A new round of independent testing has put a number on a problem Nvidia has never fully solved: how much heat its flagship graphics card can push through a single power connector. Korean outlet QuasarZone measured a GeForce RTX 5090 Founders Edition pulling 646.7 watts at peak with Nvidia’s new DLSS 5 neural rendering feature switched on, and found the card’s 12V-2×6 power connector hit 91.7°C at its hottest point, according to results reported by Tom’s Hardware on September 25, 2026. That is hotter than many people expect the GPU die itself to run under normal gaming loads, and it reopens a safety conversation that has followed Nvidia’s high-wattage connector design since the RTX 4090 launched in 2022.
The finding matters beyond one Korean lab bench. DLSS 5 is Nvidia’s marquee feature for the RTX 50 generation, and it is rolling out into major titles, starting with NBA 2K27 on September 3, 2026. If a headline rendering feature meaningfully raises sustained power draw on the single most failure-prone component in a modern gaming PC, that is a story with real consequences for buyers, builders, and the cable and power-supply industry that serves them.
What QuasarZone Actually Tested
QuasarZone ran an RTX 5090 Founders Edition through a demanding game workload with Nvidia’s own PCAT (Power Capture Analysis Tool) monitoring power flowing through the 16-pin 12V-2×6 connector, while a thermal camera recorded temperatures on both sides of the plug. The team tested the card twice: once with DLSS 5 disabled, and once with it enabled, isolating the feature’s real-world effect on power and heat rather than relying on Nvidia’s own marketing figures.
With DLSS 5 off, the card peaked at 571.7 watts and the hottest point on the connector reached 80.9°C. With DLSS 5 switched on, peak power climbed to 646.7 watts, an increase of roughly 75 watts, or about 13%, and the connector’s hottest point rose to 91.7°C. Average power draw with DLSS 5 active landed at 582.7 watts, already above the RTX 5090 Founders Edition’s nominal 575-watt board power limit. Club386, summarizing the same data, put it plainly: “Looking at the 12V-2×6 connector, the thermal camera recorded a 13% / 10.8°C temperature increase with DLSS 5 on, reaching 91.7°C at the hottest point compared to 80.9°C with DLSS 5 off.”
RTX 5090 Power Draw: DLSS 5 On vs. Off
The table below lays out QuasarZone’s core measurements side by side. The two connector readings come from opposite sides of the 16-pin plug: the “sense-pin side” runs hotter in both conditions because that is where the connector’s four small sense pins sit alongside the twelve power-carrying pins.
| Metric | DLSS 5 Off | DLSS 5 On | Change |
|---|---|---|---|
| Peak board power (16-pin input) | 571.7W | 646.7W | +75W (+13%) |
| Average power draw | Not separately reported | 582.7W | Exceeds 575W board limit |
| Rated board power limit (RTX 5090 FE) | 575W | 575W | Unchanged spec |
| Connector temp, sense-pin side (hottest) | 80.9°C | 91.7°C | +10.8°C |
| Connector temp, latch side | 77.1°C | 83.4°C | +6.3°C |
Nvidia’s official board power limit for the RTX 5090 Founders Edition is 575 watts. An average draw of 582.7 watts with DLSS 5 running means the card was operating above its own rated ceiling for sustained stretches, not just during brief transient spikes. That distinction matters for cable and connector design, since sustained current is what drives sustained heat, and sustained heat is what degrades plastic housings and solder joints over months of use rather than one gaming session.
Why the Connector Runs Hotter Than the Die
Tom’s Hardware framed the story around a specific irony: on this test, the power connector ran hotter than reviewers typically expect the GPU’s silicon to run under a well-cooled open-air loop. That happens because a GPU die has an enormous, actively-cooled surface area moving heat into a vapor chamber or heat pipes, while a 12V-2×6 connector has to dissipate its heat through a few square centimeters of plastic housing and metal terminals with no dedicated cooling at all. Every watt that a connector burns off as resistive heat, rather than delivering to the GPU, is a watt lost to friction inside a plug the width of a USB-C port.
TweakTown’s writeup of the same test summarized the stakes directly: “On the RTX 5090, peak draw reached 646.7W and the 16-pin connector temperature hit 91.7°C, highlighting increased power-delivery and connector concerns.” Contact resistance is the usual culprit in these cases: if a connector isn’t fully seated, if a pin is slightly bent, or if a cable has been flexed too sharply near the plug, the effective resistance at that contact point rises, and resistance under high current produces heat almost immediately. DLSS 5 doesn’t cause bad seating on its own, but by pushing sustained current higher, it shrinks the safety margin available for any imperfection that was already present in a given build.
