Why Does Your GPU Need Two or Three Power Cables? PCIe Slot 75W + External Power Explained

2026-10-23 ｜ GPU Repair ｜ About 28 minutes ｜ HK GPU Fix Editorial Team

Summary

A graphics card draws power from two places: the motherboard PCIe slot and its external power cables. This guide explains the verified limits behind each connector, why every socket must be filled, how sense pins work, and how to diagnose black screens and restarts by changing one variable at a time.

Two Sources of Power, One Combined Limit

A graphics card does not run on a single cable. It receives power from two directions at once. One direction is the motherboard PCIe x16 slot into which the card is seated. The other direction is the set of external cables that run directly from the power supply unit (PSU) to the sockets on the edge of the card. Inside the card, the power delivery circuitry distributes that incoming power to the graphics processor, the video memory (VRAM), the fans, and the other components that need it.

This two-source arrangement is the key to almost every confusion about GPU power cables. The card's total power ceiling is the sum of the rated limits of each source that is connected. The PCIe slot has its own limit. Each external connector has its own limit. Adding them together gives the maximum the card is designed to be able to draw through those paths. It does not mean the card always draws that maximum. A limit is a ceiling, not a consumption figure, and the distinction matters throughout this guide.

It also explains why the wattage printed on the side of a PSU is not, on its own, a complete answer. A PSU can have a large total wattage rating and still be the wrong PSU for a particular card, if it does not provide the right number and type of connectors, or if its 12V output is not sufficient for the card. Total wattage and connector provision are two separate checks. Both must pass. Failing either one means the PSU is not a suitable match for that card, and no adapter can change the PSU's own capability.

The Rated Limits, Stated as Limits

The following table uses only the rated limits that are documented in the specification and in manufacturer material. Read every figure as an upper limit for that path. The right-hand column shows what the slot plus the external connectors add up to. It does not state what any particular card consumes, because actual consumption depends on the card's own design and on the workload.

Two cautions belong with this table, and they apply to every row. First, the combined ceiling is not the card's power draw. A card with two 8-pin connectors has a 375W ceiling through the slot and connectors combined, but that does not mean it draws 375W. Its actual draw is set by its own design and by what it is doing at the time. Second, always check the card's own official specification first. The table describes the connector system in general; the card manufacturer's specification for your exact model and variant is the authority for how many sockets that card has and what PSU it recommends.

One worked example makes the limit-versus-draw distinction concrete. The Intel Arc A750 and A770 use one 8-pin and one 6-pin connector. Their combined ceiling is 300W: 75W from the slot, 150W from the 8-pin connector, and 75W from the 6-pin connector. Their total board power (TBP) is 225W. The 300W figure is the ceiling of the paths provided. The 225W figure is the board's rated power. Confusing those two numbers — treating the ceiling as the consumption, or treating the consumption as the ceiling — is one of the most common errors in GPU power discussions, and it leads directly to wrong PSU choices.

A related reservation applies to the slot figure itself. The specification limit for the PCIe x16 slot is 75W, and that is the number to plan around. Individual small systems or older motherboards may have slot power designs below the specification, so the safe statement is that the slot limit is 75W subject to the motherboard and graphics card official specifications, not that every motherboard in existence guarantees a full 75W to every card. When in doubt, the motherboard manual and the card specification outrank any general table.

Why the Slot Alone Is Not Enough

A card that needs no external power cable at all is a normal design, not a card that is missing something. Low-power cards are designed to stay within what the slot can provide, and for those cards the slot alone is the whole power story. The moment a card's design power goes beyond 75W, however, the slot cannot carry the card on its own, and external power becomes a requirement rather than an option.

That is why higher-performance cards carry one, two, three, or (in the 16-pin generation) one larger external socket. The number of sockets on the card is the manufacturer's statement about how many external paths the design expects. Every socket must be filled with the correct cable, fully seated, with its latch engaged. A socket left empty is not a minor omission. It removes one entire rated path from the card's power supply, and the card may refuse to start, display a warning asking for the power cable to be connected, or behave unpredictably under load.

