Nvidia DLSS 5 Power Draw Hits 647W as the RTX 5090 Connector Runs Hotter Than Its GPU

Nvidia DLSS 5 power draw pushed an RTX 5090 Founders Edition connector to 91.7°C, while its cable input briefly reached 646.7W. The GPU core peaked at 86.3°C during the same test. That reversal puts the hottest measured point outside the chip that performs the work.

Korean hardware outlet QuasarZone recorded the result while running Cyberpunk 2077 at 4K for 20 minutes. The test used the Ultra preset, DLSS Quality mode, and a controlled ambient temperature of 27.5°C. It compared the RTX 5090 Founders Edition with the substantially lower-powered RTX 5080 Founders Edition.

The measurements do not establish that DLSS 5 damages power connectors. They show that a demanding neural-rendering workload can keep Nvidia’s flagship card near its power ceiling. That sustained load leaves less thermal and electrical margin at its single 12V-2x6 input.

This distinction matters because the RTX 5090 connector was already under scrutiny before DLSS 5 arrived. Independent testers have recorded high cable temperatures under rendering, gaming, and synthetic workloads. Some owners have also reported damaged plugs, although individual reports cannot establish a general failure rate.

The central conflict is therefore not DLSS 5 against conventional rendering. It is Nvidia’s expanding visual ambition against a power-delivery system operating close to its practical limits. The new software exposes that tension more clearly because its neural workload can maintain unusually high utilization.

The Nvidia DLSS 5 Power Draw Test Reached the Connector’s Edge

The most important measurement was not a momentary board-power spike. It was the sustained load passing through one compact connector.

QuasarZone used two measurement paths. Software telemetry reported the card’s internal power estimate, while Nvidia’s Power Capture Analysis Tool measured power entering through the slot and external cable. The second method helped separate total board consumption from the load carried by the 16-pin connection.

With DLSS 5 disabled, the RTX 5090’s software-reported average consumption was 509W. Enabling it raised that figure to 575W, matching Nvidia’s published total graphics power rating for the Founders Edition.

The external measurements were slightly higher. Total board power averaged 593.9W and peaked at 658W. After excluding power delivered through the motherboard slot, the 12V-2x6 input averaged 582.7W and briefly reached 646.7W.

That distinction prevents a common misunderstanding. A 658W board reading does not mean all 658W traveled through the cable. However, the measured cable peak still exceeded the connector’s familiar 600W nominal rating.

A short transient does not carry the same thermal meaning as a continuous 647W load. Temperature follows sustained current, resistance, cooling, and time. The more consequential result was the cable averaging nearly 583W during the test.

Nvidia lists the RTX 5090 Founders Edition at 575W total power. It specifies either a compatible PCIe Gen 5 cable or the included multi-cable adapter. Nvidia also recommends a high-capacity system power supply, with requirements varying by configuration.

The RTX 5080 offered a useful control. Its reported consumption rose from 298W to 350W after DLSS 5 was enabled. Yet its connector reached only 52.2°C, far below the RTX 5090 result.

That comparison points toward load density, not the DLSS label alone. Both GPUs ran the same class of neural-rendering workload. The card drawing roughly 300W through its connector retained far more thermal headroom.

The original connector test therefore raises a narrower concern than its headline suggests. DLSS 5 did not make every tested connector dangerously hot. It placed the RTX 5090 near a boundary that its lower-powered sibling never approached.

The test also used one card, one workload, and one controlled setup. Those conditions make the comparison informative, but they do not describe every RTX 5090 installation. Cable construction, connector seating, case airflow, ambient temperature, and manufacturing variation can all change the outcome.

What changed is measurable: DLSS 5 converted spare compute capacity into sustained work. On the RTX 5090, that work pushed external power delivery close to its nominal ceiling. The next question is why the connector became hotter than the silicon.

Why the RTX 5090 Connector Hit 91.7°C

A connector becomes hot when high current meets resistance, and even a small imbalance can concentrate that heat in one contact.

The 12V-2x6 connector uses six power contacts and six corresponding ground contacts. Four smaller sense contacts communicate connection and power information. Together, they form the 16-pin interface commonly discussed around recent high-end graphics cards.

Electrical heating rises with the square of current. This relationship means modest increases can produce disproportionate heat when contact resistance remains unchanged. A loose fit, imperfect crimp, contaminated surface, or uneven current distribution can raise local resistance further.

QuasarZone measured both major faces of the RTX 5090 connector with a thermal camera. With DLSS 5 disabled, the sense-pin side reached 80.9°C. The opposite, latch-side surface reached 77.1°C.

