How Many Watts Does Your GPU Upgrade Really Need? PSU Sizing, TDP, Transient Headroom & 12V Rail Explained

2026-10-22 ｜ GPU Repair ｜ About 19 minutes ｜ HK GPU Fix Editorial

Key Takeaways

Before buying a graphics card, budget from its official board power, not its TDP, and add the processor's maximum turbo power plus an allowance for the rest of the system. Check the power supply's combined 12V wattage and native connectors, and favour ATX 3.0 or 3.1 for newer high-end cards. Do not size exactly to the edge.

A Planning Question, Not a Fault-Finding Question

This article answers a question asked before anything is bought and before anything has gone wrong: how much power will the upgraded system need, and does the existing power supply already provide it? The reader's current system is assumed to be working normally. Three questions carry the whole decision: how much power does the new card need, how much does the whole system need at peak, and is the existing power supply's 12V capacity and connector set sufficient for both?

That framing matters because sizing and diagnosis are different jobs. Tripping, restart and black-screen faults, loose-connector checks, and mixed modular-cable problems are separate problems and are not diagnosed here.

The usual mistake is to treat one number as the answer. The wattage printed on the power supply box, the TDP printed on the card's specification page, and the recommended power supply wattage printed on the card manufacturer's page are three different figures, and none of them alone tells you whether an upgrade will work. Each has a specific meaning, and the budget only becomes reliable when each is used for its own purpose.

Why the Box Wattage Misleads: Follow the 12V Rail

Modern processors and graphics cards draw almost entirely from the 12V supply. That single fact reorders the whole sizing exercise. The headline wattage on the box is the power supply's total rated output across all of its voltage rails, not a statement about how much 12V power it can deliver. A unit can carry an impressive total figure while its 12V section is comparatively modest, and it is the 12V section that feeds the two components that dominate the budget.

The figure to read is on the power supply's own label: the +12V line, stated in amps, together with the combined +12V limit where one is printed. The conversion is basic electrical arithmetic: watts equal volts multiplied by amps, so the 12V wattage equals 12 multiplied by the +12V amps. A label-reading example from a secondary source illustrates the method: a quality 750W unit with 62A or more on the 12V rail provides 62 multiplied by 12, which is 744W, on that rail. That example demonstrates how to read a label; it is not a claim that every 750W unit delivers 744W on 12V, and it must not be turned into a pass mark that other units are judged against. The only figure that counts for a given unit is the figure printed on its own label.

Multi-rail units need one extra caution. A power supply with several 12V rails does not allow the rails to be added together freely to reach a total. Each rail has its own limit, and the label's stated combined +12V limit governs the total that can actually be drawn. Adding the individual rail ratings and treating the sum as available capacity overstates what the unit can deliver, and the overstatement lands exactly where the budget can least afford it: at peak load, when the card and the processor draw together.

TDP, TGP and TBP Are Not the Same Number

The most persistent source of undersized budgets is the confusion between three terms that look interchangeable and are not.

TDP, or Thermal Design Power, is fundamentally a thermal-design figure. It describes the heat a component is designed to produce under its design load, and therefore the heat its cooling must be able to dissipate. It is not a direct measure of how much electrical power the board draws from the supply. Using a TDP figure as though it were board power understates the supply requirement whenever the board's actual draw exceeds its thermal design figure, which is a common situation under heavy load.

TGP, or Total Graphics Power, is the term commonly used by NVIDIA. It refers to the combined power of the power-consuming parts on the graphics card's circuit board, and it is therefore much closer to the figure a supply budget needs. TBP, or Total Board Power, is the term commonly used by AMD, and its concept is close to TGP: the power of the whole board. Some sources describe TBP as including board-level parts such as fans and lighting, which would make it slightly higher than TGP, while other sources treat the two terms as the same concept. The honest statement is that vendor wording is not uniform, and no fixed watt gap between TGP and TBP should be asserted.

The practical rule follows directly: budget from the specific card model's official board-power figure, whether the manufacturer calls it TGP or TBP, and never from a single TDP figure alone. Where the manufacturer's page gives both a thermal figure and a board-power figure, the board-power figure is the one that belongs in the budget. Where only one figure is given, check what that figure actually measures before using it, and confirm it against the card manufacturer's own specification page rather than a retailer summary or a forum post.

The Processor Counts Too, and Its Maximum Matters More Than Its Base

The same error repeats on the processor side. A processor's base power figure describes sustained operation at its base frequency, but processors under load boost well above base, and many motherboards relax power limits by default, allowing sustained draw above the base figure. Budgeting from base power alone therefore understates the processor's contribution at exactly the moment the graphics card is also drawing heavily.