The temptation to "try one cable first and add the second later if needed" should be resisted. The sockets are not spares, and they are not stages to be added one at a time as an experiment. They are part of the design. If a card has two sockets, it is designed to be run with two cables connected. Testing with a socket deliberately left empty does not produce useful diagnostic information; it produces a card running outside its designed power arrangement, and any symptom observed in that state cannot be trusted as evidence about the card, the cables, or the PSU.

6-Pin and 8-Pin: More Than a Pin Count

The difference between a 6-pin and an 8-pin connector is not cosmetic, and it is not only about having two extra pins to hold the plug in place. A 6-pin connector provides three +12V lines and ground lines, and is rated at 75W. An 8-pin connector, or a 6+2-pin connector that is assembled into the 8-pin form, provides an additional +12V and ground pair, and is rated at 150W.

The two additional pins on the 8-pin connector are sense pins. They do not carry the main current. Their job is signalling: they tell the graphics card that a 150W-class connector is attached, so the card can set its power behaviour accordingly. This is why physical fit is a poor guide to electrical adequacy. A 6-pin plug can be inserted into an 8-pin socket in some configurations, and it may feel as though it fits, but without the sense pins connected the card understands the connection as a 75W-class connection. The card is not being stubborn. It is reading the only information it has been given, and it is limiting itself to the power level that the connection it can detect is rated to provide.

The practical rule follows directly. To obtain the 150W rating of an 8-pin path, the full 8-pin (or 6+2-pin fully assembled) plug must be inserted, fully seated, with the sense pins making contact. A partially inserted plug, a 6-pin plug in an 8-pin socket, or a plug whose sense pins are making poor contact can all leave the card with a wrong or reduced understanding of the power available to it. That is one of the mechanisms behind the frustrating pattern in which a card works at idle but fails under load, or works after reseating and then fails again: the connection is marginal, and load is what exposes it.

Labels matter here. PSU cables are normally marked on the connector or on the cable itself. PCIe or VGA markings indicate the GPU power cables. CPU power cables, which use a 4+4-pin arrangement, are a different cable with a different pin definition, and they are discussed separately below because confusing the two is a real and potentially damaging error.

The 16-Pin Generation: 12VHPWR and 12V-2x6

The 16-pin connector is a different generation of design, introduced to replace the practice of running two or three 8-pin cables to a single high-power card. It has 12 power pins and 4 signal or sense pins, and a single 16-pin cable carries a rated limit of 600W. Combined with the 75W slot, the theoretical ceiling is 675W. As with every other row in the table, that 675W figure is a theoretical ceiling for the paths, not a statement that any particular 16-pin card draws 675W. For any specific card, the card manufacturer's official specification states the card's own power figure, and that specification governs.

The revised version, 12V-2x6, is physically compatible in outline with the earlier 12VHPWR, but it is not identical. Its sense pins are shortened so that the power pins must be fully inserted before the connection is confirmed. The purpose of that change is to reduce the risk of a half-inserted connector passing unnoticed and then overheating through poor contact. The two generations look alike, and that similarity is itself a source of error. When buying a replacement cable, match it to the card and the PSU specifications by model, rather than assuming that any 16-pin cable is equivalent to any other.

Seating a 16-pin connector deserves particular care. It must be inserted fully, along its whole length, with no visible gap remaining. A connector that looks nearly seated but still shows a gap is not seated. The sense-pin change in 12V-2x6 exists precisely because a small remaining gap used to be easy to miss. After inserting a 16-pin cable, check it visually along the full connector and confirm that the latch, where fitted, is engaged. If the cable run puts sideways strain on the connector, reroute the cable so the connector is not being pulled at an angle.