After enabling DLSS 5, those readings increased to 91.7°C and 83.4°C. The hotter side gained 10.8°C under the tested conditions. That rise accompanied the card’s movement from an already high workload toward its power limit.

The GPU’s average temperature moved from 79.4°C to 83°C. Its recorded peak reached 86.3°C, leaving the connector’s hottest surface 5.4°C higher. The result does not mean the connector produced more total heat than the GPU. It identifies a hotter localized measurement point.

The location matters because the GPU die has a large cooling system built around it. Heat pipes, vapor chambers, heatsinks, and fans move energy away from the silicon. The connector relies more heavily on its metal contacts, attached wiring, nearby airflow, and surrounding material.

Independent testing has already shown that connector cooling can behave counterintuitively. Hardware Busters found that an RTX 5090 cable could become hotter under a lighter rendering workload than under a heavier benchmark.

Its proposed explanation involved fan speed. A severe GPU workload makes the card’s fans spin faster, increasing airflow around the connector. A somewhat lighter load can still move substantial current while producing less local cooling.

In one load-balancing analysis, the connector reached 64°C at a 21°C ambient temperature. The tester warned that a much hotter case interior could move the same setup closer to 90°C.

QuasarZone’s room was already warmer at 27.5°C. However, room temperature is not the same as the air surrounding a connector inside a closed case. A graphics card, processor, radiator, and restricted exhaust path can increase local intake temperatures.

This makes the 91.7°C result relevant without making it universal. An open test bench, a spacious tower, and a compact workstation can produce different connector temperatures. Cable routing can also direct or obstruct airflow.

The measurement method has limitations too. Thermal cameras report surface temperatures, not the temperature deep inside every electrical contact. Surface emissivity and viewing angle can affect readings. Still, the large before-and-after difference deserves attention because the equipment and setup remained consistent.

The result should be read as a stress signal. One tested connector became substantially hotter when the workload raised sustained current. That finding does not demonstrate melting, but it shows how narrow the available margin can become.

DLSS 5 Exposes a Power-Delivery Tradeoff

DLSS 5 is the trigger in this test, but the single-connector design is the deeper constraint.

Nvidia describes DLSS 5 as a suite built around 3D-guided neural rendering. The system uses AI models to add more realistic lighting and material behavior while reconstructing and generating frames. Those models run on the GPU’s Tensor Cores.

Earlier DLSS versions often reduced rendering pressure by starting from a lower internal resolution. DLSS 5 adds more neural work rather than functioning only as conventional upscaling. That difference can keep specialized hardware busy even when fewer traditionally rendered pixels reach the screen.

Nvidia’s current DLSS description says Dynamic Multi Frame Generation can create up to five frames for each rendered frame. The company also positions neural rendering as a way to improve scene materials and lighting.

The experience can be visually impressive, but generated frames do not erase the cost of the underlying inference. The GPU must execute the models, manage intermediate data, and continue conventional rendering work. A demanding implementation can therefore increase total utilization.

Tom’s Hardware found a similar pattern while testing the first official DLSS 5 implementation across Nvidia’s Blackwell lineup. Both the RTX 5080 and RTX 5090 reached their power limits during its performance testing.

That review also tested MSI’s RTX 5090 Lightning Z, a specialized card with two 12V-2x6 inputs. It reportedly averaged nearly 850W under one DLSS 5 workload. The card delivered 22 percent more performance than the Founders Edition while consuming 48 percent more power.

Those figures illustrate diminishing returns. Removing a power-delivery constraint can unlock more performance from Nvidia’s GB202 processor. The additional performance does not scale evenly with the extra electricity.

The Lightning Z is also an important engineering comparison. It distributes current across two connectors rather than forcing the entire external load through one. That arrangement reduces current per contact and gives each cable more headroom.

In separate thermal testing, an RTX 5090 Founders Edition cable reached 58°C on an open bench with active cooling. The Lightning Z’s two cables stayed between 46°C and 47°C under the same demanding game.

Under a synthetic 1,000W configuration, both Lightning Z cables reportedly stabilized around 52°C. The dual-connector results do not prove that every two-cable design is safe. They demonstrate how spreading current can reduce thermal concentration.

The tradeoff is physical and commercial. Two connectors need more board space, stronger power delivery, additional cables, and a larger power supply. They also complicate installation in systems already struggling with cable clearance.

A single compact connector gives card makers a cleaner product. It can support a 600W class of power delivery without several bulky legacy plugs. However, compactness also concentrates the consequences of resistance, uneven contact, or incomplete insertion.

DLSS 5 did not create that compromise. It introduced a real gaming workload capable of keeping the RTX 5090 close to its ceiling. That is more relevant to owners than a synthetic stress test they would never run voluntarily.