Intel defines two relevant figures: PL1, the sustained power limit, roughly equal to the stated base power, and PL2, the short-term maximum turbo power. AMD uses PPT, or Package Power Tracking, the socket power limit, together with related limits. The budget must use the processor's maximum turbo power or PPT figure, not its base TDP.

Examples from secondary summaries show why the distinction matters, and they must be read as examples tied to specific models, always verified against the official specification of the processor actually installed. A Core i5-13400 is summarised at 65W base with 154W maximum turbo power; a Core i7-14700K and a Core i9-14900K are summarised at 125W base with 253W maximum turbo power; a Ryzen 5 7600 is summarised at 65W TDP with 88W PPT; a Ryzen 9 7950X is summarised at 170W TDP with 230W PPT. In each case the maximum is far above the base, and a budget built on base figures would be short by the difference. One secondary source suggests an approximate conversion of TDP multiplied by 1.35 for AMD PPT on AM4 and later platforms, but that is an experience-based approximation, not an official formula, and it must not be presented as one. The official PPT figure for the exact model governs.

Connector Ratings: Watts and Connectors Are Separate Checks

A graphics card receives power from two places added together: the motherboard PCIe slot and its external power connectors, which are then distributed on the card to the graphics processor, the video memory, the fans, and the other board parts. The slot and each connector type carry a defined rating, and the card's total supply is the sum of what its slot and its fitted connectors can provide.

Two consequences follow. First, watts and connectors must be checked separately: a power supply with enough total wattage but only one 8-pin connector cannot feed a card that needs two, and a supply with enough connectors but insufficient 12V wattage cannot feed the same card either. Either mismatch alone disqualifies the unit. Second, native connectors are what count. If the required connector type or count is not natively present, the correct conclusion is that the power supply does not fit the upgrade, not that an adapter can be used to conjure the missing connector. An adapter does not increase the supply's own 12V capacity, and the detailed risks of mixing modular cables belong to a different discussion.

ATX 3.x and Transient Headroom: Why Rated Watts Are Not the Whole Story

Graphics cards do not draw power smoothly. Their draw spikes briefly during load changes, and a power supply must ride through those short excursions without shutting down. This is where the power supply's ATX version becomes a sizing factor in its own right.

The Intel ATX 3.0 specification, Table 3-3, sets out a power excursion requirement: the short-duration overload, expressed as a percentage of the power supply's rated power, that a compliant unit must tolerate. For units above 450W fitted with a 12VHPWR connector, the requirement is 200% for 100 microseconds, 180% for 1 millisecond, 160% for 10 milliseconds, and 120% for 100 milliseconds. For units of 450W or below, or without a 12VHPWR connector, the requirement is 150% for 100 microseconds, 145% for 1 millisecond, 135% for 10 milliseconds, and 110% for 100 milliseconds.

Percentages are relative to rated power, and the requirement is a short-duration ride-through capability, not a continuous load rating. Nobody is expected to run a power supply at 200% of its rating continuously, and the specification does not license doing so. The practical point is narrower and more important: ATX 2.x units were never required to meet this excursion requirement at all, so an older unit can carry a sufficient rated wattage on paper and still lack any guaranteed ability to ride through the same class of spike. For newer high-end cards, an ATX 3.0 or 3.1 unit is therefore the prudent choice, and an adapter cable does not turn an ATX 2.x unit into an ATX 3.x unit.

The Seven-Step Budget

With the definitions settled, the budget itself is straightforward arithmetic done in the right order.

Step 1: Note the target card's official board power, TGP or TBP, and its connector type and count. Both facts come from the card manufacturer's specification page for that exact model. A board-power figure borrowed from a different model, even a closely related one, does not belong in this budget.

Step 2: Note the processor's maximum turbo power or PPT figure, again for the exact installed model, from the processor manufacturer's specification. Base power and base TDP have no role here.

Step 3: Add an allowance for everything else: the motherboard, memory, storage drives, fans, liquid-cooling pump, and USB devices. No verified universal fixed figure exists for this remainder, and none is offered here. Configurations differ too much: a system with several hard drives, many fans, and extensive lighting needs a larger allowance than a minimal build. Estimate item by item for the actual configuration, or cross-check the total against the card or power supply manufacturer's official calculator, and treat any calculator result as a cross-check on the manual budget, not a replacement for understanding it.

Step 4: Convert the total into a rated power supply size using a planning convention. The convention used in secondary guidance is to keep everyday heavy load at roughly 50 to 80 percent of the unit's rated capacity, with an additional 20 to 30 percent of headroom allowed for transient excursions and future upgrades.