Older PSUs without a native 16-pin output are a common boundary case. Using the adapter supplied or specified by the manufacturer is one manufacturer-recognised approach, but an adapter is not the same thing as a native cable. An adapter adds an extra set of contact points, one on each side, and every one of those contacts must be fully seated. The adapter does not increase the PSU's own 12V capability, and it does not turn a PSU that lacks the required output into one that has it. The correct statement about adapters is balanced: they are not automatically dangerous, and they are not equivalent to a native cable. Each side must be checked, and the PSU must still pass both of its own checks — total and 12V output, and native connector provision.

"Power Imbalance": What the User Can and Cannot Control

The phrase "power imbalance" is often used loosely, and it can mislead. A graphics card's internal power design decides how current flows through its own power stages. The user has no control, knob, or setting that redistributes current between the card's internal paths. There is no user-level "balancing" to perform, and no setting that compensates for a faulty connection.

What the user can control, and what actually goes wrong in practice, is the condition of each individual path into the card. Problems cluster into a small number of patterns, each of which removes or degrades one path:

First, a socket left empty, a cable not fully inserted, or a sense pin making poor contact. The card then either refuses to start, warns that power is not connected, or understands the available power as lower than the design expects. At idle the card may appear normal, because idle draw is low. Under load the shortfall appears as a black screen, a restart, or a crash.

Second, a daisy-chain arrangement: one cable from the PSU split through a pigtail or Y-splitter to feed two sockets on the card. In that arrangement, the full current for both sockets travels through a single cable and a single PSU terminal. At high load, that single path carries more current than it would if each socket had its own cable, with greater voltage drop and greater heating at the shared points. The recommendation for higher-power cards is therefore to run a separate, independent cable from the PSU to each socket on the card. This is not a claim about a universal wattage limit for a single cable — no such universal figure is stated in this guide, because the safe arrangement depends on the PSU, the cable, and the card together. The recommendation is comparative and practical: independent cables are the more robust arrangement, and a daisy-chain is the arrangement to avoid on a higher-power card.

Third, adapters and mixed modular cables. An adapter adds a contact point, and every additional contact point is an additional place where resistance can rise if the connection is imperfect. Modular cables are a sharper hazard: the PSU end of a modular cable is not standardised across brands, and in some cases not across models from the same brand. A cable from another brand's PSU may plug into the socket and yet connect the wrong voltages to the wrong pins. The rule is absolute: use only the cables supplied with, or specified for, your exact PSU model. Never mix modular cables across brands or across PSU models. A cable that fits is not proof that it is wired correctly.

Fourth, wear and damage at the connector itself. Contacts oxidise over time. The spring tension of the connector's contacts weakens with repeated insertion and removal. A plug that has overheated may show discolouration, deformation, or a burnt smell. Any of these raises contact resistance, and contact resistance is what turns into voltage drop under load. This is the mechanism behind the misleading pattern in which reseating a cable "fixes" the problem for a while and then it returns. Reseating scrapes the contact surfaces against each other and can temporarily cut through an oxide layer, but it does not restore worn spring tension, and it does not repair a damaged contact. A fault that returns after reseating should be treated as a contact problem to be investigated — by trying a different original cable for the same PSU, or a different PSU terminal — not as a problem that has been solved.

Two Checks, Not One: Wattage and Connectors

Choosing or checking a PSU for a card means passing two independent checks, and the order does not matter as long as both are done:

Check one: capacity. Does the PSU's total wattage, and specifically its 12V output, meet the card manufacturer's recommendation for that exact card, with the rest of the system taken into account? A PSU whose total wattage looks generous on the box but whose 12V output is weak for the card's needs fails this check.

Check two: connectors. Does the PSU natively provide the number and type of connectors the card requires — the right count of 6-pin, 8-pin, or 16-pin outputs, without relying on splitters or adapters to invent a connector the PSU does not have? A PSU that passes the wattage check but has only one 8-pin output cannot properly feed a card that requires two 8-pin inputs.

