Foreword
The measurements on the RTX 5090 Founders Edition and the MSI RTX 5090 Suprim show remarkable similarities in the power distribution, which point to a fundamental problem. Both cards exhibit minimal unbalanced loads in terms of power consumption, although these are not caused directly by the cards themselves, but arise somewhere between the power supply unit and the cards. It is interesting to note that the measurements were later repeated with two other different 12VHPWR cables. However, there were hardly any extreme deviations.
The cause of the slight unbalanced loads appears to lie in the combination of cables, connectors and the shared power supply unit, as the cards themselves do not actively distribute the load or balance it on the input side. Certain pins of the 12V2X6 connector are loaded more heavily than others, which leads to a slightly uneven distribution of the currents. These curves can be observed regardless of the choice of cable or card, which indicates that the current distribution is influenced by the mechanical and electrical properties of the overall system or the power supply unit.
The effect occurred on a be quiet! Dark Power Pro 13 with 1600 watts of rated power, whereby the power supply had to be set to OC mode with a switch so that it did not switch off when the spikes of the gaming load came. However, it managed the Furmark in both modes. This is precisely why I use the multi-rail mode for this measurement, as I can supply and measure cables 1 to 3 and 4 to 6 separately. However, these measurements show that the combination of power supply, cable and plug can have a considerable influence on the load distribution and that the weaknesses of the 12VHPWR design become more apparent in practice. Here we go.
Measurements in the visible range of the power supply units (supervisor)
Let’s measure at 10 ms intervals and play with the supervisor chip in the power supply unit. It is precisely these intervals that a supervisor chip still ‘sees’ for monitoring and in which it leads to switch-off if they are exceeded (trigger point). The first measuring clamp covers lines 1 to 3, the second covers lines 4 to 6. While the Founders Edition deviates a full 400 mA at around 590 watts on average over a minute, the peak of the 10 ms intervals is only 1.3 amps. Neither of which is anything to worry about. The original native cable of the power supply unit (be quiet! Dark Power Pro 13 1600 Watt) is used here, which is also connected to the 12V2X6 connector on the power supply unit.
If I go to 10 µs, there is still an average difference of 400 mA in the currents (that’s nothing) and the peaks are again 1.3 amps.
The thermography is also completely unsuspicious. No connector glows even after 30 minutes.
Now I’m measuring the MSI RTX 5090 Suprim with the same cable (but newly plugged in) at 10 ms intervals and I can already see a difference of 1.3 amps in the average value. That’s actually not much either, but a little more. What I have noticed, however, is that the voltage with the card suddenly drops to 11.7 or 11.8 volts.
If you resolve a little higher, it is a maximum of 1.7 amps, which is also not very much on average from 3 cables. But it is more than the first time, even if the currents are now slightly higher overall.
Of course, I checked the IR image again to double-check: nothing to be afraid of:
What can be concluded from this? If something goes wrong, then at most it is the cable and connector. Two plugs, four results? It’s not quite that extreme, but another cable change shows: The values change slightly each time they are plugged in, which indicates the general deficiencies of the plug connection (clamping surface, contact). Added to this is the voltage drop, which also depends on chance.
Ground on the wrong track: Why only certain power connector pins melt and we should rethink PCs
Summary and conclusion
The shortcomings of the 12VHPWR connector, in particular the uneven current distribution through the cable and connector, can cause unbalanced loads where individual pins are loaded more than others. These local overloads lead to increased contact resistance and heat generation, which under certain conditions can cause thermal damage to contacts and cables. In addition, by dispensing with active balancing and splitting the power supply across several rails in the board topology, NVIDIA has itself abandoned possible protective and corrective measures. As the cards directly take over the faulty distribution of the input side, the power load remains uncontrolled, which can lead to escalation under the wrong conditions.
This situation shows how several factors can interact: The inadequate plug connection as a starting point, the resulting thermal issues as a potential symptom, and the lack of protection measures on the board as an untapped opportunity to remedy the situation. Although such problems do not necessarily have to occur, the system remains susceptible to this concatenation if the load and the external conditions coincide unfavorably.
The symptoms of melting contacts and overheated cables in modern GPUs can be explained as a chain of unfortunate circumstances that do not necessarily have to occur. On the contrary, it will probably remain the exception. But it can happen.
377 Antworten
Kommentar
Lade neue Kommentare
Mitglied
1
Mitglied
1
Mitglied
1
Urgestein
Mitglied
1
Urgestein
Mitglied
Veteran
1
Urgestein
Mitglied
Urgestein
Urgestein
Urgestein
Urgestein
Alle Kommentare lesen unter igor´sLAB Community →