
Like I said, driving around at the redline in 1st gear as long as you don’t go OVER that redline and the temps stay out of the red doesn’t sound like a good idea, even if according to their engineering team the thing is fine as long as you stay out of the red.
It’s moot, though, as you will never convince me that you’re not overdriving it and apparently I will never convince you that despite what the claims are, sh!t happens.
No, if it was a single conductor then yes, there wouldn’t be anything to worry about but it isn’t. It’s eight 12V conductors and eight ground conductors and your safety margin is roughly 17% which is completely insufficient at these currents. If only one or two pins have slightly higher resistance then you run into problems. The way this works it is entirely possible that you exceed the absolute maximum rating by so much that things start to go wrong, hence the testing. The card isn’t the problem per se, it is the connector that lacks safety margins.
With the PCIe 8-Pin cables you simply don’t run into this problem because they are rated for 150W but as long as every pin makes a proper connection it is completely fine carrying 300W. This is how you spec a cable like this. Now if some pins make a bad connection it doesn’t matter because they are absolutely capable of carrying double the rated current without issue.
You’re talking like Furmark will sit there and test only the cables and connectors you wish like it was a handheld device.
You’re telling me that things I have seen with my own eyes did not happen?
Of course I do not believe you. In what world would I?
)