If either check fails, the conclusion is the same: that PSU is not a suitable match for that card. The remedy is a PSU that passes both checks, not an adapter that makes the cable fit. An adapter can change the shape of a plug. It cannot add 12V capacity the PSU does not have, and it cannot create a native connector the PSU was not built with. This is the single most useful sentence in PSU selection, and it is worth repeating to yourself before buying anything: fit is not capacity, and capacity is not fit.

Reading the Symptoms Without Jumping to Conclusions

Power problems produce symptoms that overlap with several other faults, including overheating, driver problems, and faults in the card or motherboard. Symptoms are clues, not verdicts. The table below groups the common patterns and states what each pattern suggests checking first, while keeping the alternative explanations visible.

The last row is not a diagnostic suggestion. It is a stop instruction. A connector that shows heat damage has already told you everything it needs to tell you. Continuing to run high loads through it in the hope that it will "hold for one more test" risks further damage to the cable, the PSU, or the card. Remove that power path from service, do not force the system to run high loads through it, and have the card and PSU inspected.

The broader diagnostic discipline is equally important. A fault that appears at idle as well as under load, that survives reinstalling several different driver versions, and that occurs while temperatures are normal is not behaving like a straightforward power-path fault. In that situation, repeatedly reinstalling drivers or replacing the PSU without result is not persistence; it is repeating a test that has already answered. The investigation should widen to the card itself, the motherboard slot, cooling, and other directions, rather than narrowing further onto the power supply. Knowing when to stop pursuing one hypothesis is part of diagnosis, not a failure of it.

A Diagnostic Sequence That Changes One Variable at a Time

The diagnostic method for a suspected power problem is deliberately slow, because speed is what destroys evidence. Change one thing, observe, and only then change the next thing. If two things change at once and the symptom disappears, you have not learned which change mattered, and the fault may return with the next change you make.

Work in this order:

1. Identify the card exactly. Read the model and variant from the card's label and the manufacturer's official specification: how many sockets, of which type, and what PSU the manufacturer recommends for that exact card. The official specification outranks any general guide, any forum post, and any assumption based on the chip name alone, because the same chip can appear in variants with different socket combinations — for example, one variant with a single 8-pin socket and another with two.
2. Count sockets and count cables. Every socket on the card must have a cable, fully inserted, latch engaged. For a 16-pin connector, confirm there is no visible gap along the connector. Check the PSU end as well: a cable that is seated at the card but loose at a modular PSU terminal is not connected.
3. Check the PSU twice, as two separate checks: total wattage and 12V output against the card manufacturer's recommendation, and native connector type and count against the card's sockets. Record which check, if either, fails.
4. For a card with two or more sockets, replace any daisy-chain with independent cables: one cable from the PSU to each socket. Do not test a high-power card on a daisy-chain and then conclude the card is at fault.
5. Verify cable identity. Confirm that GPU cables are marked PCIe or VGA, not CPU or EPS, and that modular cables belong to this exact PSU model. If there is any doubt about a modular cable's origin, do not use it. The cost of a correct replacement cable is small compared with the cost of a wrongly wired one.
6. Substitute one element at a time, in this order of increasing disruption: a different original cable for the same PSU; a different PSU terminal for the same cable; a different PSU that passes both checks; the card in a different machine, if one is available. After each single change, test under the same load that produced the symptom before. Only one change per test.
7. Observe the pattern, not just the outcome of one test. A fault that disappears after reseating and then returns is contact trouble until proven otherwise. A fault that follows the card into a different machine points at the card. A fault that disappears with a different PSU and stays away points at the PSU or its cables. A fault that follows neither points elsewhere, and the investigation should widen rather than repeat.

Two warnings belong inside this sequence. First, never mix modular cables across brands or models during substitution testing. A substitute cable must be a cable for that PSU. Substituting a cable from another PSU is not a test; it is a new hazard. Second, do not use a damaged connector as a test subject. If a connector shows discolouration, deformation, or melting, that path is out of service until inspected, and no further load testing runs through it.

