Cause research: Capacitor defect on a PNY GeForce RTX 5070 with announcement and prior warning | Update

The PNY RTX 5070 OC is positioned as a classic butter and bread card, it should serve the mid-range segment, deliver solid frame rates and make the red pen of the purchase visible without exceeding the functional limit. It is precisely in this area of tension that the typical component mix of modern mainstream cards is created: the GPU and the main VRMs are equipped with proper polymer banks, while the peripherals and secondary paths are fitted with inexpensive electrolytic capacitors. The cost advantage adds up over dozens of items on the parts list, but the risks end up where thermal and electrical reserves become scarce at the same time.

A bug report recently appeared on Reddit that caused quite a stir: A user described that the cup electrolytic capacitor labeled “C307” on his PNY RTX 5070 OC literally exploded under load. According to him, there were already unstable system states with sporadic crashes and graphics errors until the component finally failed mechanically. Photos in the thread show the burst aluminum cover and remnants of the inner workings on the circuit board, a classic picture of a thermally and electrically overloaded capacitor. It is worth noting that this very component was already the focus of my own investigation, which I carried out on the same card model a few months ago.

Update from 02.09.2025 15:20

A friendly tip from a reader has led me to question the capacitors again, which are not from CapXon but from Apaq and also have a different service life. Accordingly, I have re-performed the underlying calculations for the probability of failure and adjusted the respective text passages. Thanks again to the attentive reader for the correction, because order is a must 🙂

On July 7, 2025, I sent an urgent email to the manufacturer PNY pointing out a serious vulnerability. Specifically, it was about conspicuous hotspots in the area of the voltage converters and the use of questionable capacitors, which are subject to high thermal loads there. As a result, I deliberately suspended the ongoing review of the card to give PNY the opportunity to comment and clarify the issue. Unfortunately, no conclusive and plausible response has been received to date, which does not make it any easier to assess the problem and casts a significant light on how criticism is handled. Here is the news, but it does not yet contain any research into the causes. I will provide that today.

PNY RTX 5070: When the capacitor becomes a time bomb.

Thermographic image

We already know the image shown here and it comes from a calibrated Optris Pi640, which enables a very fine resolution in infrared thermography and thus makes even the smallest temperature differences on the back of the circuit board visible. This is particularly evident in the GPU supply: 104.8 °C is already reached in the “VRM VDDC” area, with the marked hotspot even reaching 107.3 °C. These are values that were not measured on the housings of the power transistors themselves, but directly on the soldering and contacts of the PCB, i.e. on the via bundles and copper rails that connect the phases to the load rail and the capacitors.

These contacts represent the thermal interface through which a significant proportion of the heat loss from the VRM is conducted into the PCB. As copper has a very high thermal conductivity, the power loss is distributed in the layer, but it is concentrated at the points where the current density is highest. The IR image shows precisely this effect: the hottest points are not found at random, but where the busbar of the GPU supply converges and the vias establish the connection between the front and rear.

For the adjacent components, especially the electrolytic capacitors used, this is a problematic situation. Their negative and positive pads sit directly on these strongly heated copper surfaces and are therefore permanently heated from below. Even if the ambient air temperature is lower, the heat conduction through the pads and the contacts drives the core temperature of the capacitor significantly upward. Applying the rule of thumb for 105 °C, every further increase in core temperature by 10 Kelvin means a halving of the service life. I will come back to this in a moment.

The thermography thus clearly shows that the capacitors in this zone are not only stressed by ripple currents inside, but also by the direct thermal coupling to a contact surface that is over 100 °C hot. In practice, the capacitor heats up faster than intended, the ESR increases as it ages, which further increases the self-heating – a classic vicious circle that only accelerates failure. The fact that one of these capacitors (C307) ultimately failed mechanically in the Reddit case is therefore the logical consequence of a continuous thermal load that far exceeds the specification of the component.

Circuit board analysis and “predetermined breaking point”

This pattern is clearly visible on this circuit board. Along the phases are the gray coils and a chain of capacitors with a low operating voltage, which are designed for the GPU rail. In between and near the output sockets and auxiliary voltages, there are further cup capacitors from the manufacturer Apaq, all recognizable by the circle logo and the half-black cover marking. The printed indications 100 and 16 V refer to 100 microfarads at 16 volts, the internal print code 5KE44 is used for traceability. The halved black segment area on the lid is a series marker, here it assigns the parts to the AR5K series. This series is a long-term capacitor for 105 degrees, the data sheets typically state 5000 hours test life at 105 degrees, moderate ESR and ripple current in the low to medium range.

Technically, this means that the capacitor meets the usual requirements of simple secondary paths when new, but can age under high ripple and high ambient temperature because the power dissipation in the ESR additionally heats the can and degrades the polymer inside. Polymer parts generally tolerate much higher ripple currents at a much lower ESR, so they are much less prone to self-heating. The capacitor identified here is a cost compromise that becomes critical in the vicinity of the GPU supply or heavily loaded auxiliary rails if the maximum temperatures are exceeded.

