Are there machines requiring cooling for their wiring?
Are there machines requiring cooling for their wiring?
This isn't an anti Nvidia Sea Lion issue. It's just an engineering question that someone here might help with. The well-known connector Nvidia uses can overheat, so it's advised not to bend the wires excessively. The only situations where wires become flexible are when a lot of heat is applied through them using a soldering gun. At least they should allow for ambient cooling now. We also have devices with fans to manage the actual wires. It has been mentioned here that removing the sheaves is a good idea if the wires come with them.
I've been trying to understand this for months and still can't solve it. The only comparison I can make is tube/valve amplifiers—300 VDC but generally low amps. Their main wire concerns are noise, not excessive heat, and definitely avoid checking wires for warmth in any setup, especially tube amplifiers.
Of course, I don't have much experience with large engineering projects, but I can't think of a design that really needs a fan for the wires due to heat. It seems like industrial applications tend to handle this differently—more robust or perhaps unnecessary. It probably means most other uses would be too weak in this regard. I'm sure I'll pick up some new knowledge about power supplies from you all.
DC fast chargers for EVs usually feature water-cooled wires. Big transformers are frequently placed in mineral oil, serving both as an insulator and a coolant. This approach is also seen in other industrial uses. It isn’t common in consumer electronics due to safety concerns. The primary concern with the 12VHPWR and the newer 6x2 version is the absence of current sensing on the pins to manage load balance. A wider ribbon would have been a better alternative. I prefer the concept of PCB power delivery. One problem lies in the large size of current GPUs combined with older chassis that aren’t wide enough to allow proper connector bending. Connecting directly to the motherboard might have addressed that issue.
The common consensus is that the problem lies with the connector itself, not merely the thin gauge wires. Until now, there have been cases of melted connectors, and the reasons behind this issue remain unclear, as cards with properly seated connectors or high-quality harnesses from trusted PSU manufacturers are still failing.
DC fast chargers for electric vehicles usually feature water-cooled wiring. Big transformers are frequently placed in mineral oil, serving both as an insulator and a cooling medium. This approach is also seen in other industrial uses. Consumer electronics rarely employ this method due to safety concerns.
Most agree that the primary problem with the 12VHPWR and the newer 6x2 version lies in the absence of current sensing on the connectors, which affects load balancing. A wider ribbon would have been a better alternative. The concept of PCB-based power delivery seems appealing. One challenge is the oversized current GPUs combined with chassis that aren’t wide enough to allow proper connector bending. Connecting directly to the motherboard might have resolved this issue.
Because the original 12VHPWR cable has a smaller contact area, it can cause problems if the connector isn’t fully seated. The newer revision greatly increases the contact area, yet it doesn’t address current balancing. Some GPUs and power supplies are adding this feature, but it remains inconsistent.
The main concern is that the connector was rated for 600W with only a 75W safety buffer (12.5%). This pushed GPU manufacturers to push the limit to 600W for cards like the 5090. With such a low margin, even a 10% deviation becomes critical.
In contrast, the 8-pin PCIe connector has a much higher safety margin from a simpler view. The 8-pin EPS is rated for 336W (84W per pin at 18 gauge), and the three 12V pins could handle up to 252W. However, the cable is only rated for 150W, providing a 66% safety buffer. Many power supplies use 16-gauge wire, which can handle even more current. Although 12VHPWR uses 16 gauge, most failures have occurred at the pins.
It would have been better to adopt dual EPS connectors, which were standard for enterprise and data center GPUs. Other ideas include boosting voltage to reduce current, but this would require entirely new power supplies or motherboards with large boost converters.
It was a warning sign for me. Usually upgrades should at least keep up with stock performance, but a lot of different PSU companies' cables are failing. The cable mods cables were also melting, which is concerning. That's a significant issue. In the examples you mentioned, the actual wires would be hard to find if someone was properly trained—no mention for a regular person. But the 4090 and 5090 connectors look fine. For anyone to move, that's a problem. Then there are those connectors that still fumble with 90-degree cables and need extra care to avoid disturbing the wires. It's quite sloppy.
When searching for NorthRidgeFix on YouTube, you'll notice Alex received numerous CableMod adapters and cables, likely in collaboration with GamersNexus. It seems they were investigating the problem together, possibly through reverse engineering, about a year or two ago. Even now, NorthRidgeFix continues to receive cards with damaged connectors.
He points out the AMD GPU's issues as well, but I'm puzzled he doesn't face more severe problems with the 4090 and 5090 models. Eximo really helped me grasp the matter. It was the straightforward access to the problematic wires that made a difference.