A feature marketed around better image quality and frame generation can now act like a practical power virus. That description concerns workload behavior, not malicious intent. It means normal software can sustain electrical conditions once associated mainly with specialized benchmarks.

The pressure now falls on Nvidia, board partners, power-supply vendors, and connector suppliers. They must ensure that a supported software feature remains safe across ordinary variations in cables, cases, and installation quality.

The Test Does Not Prove DLSS 5 Melts Connectors

A 91.7°C surface reading is concerning, but it is not evidence that DLSS 5 caused a connector failure.

No melting occurred during QuasarZone’s reported 20-minute test. The RTX 5090 completed the workload while operating within its programmed board-power behavior. The measured temperatures and peaks show stress, not a documented failure.

That difference is essential because several variables sit between elevated temperature and physical damage. Connector materials have temperature ratings, while cables and terminals have their own limits. Duration also matters because brief peaks and sustained exposure produce different outcomes.

The current distribution across individual contacts is equally important. A connector can carry high total power safely when its contacts share current evenly. One weak contact can force neighboring paths to carry more current, creating a localized hotspot.

QuasarZone reportedly examined this risk by considering a lost or impaired connection. Under normal operation, its individual pin-current measurements remained below the cited 9.2A threshold. A simulated missing path pushed one remaining contact to about 10.1A.

That scenario demonstrates reduced fault tolerance rather than normal overload. It asks what happens after one current path stops contributing properly. The answer depends on whether the card or power supply detects the imbalance before heat causes damage.

The RTX 5090 Founders Edition does not expose individual contact temperatures through standard monitoring software. Owners can watch total board power and GPU temperature while missing a developing hotspot at the plug.

That monitoring gap explains the market for temperature-sensing cables, inline meters, and current-balancing power supplies. Those products address symptoms and detection. They do not change the physical connector installed on the card.

The connector itself is the revised 12V-2x6 design, which succeeded the earlier 12VHPWR version. The revision changed contact sequencing so the sense pins engage after the main power contacts. That design helps prevent high-power operation when a plug is not fully seated.

It does not remove every failure mode. A cable can work loose after installation, bend sharply near the housing, contain a poor crimp, or develop uneven resistance. Manufacturing tolerance can also influence how tightly contacts mate.

Reports of melted RTX 5090 connectors existed before DLSS 5. Those cases occurred under other games and workloads, weakening any simple claim that the rendering technology created the problem. They support a broader interpretation involving high current and limited margin.

A separate experiment replaced the 12V-2x6 socket on an RTX 5090 with multiple conventional 8-pin inputs. The modified card handled extreme loads, but the project was not a consumer fix. It removed standard signaling and required invasive hardware work.

The experiment’s useful lesson concerned current density. Each 8-pin connection carried less current per powered contact because the load was spread across more plugs. Hardware Busters described the project as an argument for engineering margin, not a shipping recommendation.

The same analysis cautioned that one reported melted connector after a DLSS 5 session did not establish causation. Its 8-pin experiment also noted that total board power differs from cable power because the PCIe slot contributes energy.

Owners should therefore avoid treating DLSS 5 as a guaranteed hardware hazard. The evidence supports a more precise conclusion: some DLSS 5 workloads can raise sustained RTX 5090 power enough to expose connector weaknesses.

That conclusion is still consequential. Supported gaming software should not depend on perfect contacts, ideal airflow, and flawless cable routing to remain safe. Consumer systems need tolerance for normal installation differences and gradual wear.

The unresolved issue is frequency. Public tests show that severe connector heating is possible. They do not reveal how often retail systems approach those temperatures or how frequently high readings turn into damage.

The RTX 5080 Shows Why Headroom Matters

The RTX 5080 result separates a DLSS 5 workload problem from an RTX 5090 power-density problem.

Both Founders Edition cards gained power consumption after DLSS 5 was enabled. The RTX 5080 moved from 298W to 350W, while the RTX 5090 rose from 509W to its 575W software limit.

The percentages were similar, but the thermal outcomes were not. The RTX 5080 connector reached 52.2°C. The RTX 5090’s hottest connector surface reached 91.7°C under the same reported test conditions.

A similar relative increase therefore produced a dramatically different absolute temperature. The likely reason is that the RTX 5090 began much closer to the connector’s upper operating envelope.

This comparison also weakens the claim that DLSS 5 itself makes 16-pin connectors unsafe. If the software alone caused extreme heating, the RTX 5080 should have displayed a comparable result. Instead, its lower total current preserved substantial margin.