The IR measurement image of the back of the card illustrates the thermal situation. Up to 107.3 degrees are shown at the rear hotspot, the VDDC supply region is around 104.8 degrees, the area under the GPU socket is around 70 degrees and the memory modules show values between around 59 and 74 degrees. Experience has shown that the actual housing temperature of a component on the front is several Kelvin higher in an identical location, as copper layers and via fields draw the heat loss of the MOSFETs and chokes into the immediate vicinity.

The layout makes a decisive contribution to this. Wide busbars and via bundles connect the phases with busbars and capacitor arrays, so the losses of the power semiconductors are not distributed over a wide area but are concentrated along these conductive structures. If the backplate is more ornamental than a heat sink without thermal pads above the VRMs, the heat builds up in the exact area where the capacitors are located. The result is a local temperature plateau that simultaneously heats the capacitors from the outside and heats them from the inside due to increased ripple. The documented Reddit case of an exploded C307 fits this pattern, gas formation due to electrolyte degradation drives the internal pressure, the predetermined breaking notch in the lid opens, in the worst case with an abrupt event. this is exactly what I had already discussed months ago (link below)

The common temperature halving rule can be used to calculate the service life. The data sheet service life L0 is 5000 hours at a reference temperature of 105 degrees. The expected service life L at a core temperature Tcore is approximately calculated as L equals L0 times 2 to the power of (T0 minus Tcore) divided by 10. The core temperature in the component is decisive, not the back surface temperature. If one conservatively assumes that 107 degrees on the back correspond to a housing temperature at the capacitor of around 112 to 120 degrees on the front, and also takes into account a few Kelvin of self-heating due to ripple, then sensible calculation points are 110, 115 and 120 degrees. This results in around 3525 hours at 110 degrees, 2500 hours at 115 degrees and around 1775 hours at 120 degrees. With four hours of gaming per day, this corresponds to around 881 days to 3525 hours, 625 days to 2500 hours and 444 days to 1775 hours. These average values alone show that although some of the cards will exceed the twelve-month mark, the thermal load still significantly reduces the effective service life if the card is regularly operated at the limit.

In order to make the probability of failure more tangible, an exponential distribution can be derived from the expected values determined above, which is not an exact representation of the test service life, but provides a conservative orientation. For a year of use with 1460 hours in full load mode, the probability of failure is around 34 percent at 110 degrees, around 44 percent at 115 degrees and around 56 percent at 120 degrees, in each case in relation to the individual capacitor at the hot spot. If the core temperature shifts to 125 degrees, which is possible in installation cases without draughts and with a high ripple load, the calculated average service life is around 1250 hours and the one-year failure probability is around 69 percent. It should be noted that capacitors rarely work alone, several parallel components share the ripple, but a single aged capacitor with a greatly increased ESR is sufficient to cause the local ripple to increase, which places additional stress on the neighbors, so that the first failure can accelerate the cascade.

The technical cause is therefore to be found in the interaction of component selection, layout and cooling. The capacitors from a low-cost supplier, which guarantee 5000 hours at 105 degrees, are placed in a zone that reaches well over 100 degrees on the front and rear under real load. The via fields and copper traces lead the heat exactly there, the backplate decouples thermally rather than dissipating it, and there is no targeted airflow on the rear. This is understandable for a butter and bread card, but a risk for the service life of the wet cup electrolytic capacitors. Against this background, the documented Reddit failure is not an outlier, but a plausible consequence. Anyone who regularly moves such cards for several hours a day under high load must expect a short time until the first failure if the condition remains unchanged. From a purely technical point of view, conductive polymer types or a relocation of the electrolytic capacitors from the hotspot, combined with a thermally conductive backplate and a slight airflow on the back, would be effective countermeasures, they would lower the core temperature and shift the calculation significantly in favor of the service life.

Conclusion

The conclusion is sobering. The thermography impressively demonstrates that the voltage converter areas of the PNY RTX 5070 OC reach temperatures of over 100 °C on the back of the PCB and even exceed this at individual contacts. Although these values are not critical for the DrMOS components used in the short term, as they can tolerate 150 °C according to the specification, they pose a considerable risk for the surrounding components. The capacitors in particular suffer considerably in such an environment. With a rated service life of only 5000 hours at 105 °C, the calculated durability in real-life gaming is reduced to a few months before the first failures inevitably occur. The documented case on Reddit should therefore not be seen as an isolated incident, but rather as a direct consequence of a design weakness that is likely to be systematically present throughout the series.

Added to this is the manufacturer’s handling of this problem. Back in May, I sent PNY detailed ASTM measurements of the heat conducting materials used, which proved the thermal load in the hotspot zones. The only response was the terse remark that the DrMOS components could withstand temperatures of up to 150 °C without any problems. I received no further response to my explicit warning that it was not the DrMOS that were the bottleneck, but the surrounding passive components such as capacitors. Incidentally, I have optimized my specimen:

This shows a worrying pattern: instead of looking at the overall picture and also taking the thermal sensitivity of secondary components seriously, the manufacturer relies on isolated specifications for individual components. For the end customer, this means that the service life of a card that was actually intended as a reliable bread-and-butter solution is considerably reduced by avoidable cost-cutting measures. The risk of premature failure is therefore not theoretical, but can be proven in practice – both by my measurements and by documented cases of damage in practice.

Local hotspots on RTX-5000 cards: When board layout and cooling design don’t work together and a pad mod has to help