A Second Independent Test Points the Same Direction
QuasarZone’s numbers are not an outlier. Tech publication Digit ran its own RTX 5090 comparison around the same NBA 2K27 launch window and found an even sharper jump in GPU-side power: “On our RTX 5090, enabling DLSS 5 pushed average GPU power from 166.6 W to 314.1 W,” according to Digit’s testing. That figure measures a different slice of the power budget than QuasarZone’s board-level reading, likely isolating GPU compute draw in a specific scene rather than total system input, and the two outlets used different games and test conditions. But the direction of travel is identical: DLSS 5’s neural rendering pipeline is measurably more power-hungry than the upscaling techniques it replaces, and it moves the RTX 5090 much closer to, or past, its rated limits.
A third outlet, XenoSpectrum, ran its own headroom test and reached a matching conclusion about where the ceiling sits: “The RTX 5090 Founders Edition (FE) hits its 575W total board power limit when running DLSS 5,” according to XenoSpectrum’s independent test. Three separate labs, three different methodologies, and all three land on the same story: DLSS 5 asks more of the card’s power delivery than Nvidia’s marketing slides about “AI efficiency” would suggest.
A Problem Nvidia Has Never Fully Closed
None of this happens in a vacuum. Nvidia introduced the 12-pin, later 16-pin, high-power connector alongside the RTX 4090 in October 2022, and within weeks builders were posting photos of melted, scorched connectors on forums and Reddit. The root causes were usually a mix of incomplete insertion, tight bends near the plug, and manufacturing tolerances that let contact resistance climb on individual pins. Nvidia and PCI-SIG responded with the revised 12V-2×6 standard for the RTX 50 series launch in January 2025, adding sense-pin changes meant to detect a poor connection before it became dangerous.
That redesign reduced, but did not eliminate, the problem. Cable retailer MODDIY maintains a running public log of 12VHPWR and 12V-2×6 melting cases that continues to add entries through 2026, tracking the cable, connector, GPU, and power supply involved in each incident. Tom’s Hardware itself maintains an explainer on how to safeguard against melting power connectors, recommending full insertion, gentle cable routing, and third-party monitoring hardware as ongoing mitigations, four years after the connector’s original debut.
The Connector Timeline: 2022 to Today
| Date | Event |
|---|---|
| October 2022 | RTX 4090 launches with the original 12VHPWR connector; melting reports surface within weeks |
| 2023 | Nvidia and PCI-SIG study contact-resistance failures; RTX 40 Super refresh ships largely unchanged |
| January 2025 | RTX 50 series launches with the revised 12V-2×6 connector and updated sense-pin design |
| 2025-2026 | MODDIY’s public case log continues to document isolated melting incidents on both connector generations |
| Earlier in 2026 | PNY denies an RTX 5090 warranty claim tied to a melted 12V-2×6 cable, citing improper installation |
| September 25, 2026 | QuasarZone’s DLSS 5 test measures 646.7W and a 91.7°C connector peak on an RTX 5090 FE |
That PNY dispute is worth dwelling on, because it shows how these thermal margins turn into real financial and legal friction. Tech Insider previously reported on PNY denying a warranty claim over a melted 12V-2×6 cable, with the manufacturer pointing to installation error rather than a design flaw. DLSS 5’s higher sustained power draw doesn’t change who is at fault in any individual case, but it does shrink the thermal buffer that used to separate a “fine” installation from a “marginal” one, meaning more borderline builds could tip into failure territory even without any new mistake by the user.
Nvidia’s Silence Is Itself a Data Point
As of this writing, Nvidia has not issued a public statement addressing the QuasarZone findings, the 91.7°C reading, or DLSS 5’s effect on sustained power draw. There has been no recall, no safety advisory, and no acknowledgment from Nvidia that DLSS 5 changes the thermal profile of the connector in a way that requires new guidance to buyers. Neither the Consumer Product Safety Commission, PCI-SIG, nor UL appears to have opened any public proceeding tied to this specific test as of September 29, 2026.
That silence cuts two ways. On one hand, a single 91.7°C reading from one outlet’s test bench is not proof of an imminent failure epidemic, and Nvidia may reasonably be waiting for more data before commenting on a feature that only launched in its first major game 22 days before QuasarZone’s test ran. On the other hand, Nvidia has a four-year track record of downplaying connector concerns until public pressure forces a response, and reviewers who track this beat have grown accustomed to filling that silence with their own safety guidance rather than waiting on Santa Clara.