The Founders Edition design places Nvidia in a difficult position. The RTX 5090’s performance depends on a large processor, 32GB of GDDR7 memory, and aggressive power allocation. Reducing power can protect the connector, but it can also reduce clocks or performance in demanding scenes.

Board partners face the same choice. They can retain one connector and remain near the reference envelope. They can add monitoring and current balancing, or use two connectors with greater cost and physical complexity.

The Lightning Z illustrates the second path at an extreme scale. Its cooling and power systems let the processor move well beyond reference limits. Yet its measured performance gain remained smaller than its power increase.

That creates a broader question for enthusiast GPUs. Should the final portion of performance justify operating so close to a connector’s capacity? The answer depends on more than average frame rate.

System builders must account for heat dumped into the case, power-supply transients, cable placement, and long gaming sessions. A benchmark score cannot describe those installation constraints.

The RTX 5090’s connector can also behave differently as the card ages. Repeated plugging, cable tension, oxidation, and thermal cycling can change contact resistance. A new review sample does not reproduce years of ownership.

The RTX 5080’s lower reading suggests that headroom provides resilience against those unknowns. It gives the connection more room to absorb a warm room, restricted airflow, or small resistance increase.

This principle applies beyond Nvidia. Any component that operates close to a connector’s continuous capacity becomes more sensitive to contact quality. Higher nominal ratings do not eliminate the need for balanced current and fault detection.

For buyers, the comparison reframes product segmentation. The RTX 5090 provides Nvidia’s highest gaming performance, but it also demands closer attention to power delivery. The RTX 5080 sacrifices performance while remaining farther from the same connector’s limits.

That does not make one card universally better. It means the flagship’s advantages come with infrastructure requirements that deserve equal prominence. Power and thermal behavior are part of the product experience, not footnotes to frame-rate charts.

What RTX 5090 Owners Should Watch Next

Three signals will show whether this was an isolated thermal result or evidence of a wider design-margin problem.

The first signal is independent reproduction. Reviewers need to repeat the 20-minute workload across several RTX 5090 Founders Edition samples, cables, and power supplies. Tests should report ambient temperature, case configuration, insertion depth, individual contact current, and connector-surface temperature.

Reproduction would strengthen the concern if multiple properly installed systems approach 90°C. Lower readings across broader samples would suggest QuasarZone’s setup represented a harsh but uncommon combination.

The second signal is monitoring behavior from Nvidia and its partners. Firmware or software that exposes connector temperature, detects uneven current, or reduces power automatically would address the failure pathway more directly.

Some partner cards and power supplies already offer forms of current monitoring or thermal protection. Broader adoption would indicate that manufacturers consider local connector conditions important enough to track.

The third signal is DLSS 5 optimization. Its first workloads place heavy pressure on Tensor Core resources and board power. Future game patches or driver revisions might preserve visual gains while reducing sustained consumption.

If optimized releases lower average RTX 5090 connector power, the QuasarZone result will look partly like an early software-efficiency problem. If later games repeatedly hold the card near 575W, the concern will shift firmly toward hardware margin.

Owners do not need to abandon DLSS 5 while those answers develop. They should treat the RTX 5090 as a high-current device and follow careful installation practices.

The connector should be fully seated, with no visible gap between the plug and socket. The cable should not bend sharply immediately after leaving the housing. Side panels should not force the wire against the card.

Users should also follow their card and power-supply manufacturers’ compatibility guidance. A native, approved cable reduces the number of connections, although it cannot guarantee balanced current. Adapters should use every required power lead from the supply.

Unusual odor, discoloration, unstable power readings, or a connector that becomes visibly distorted warrants an immediate shutdown and inspection. Owners should not unplug a hot connector or continue testing suspected damage.

Software power limits and undervolting can reduce consumption, but they are user-managed compromises. They should not substitute for a safe stock configuration. Any tuning also requires stability testing across different applications.

The Nvidia DLSS 5 power draw result ultimately reveals a mismatch in visibility. Gamers can see generated frames, latency, GPU utilization, and chip temperature. They generally cannot see the temperature at the electrical interface carrying almost 600W.

That blind spot matters because the connector can become the hottest measured part of the card. QuasarZone’s test did not prove that DLSS 5 melts plugs. It showed that a supported feature can push the RTX 5090 into a range where small imperfections matter more.

The next round of testing should focus less on spectacular peak numbers and more on repeatability. How many cards reach 90°C, how evenly do their pins share current, and how quickly do protections respond?

Those answers will determine whether 91.7°C was an outlier or a warning. Until then, RTX 5090 owners should watch the connector as closely as the GPU, especially during sustained DLSS 5 workloads.