CPU Cables and GPU Cables Are Not Interchangeable

This error deserves its own section because it is common, it looks plausible, and it can damage hardware. A CPU power cable uses a 4+4-pin arrangement. A GPU power cable uses a 6+2-pin arrangement. They are keyed differently, they carry different pin definitions, and they serve different components. They are not two versions of the same cable.

The keying is designed to prevent insertion, but keying is a deterrent, not a guarantee, and a determined or inattentive insertion can defeat it or damage the connector in the attempt. The defence is reading the marking before inserting anything: PCIe or VGA for the graphics card, CPU or EPS for the processor power. If a cable is unmarked and its origin is uncertain, do not guess. Trace it back to the PSU it came with, or replace it with a known cable for that PSU. Guessing with power cables is the one place in PC assembly where a wrong guess can destroy the component it was meant to power.

The same principle applies to modular cables generally. The end that plugs into the graphics card follows the PCIe standard, which is why cables from different PSUs can look identical at the card end. The end that plugs into the PSU does not follow a single standard, which is why they can be wired differently at that end while looking the same at the other. Identical appearance at one end proves nothing about the other end. Only the PSU manufacturer's own cable list for that exact model proves it.

Common Misconceptions, Corrected One by One

• "The PSU wattage is enough, so I can make up missing connectors with adapters." Incorrect. Wattage and connector provision are independent checks. An adapter changes the plug shape; it does not add 12V capacity or create a native connector.
• "A 6-pin plug fits the 8-pin socket, so it must deliver 150W." Incorrect. Without the sense pins connected, the card understands the connection as 75W-class. Full 150W delivery requires the full 8-pin or 6+2-pin connection.
• "I will try one 8-pin cable first, and add the second only if needed." Incorrect. Sockets the manufacturer fitted must all be connected from the start. Deliberately running with a socket empty is running outside the design.
• "Two cables share the load evenly, so one slightly loose cable does not matter." Incorrect. The user cannot control how current divides between the card's internal paths, and a degraded path raises the burden and the voltage drop on the remaining paths. Every path must be sound.
• "CPU 8-pin and GPU 8-pin cables are interchangeable." Incorrect. Different pin definitions and different keying. Using the wrong cable risks damage to the component.
• "The slot gives every card a free 75W on top of everything else." Needs a reservation. The specification limit is 75W, but individual small or older motherboards may provide less, and the motherboard and card specifications govern for the actual machine in front of you.
• "12VHPWR and 12V-2x6 are exactly the same thing, so any 16-pin cable will do." Inaccurate. The outlines are compatible, but the sense-pin length and the protection against half-insertion differ. Match replacement cables to the card and PSU specifications by model.
• "An adapter is just as good as a native cable." Incorrect, and so is the opposite extreme. An adapter is a manufacturer-recognised approach in some situations, notably for older PSUs without a native 16-pin output, but it adds contact points that must each be fully seated, and it does not increase the PSU's capability. It is neither equivalent to a native cable nor automatically dangerous. Judge it by whether every contact is sound and whether the PSU passes both checks on its own.
• "Reseating fixed it, so the problem is gone." Not proven. Reseating can temporarily improve a contact by scraping through oxidation. If the fault returns, the contact itself is the suspect, and a replacement original cable or a different PSU terminal is the next step, not another reseat.

Boundary Cases Worth Knowing Before You Start

Low-power cards with no external socket are not incomplete. They are designed to live within the slot's 75W, and nothing needs to be added to them. Adding a cable is not possible on such a card, and it is not needed.

The same chip name does not guarantee the same sockets. Board partners produce variants of the same chip with different socket combinations, different cooling, and different power designs. Always work from the label on the card in front of you and from that variant's official specification. Advice written for a different variant of the same chip may describe sockets your card does not have, or omit sockets it does.

Overclocking and undervolting change actual power draw without changing any rated limit in the table. Overclocking increases power draw and therefore increases the pressure on every power path; undervolting reduces it. This guide gives no figures for overclocked or undervolted draw, because those figures depend on the specific card, the specific settings, and the specific workload. The rated limits stay the same. The margin between actual draw and those limits is what changes, and it changes in the direction the settings push it.