Why AMD Isn’t Having This Conversation
Nvidia’s approach of concentrating up to 600-plus watts through a single compact 16-pin connector is a deliberate design choice, not an industry standard every GPU maker has adopted. AMD’s current Radeon lineup, including the RX 9070 XT and its high-end cards, continues to rely on standard 8-pin PCIe auxiliary connectors, typically two or three per card on the highest-power models, spreading current across multiple independent connectors rather than routing it all through one point of failure.
| Approach | Nvidia (RTX 5090) | AMD (high-end Radeon) |
|---|---|---|
| Connector type | Single 16-pin 12V-2×6 | Multiple 8-pin PCIe connectors |
| Current concentration | Up to ~650W through one plug | Load split across 2-3 separate plugs |
| Cable count | 1 (plus adapter, if using older PSU) | 2-3 individual cables |
| Documented melting history | Ongoing since 2022 (both connector generations) | No comparable pattern reported |
| Design trade-off | Cleaner cable routing, higher failure concentration risk | More cable clutter, distributed risk |
Neither approach is immune to bad cables or poor seating in principle, but the practical record over four GPU generations shows a clear pattern: Nvidia’s single high-current connector concentrates risk in one place, while AMD’s distributed 8-pin approach has not generated a comparable public log of melting incidents. That’s the trade-off Nvidia accepted for a tidier-looking build, and DLSS 5’s extra wattage is now testing how much margin that trade-off actually had.
Market Impact: RMA Costs, Brand Risk, and the Cooling Ecosystem
The financial stakes here run in several directions at once. For Nvidia, each publicized connector incident feeds a narrative that its flagship card, already selling well above $2,000 in many markets, carries hidden operational risk that competing architectures don’t. For board partners and retailers, connector failures generate warranty and RMA costs that eat into already-thin margins on a card whose bill of materials is dominated by an expensive Blackwell die and GDDR7 memory. For the third-party cooling and monitoring accessory market, the opposite is true: rising anxiety about connector temperatures is direct demand generation. Thermal Grizzly’s WireView product line, including the recently launched WireView Pro II Noctua Edition, exists specifically to give builders per-pin current and temperature readings that neither Nvidia nor most motherboard software provides natively.
There’s also a quieter cost for game performance narratives. DLSS 5 is being sold to buyers as the reason to own an RTX 5090, and reviewers benchmarking titles like the recently remastered Witcher 3 have already been documenting how close the card runs to its limits even before DLSS 5 entered the picture, with one widely cited test showing the RTX 5090 landing at 59 FPS rather than a clean 60 in that title’s most demanding settings. A feature meant to widen the RTX 5090’s performance lead now comes with an asterisk about power delivery that reviewers will have to keep re-testing with every major game that adopts it.
What PSU and Cable Makers Are Doing
No major power supply or cable manufacturer, including Corsair, Seasonic, CableMod, or MODDIY, has issued a new advisory specifically tied to the September 2026 DLSS 5 findings. That’s a meaningful absence: it suggests the manufacturers who build and sell these cables don’t yet see this as a defect requiring a coordinated response, as opposed to a workload pushing an existing design closer to its documented limits. The standing guidance from that industry, dating back to 2022, hasn’t changed: fully seat the connector until it clicks, avoid sharp bends within the first few centimeters of the plug, use the PSU manufacturer’s native cable rather than a generic adapter where possible, and replace any cable that shows discoloration, deformation, or an odor after use.
What has changed is the value proposition of third-party monitoring hardware. A connector that idles at 91.7°C under a popular new feature is a connector worth watching in real time, and that single data point does more to justify a monitoring accessory than four years of general melting headlines managed to do on their own.
How to Protect Your RTX 5090 Right Now
- Fully reseat the 12V-2×6 connector at both the GPU and PSU ends, listening for the latch to click, rather than assuming it’s seated because it looks plugged in.
- Avoid any bend in the cable within roughly 3.5cm of the connector housing on either end, since tight bends near the plug are the single most common precursor to contact-resistance failures.
- Use the power supply’s native, PSU-specific 12V-2×6 cable rather than a generic third-party adapter whenever one is available for your PSU model.
- Install a per-pin monitoring tool, such as a WireView-style adapter, if you plan to run DLSS 5 for extended sessions, since it will flag a developing hot spot before you can smell or see one.
- Check the connector by hand periodically after long DLSS 5 sessions; a warm connector is normal, but one that is uncomfortably hot to hold a finger against for more than a second or two warrants immediate shutdown and inspection.
- Keep an eye on driver and firmware updates, since Nvidia has previously used driver-level power limit adjustments to manage thermal margins on earlier GPU generations without a full recall.
Historical Context: High-Wattage GPUs Have Always Had Growing Pains
This isn’t the first time a flagship consumer GPU has outrun the power delivery infrastructure built around it. The jump from single 8-pin and 6-pin connectors to the 12-pin mini connector on the RTX 3090 Ti in 2022 drew similar skepticism before 12VHPWR replaced it entirely just months later. Each generational wattage jump, from roughly 250W flagship cards a decade ago to the RTX 5090’s 575W rated limit today, has forced a redesign of the connector standard that carries that power, and each redesign has taken one to two product generations to fully stabilize in the field. What’s different this time is that the added load isn’t coming from a new manufacturing node or a bigger die; it’s coming from a software feature that Nvidia can, in theory, patch, tune, or gate behind a power-aware setting without touching the hardware at all.
