When the RX480 was first released, there was an uproar that it was pulling too much power from the PCIE slot. Early articles at the time of release put it up as high as 95w, well over the 75w that PCI-E slots are rated for. AMD claimed that they could fix it through drivers, and it was generally accepted that it was sorted out. For us miners their decision to stick with a single 6 pin PCI power connector is arguably helpful, as it allows us to use a wider range of PSUs- but it does push the draw up from the socket. In mining terms it pushes up the draw where your molex plugs into your PCI-E riser.
Today one of my rigs went down. It was a 3 card RX480 rig, powered by a EVGA 850w power supply. I wasted an hour trying to debug it. The error was weird. It started out as just the rig crashing after a few minutes of mining, but then one by one my graphics cards were disappearing from the device manager. Eventually windows would see none of the GPUs, and not even the fans would spin up on the cards. Until then i’d thought it was a driver error. Now I had to assume a power supply problem, and upon inspecting the PSU I found the modular cable for the molex connections was melted (as pictured) where it plugs into the PSU.
When EVGA made the cable, they used 18AWG wire. 18AWG wire is not the thickest wire around, but in a length of just about 18″, it should be able to to carry 20a at 12v quite happily. That’s 80w per card, which is more than the standard PCI-E port specification. Well apparently not, as my wire melted. So, either AMD is still pulling down way more than 75w, even on the latest drivers, or EVGA screwed up on pin quality within their connector. Probably both. I ended up having to add another PSU into that system to supply some power to the risers, as I didn’t have a replacement cable for the EVGA supply (and actually I needed two cables as a replacement, as another would also have just melted).
Had my risers been the SATA type, those no doubt would have melted too. It’s quite common to see on various forums people talking about SATA power connectors on PCI-E risers melting. Well, this illustrates it. The power draw that gets pulled through the power connectors on the PCI-E risers is actually huge. I don’t have the right equipment here to measure it, but it’s got to be at least the full 75w.
It’s worth remembering that miners run 24/7, and just because something seems to work for a day or a week does not mean it’s not slowly melting away from the inside. This is not the first time i’ve seen a wire melt, almost always at the pin within a molex type connector. So it’s worth remembering that just because the wire is thick enough to carry a given current, does not mean the pins themselves are at the connection points. This is common also on power adaptors for PCI-E 6 and 8 pin > molex type adaptors- those tend to melt.
I think the take away message here is to always over-spec connections where possible, and always spread the load over as many wires as possible. If your PSU has 3 molex outputs, load up all three rather than just using one. It saves time and wire mess to use the one, but in a month when it’s melted you’ll regret it.