Finally, a word about what this guide does not claim. It does not state how current divides between two connectors on any particular card, because that division is set by the card's internal power design and is not published as a general rule. It does not state a universal wattage limit for a single daisy-chained cable, because no such universal figure is established by the sources used here. Readers who encounter a specific wattage figure for either of those things elsewhere should ask what card, what PSU, and what cable that figure was measured with, because without those three answers it is not a transferable specification.

Frequently Asked Questions

Why does my card have two power sockets when my friend's card has only one?

Because the two cards have different power designs. The number of sockets reflects how much external power the manufacturer designed that variant to receive, on top of the 75W slot limit. A card with two sockets is designed to run with two cables connected. A card with one socket, or with no socket at all, is designed for a lower power envelope. Compare cards by their own official specifications, not by socket count alone.

Can I use a 6-pin cable in an 8-pin socket if it fits?

Physically it may insert, but electrically the card will understand the connection as 75W-class, because the sense pins that signal a 150W-class connection are not connected. For a card that requires an 8-pin connection, use the full 8-pin or 6+2-pin cable, fully seated. "It fits" is not the same as "it delivers the rated power."

Is a daisy-chain cable really a problem, or is that just caution?

A daisy-chain routes the current for two sockets through one cable and one PSU terminal. At high load that single path carries more current, with more voltage drop and more heating at the shared points, than two independent cables would. For a higher-power card, independent cables from the PSU to each socket are the more robust arrangement. The guide deliberately does not quote a universal wattage limit for a single cable, because the safe arrangement depends on the PSU, cable, and card together.

Can I use a modular cable from a different brand's PSU if the plug fits?

No. The PSU end of a modular cable is not standardised across brands, and sometimes not across models from the same brand. A cable that fits at the PSU socket can still connect the wrong voltages to the wrong pins. Use only cables supplied with, or specified for, your exact PSU model.

My card is fine at idle but black-screens in games. Is that definitely a power problem?

Not definitely. It is consistent with a power path that is adequate at low current and fails at high current, but the same pattern is also consistent with overheating, a driver fault, or a fault in the card. Use the one-variable-at-a-time sequence: check every socket is filled and seated, replace any daisy-chain with independent cables, substitute one original cable or one PSU terminal at a time, and observe whether the symptom follows the card, the cable, or the PSU. If the pattern does not point at power after those steps, widen the investigation rather than repeating the same tests.

The connector looks slightly discoloured. Can I keep using it carefully?

No. Discolouration, deformation, a burnt smell, or melting at a connector means that power path is damaged. Do not continue running high loads through it, and do not treat it as a test subject. Remove that path from service and have the card and PSU inspected. Repair pricing on this site follows the published three-tier range of HK$400–2,800, with the actual quote given after inspection.

Does a bigger PSU wattage number fix a missing connector?

No. Wattage and connectors are separate checks. A PSU with a large total wattage but only one 8-pin output cannot properly feed a card that requires two 8-pin inputs, and an adapter does not add the missing native output or the missing 12V capacity. The PSU must pass both checks for the card in question.

Where to Go from Here

Start from the card's own specification, not from the PSU box and not from another variant of the same chip. Count the sockets, fill every one with the correct cable, check the PSU on both capacity and connectors, and give a higher-power card independent cables rather than a daisy-chain. When a fault appears, change one variable at a time and let the pattern — whether the fault follows the card, the cable, or the PSU — do the deciding.

If a connector is heat-damaged, if a fault follows the card into a different machine, or if the one-variable-at-a-time sequence points away from the PSU and cables without resolving the fault, further guessing at home has diminishing returns. HK GPU Fix is at Room 1209, Dali Plaza, Shek Mun, Sha Tin, telephone 2186 8010, WhatsApp 6558 6806. Repair pricing follows the published three-tier range of HK$400–2,800, with the actual quote given after inspection.