Is the 12VHPWR amperage distribution suitable?
Is the 12VHPWR amperage distribution suitable?
If not made of material, the connector's stability to the wire is crucial. The latest video demonstrates that switching wires frequently causes significant current changes. This implies that wire position affects conductivity. It seems unlikely any part of the card moves when wires are rearranged, suggesting poor crimp quality if wire consistency is assumed.
I quite agree.
If you use the wrong crimp tool or you're clumsy, you can get a really bad crimp connection and in extreme cases, you can break some or all strands of the wire inside the insulation, without it being obvious.
I've done this myself when I used too small a hole in the crimp tool. Over crimping a tag can lead to open circuit connections. It usually becomes obvious because the crimp tag "flops about" and bends easily at right angles.
As I said earlier, I don't think it's the wire itself that causes variations in current, just the disposition of VRMs in the card, plus the quality of crimp connections and how well the pins mate together in the plug/socket shells.
If they'd used pairs of connectors costing tens or hundreds of dollars like the ones I specified at work, most of these problems would disappear.
Instead they cheaped out and specified barely adequate connectors, with very little derating, plus the design fails to monitor over-current in each individual wire and trigger a shut down.
If you fit connectors costing a few bucks and don't include individual wire current sensing, you can expect trouble. Still, that's big business where a few cents saved per card can make thousands of dollars profit on a production run.
No sense in over engineering things as they do in satellites. Most people wouldn't buy GPUs or PSUs if they cost $50 more because they had decent connectors. You gets what you pays for.
It's a shame a $2,000 or $3,000 card is compromised by fitting a pair of $2 or $3 connectors. Another $20 or $50 on a top-end card might seem worth the increase if GPU cable fires disappear.
I'd be happier with a pair of thick heavy 10awg (52A), 8awg (75A) or 6awg (95A) wires (red and black) with crimp tags at each end and mating terminals on GPUs and PSUs. Problem is, you'd need to learn how to use a spanner, but is it trickier than correctly mating a 12V-2X6 pair?
However, that would mean a total redesign and they can't admit they got things wrong. Much better to change 12VHPWR and call it 12V-2X6, but still not address the lack of current monitoring on individual wires.
https://www.club386.com/prototype-12v-2x...n-sensors/
It's a mess of their own making. Penny pinchers.
Hi All, it was definitely the cable. After trying the replacement cable, amp distribution was much better:
I'm sure eliminating the balancing wasn't necessary just to save money. $k/run isn't even a factor for NVIDIA. If such savings were beneficial, they'd have acted earlier when the 4090 was released. The 3090 included balancing.
I'm sure they dropped it because otherwise they'd have faced shutdowns across all levels—designers started with a form factor and wattage, working backward which limited the current to just 10%, or about 9.2 A per pin. The person only noticed the issue during a software test and pin-by-pin current checks. Their cable wasn't melting, but without the previous balancing and monitoring, the Furmark team would have likely disabled the card.
Was the company with the largest market share, the first $5 trillion firm, attempting to cut under $1 per board on their top card that sells for more than most laptops, using a method they'd only recently perfected? Or did they opt for the negative publicity of a smaller percentage of cards failing due to connector misalignment, while risking the reputation damage from their flagship card's frequent failures? I have my opinion.
They relied on customer loyalty to suppress any protest that could stem from a credible media source. Instead, those items were promoted as a divine gift to humanity.
You're likely right. Considering using a poor connector seems reasonable. It might seem less urgent than dealing with fires.
I recall being told to follow standard derating rules for shelter wiring in a specific project, even though we thought the maximum temperatures would be +55°C outside and +70°C inside before the air conditioning started. That was a financial choice.
Luckily, electric vehicle makers don't face the same issue.