Handful of users claim new Nvidia GPUs are melting power cables again
-
Most expensive beta test ever for consumer-grade GPUs.
-
The problem seems to be load balancing, or lack thereof. A German guy on YouTube noticed that his cable got up to 150°C at the PSU end, due to one wire delivering 20 or 22 amps, while the others were getting a lot less pumped through them. 22 Ampere is pretty much half the power draw of the card, through one wire instead of three if the load was properly balanced between them. That's why it ran so hot and melted to shit.
-
The 8800 Ultra was 170 watts in 06
The GTX 280 was 230 in 08.
The GTX 480/580 was 250 in 2010. But then we got the GTX 590 dual GPU which more or less doubled
The 680 was a drop, but then they added the TIs/Titans and that brought us back up to high TDP flagships.
These cards have always been high for the time, but quickly that became normalized. Remember when 95 watt CPUs were really high? Yeah that's a joke compared to modern CPUs. My laptops CPU draws 95 watts.
-
High end GPUs are always pushed just past their peak efficiency. If you slightly underclock and undervolt them you can see some incredible performance per watt.
I have a 4090 that's underclocked as low as it will go (0.875v on the core, more or less stock speeds) and it only draws about 250 watts while still providing like 80%+ the performance of the card stock. I had an undervolt that went to about 0.9 or 0.925v on the core with a slight overclock and I got stock speeds at about 300 watts. Heavy RT will make the consumption spike to closer to the 450 watt TDP, but that just puts me back at the same performance as not underclocked because the card was already downclocking to those speeds. About 70 of that 250 watts is my vram so it could scale a bit better if I found the right sweet spot.
My GTX 1080 before that was under volted, but left at maybe 5% less than stock clocks and it went from 180w to 120 or less.
-
And it burns, burns, burns
The gpu of fire
The gpu of fire -
So, when is the first party cable going to be launched?
-
Every picture I've seen has been an outside pin. So my theory is it's the cable getting tugged for cable management and even though it's clipped in, it's not making as good of contact.
That or just a bad cable design. I've bought a few cables from cablemod and I'm not happy with the wiring they used. Their website says "Crafted with 16AWG wiring" but they also brag about the flexibility of their cables so I assume they're using stranded wires instead of a solid core so you lose a decent chunk of ampacity (and heat sinking).
-
https://youtu.be/Ndmoi1s0ZaY?si=SWRjC4AgseXSmOrm
https://youtu.be/kb5YzMoVQyw?si=eSGnUbFv6OJ4jtEg
It's poor design on Nvidias part. There is no load balancing.
-
You probably mean der8auer?
-
Alternate title:
Nvidia SLAMMED as 5090 GPUs start caching on FIRE
-
22 Ampere is pretty much half the power draw of the card, through one wire instead of three if the load was properly balanced between them.
Apparently the reference design only has one current shunt, so they can't even measure the imbalance. Madness.
-
What did it sayyyyyyy
-
-
Use the 12VHPR cable that came with your PSU
-
Well, of course. What can you expect from a Chinese PCB that costs only 2500 USD, right?
-
That IDE connector was for SLI...
-
A handful? Wow that's gotta be like 50% of the 50 series out there!
-
That's a good point and I need to start considering that option in any future upgrades.