What Comes Next: Five Predictions
- Nvidia will quietly adjust DLSS 5’s power behavior via driver update rather than issue a public statement, most likely by capping boost clocks slightly during DLSS 5 workloads to keep average power closer to the 575W spec.
- More outlets will replicate the QuasarZone test across additional games as DLSS 5 rolls out beyond NBA 2K27, and expect at least one outlet to report a reading above 95°C on a poorly-seated or budget third-party cable.
- Third-party connector monitoring sales will rise through the rest of 2026, with Thermal Grizzly and similar brands likely to reference DLSS 5 directly in future product marketing.
- AMD will use this cycle in its marketing against the RTX 5090, leaning on its distributed 8-pin approach as a stability argument rather than a raw-performance one, particularly as it pushes its own AI upscaling roadmap.
- Pressure will build on Nvidia’s next flagship connector design, expected alongside the delayed RTX 60 series, which multiple leakers now place in 2028 rather than 2027, giving Nvidia extra time to reconsider whether a single 16-pin connector is still the right call at 600W-plus power targets.
That last point carries some irony: Tech Insider has separately covered how Nvidia’s next-generation RTX 60 series has slipped to a 2028 launch window, meaning the RTX 5090 and its 12V-2×6 connector will likely remain Nvidia’s flagship for longer than any prior generation, giving DLSS 5’s power habits more time to matter in the field rather than being quickly superseded by newer hardware.
Competitive Comparison: Where This Leaves the RTX 5090’s Position
None of this erases the RTX 5090’s performance lead. It remains the fastest consumer GPU on the market by most benchmarks, DLSS 5 included, and no competing card from AMD currently matches its raw output in the workloads that matter most to the buyers willing to spend well over $2,000 on a graphics card. But the calculus for a prospective buyer has shifted slightly: the decision is no longer purely about frame rates versus price, it now includes a power-delivery risk factor that didn’t carry the same weight before DLSS 5 launched. Buyers upgrading a PSU or cable specifically to accommodate the RTX 5090 should treat DLSS 5-enabled workloads as the new worst case for thermal planning, not the DLSS-off baseline that most reviews were built around at launch.
Frequently Asked Questions
What is DLSS 5 and why does it increase power draw?
DLSS 5 is Nvidia’s latest neural rendering technology, using more intensive AI compute on the GPU to generate frames and upscale image quality. That additional compute work draws more power than earlier DLSS versions, which is what independent testing has now measured directly.
Is my RTX 5090 at risk of a melted connector because of DLSS 5?
A properly seated 12V-2×6 connector on a quality cable is not confirmed to be at imminent risk based on a single 91.7°C reading, but that temperature reduces the safety margin available for any existing installation flaw, so a marginal connection becomes more likely to fail than it would have without DLSS 5 running.
Does this affect the RTX 5080 or other RTX 50-series cards?
The 646.7W and 91.7°C figures are specific to the RTX 5090 Founders Edition. No equivalent connector-temperature data for the RTX 5080 or lower-tier RTX 50-series cards has been published as of this writing, since those cards draw substantially less total power to begin with.
What is the 12V-2×6 connector and how is it different from 12VHPWR?
12V-2×6 is the revised 16-pin power connector Nvidia introduced with the RTX 50 series in January 2025, replacing the original 12VHPWR connector launched with the RTX 4090 in 2022. It uses an updated sense-pin arrangement intended to better detect improper seating before it causes a failure.
Has Nvidia responded to the connector temperature findings?
As of September 29, 2026, Nvidia has not issued a public statement addressing the QuasarZone test results, and there has been no recall or formal safety advisory tied to this specific finding.
Should I disable DLSS 5 to protect my GPU?
That’s a personal risk decision. Disabling DLSS 5 will reduce sustained power draw and connector temperature back toward the levels seen before the feature existed, which some risk-averse builders may prefer until more long-term field data is available.
Why doesn’t AMD have this problem with Radeon GPUs?
AMD’s high-end Radeon cards distribute power delivery across multiple standard 8-pin PCIe connectors rather than concentrating it through a single high-current 16-pin plug, which spreads out the current and the associated failure risk rather than eliminating it entirely.
What can I do to monitor my GPU’s connector temperature?
Third-party monitoring accessories, such as Thermal Grizzly’s WireView product line, attach inline with the power connector and report per-pin current and temperature data that isn’t available through standard GPU monitoring software.