To be clear, the 50–80% band and the 20–30% headroom are not two allowances to be added together: they are two ways of describing the same reserve. When heavy load sits at 50–80% of rating, the remaining headroom is 20–50%, and the 20–30% figure describes the headroom left when load sits near the top of that band. To apply it, divide the step 3 total by the target load fraction: a system expected to peak at 480W, sized for 80 percent load, points to a 600W rating, while the same total sized for 60 percent load points to 800W. This is a planning convention from secondary sources, not a figure from the Intel specification, and it must be understood as such. It exists because a unit sized exactly to the expected peak has no margin for spikes, for ageing, or for the next upgrade, while a unit sized far above the load brings its own costs, discussed below.

Step 5: Run the triple check, all three parts of which must pass. First, the power supply's combined +12V wattage must be at least the system's peak total. Second, its total rated wattage must be at least the step 4 result. Third, its native connector type and count must match the card's requirement exactly. A failure on any one of the three is a failure of the fit, however comfortably the other two pass.

Step 6: Check the ATX version. For a newer high-end card, favour ATX 3.0 or 3.1, for the excursion reason given above. For a modest card whose draw is well inside the supply's comfortable range, the version matters less, but it should still be a conscious check rather than an unnoticed detail.

Step 7: Only now consider price and warranty. Reversing the order, choosing a unit by price and then discovering it fails the triple check, wastes the effort spent on every earlier step. A power supply is also a long-lived component that typically survives several graphics card generations, which is a further reason to size it for the next upgrade as well as the current one, rather than buying the smallest unit that passes today's numbers.

Reading the Manufacturer's Recommended Wattage Correctly

Card manufacturers commonly publish a recommended power supply wattage alongside each card. That figure is frequently misread in two opposite directions: as a minimum for the card alone, which invites readers to go below it, or as an exact requirement, which invites unnecessary spending.

The recommended figure is normally a whole-system figure. It assumes a typical complete system built around that card, including a processor, motherboard, storage, and cooling, and it already contains the manufacturer's own margin for typical configurations. It is therefore not a figure for the card alone, and going below it on the reasoning that the card itself draws less is a mistake: the difference between the card's board power and the recommended figure is largely the rest of the system and the margin. Conversely, a system that is unusually light, or unusually heavily loaded with drives and cooling, may need more or less than the recommendation, which is why the manual budget in the previous section remains the deciding method and the recommendation serves as a sanity check on it. If the manual budget and the recommendation disagree sharply, the discrepancy should be investigated, not resolved by picking the smaller number.

80 PLUS Means Efficiency, Nothing More

The 80 PLUS label is another figure routinely over-read. It certifies efficiency: at specified loads, the unit converts at least 80 percent of its input power into output, with the Bronze, Silver, Gold, Platinum, and Titanium tiers marking progressively higher efficiency. Efficiency is worth having, since wasted power becomes heat, but it is not a quality certification, not an output-capability guarantee, and not evidence about 12V capacity. A Gold-rated unit is not automatically sufficient for a high-end card, and a Bronze-rated unit is not automatically insufficient. The label's answer to the sizing question is the +12V figure and the connector set, not the efficiency tier.

The Oversizing Question, Answered Honestly

If margin is good, is more margin always better? Not unconditionally, and the honest answer matters because oversizing has real costs. A very large unit costs more, occupies the same or larger physical space, and may operate at low load much of the time, where its efficiency is not necessarily at its best. None of those costs is a reason to size exactly to the edge, however. The position this article recommends is the middle one stated in the planning convention: a reasonable reserve for transients and for the next upgrade, without chasing the largest unit available. A unit that idles at a small fraction of its rating for years has been bought for capacity its system never uses, while a unit with no reserve at all has been bought for a system that no longer exists the moment any component changes.

Edge Cases That Change the Answer

Overclocking and unlocked power limits invalidate a budget built on official figures. Raising clock speeds or relaxing power limits increases actual draw beyond the official board power and processor turbo figures, so a budget calculated before overclocking no longer describes the system after it. The warning is simple: recalculate for the overclocked configuration, using the raised limits rather than the stock figures. This article does not teach overclocking settings.

An old power supply is a second variable. Ageing, particularly capacitor ageing, reduces a unit's real output capability below its label over time, and no verified annual percentage decline was found for this article, so none is stated. The only safe statement is that a unit used for many years must not be assumed to still deliver its label rating, and an upgrade is a sensible moment to question an old unit rather than to inherit its assumptions silently.

Physical fit and airflow belong in the same pre-purchase check. A replacement unit must fit the case, and a small case may restrict both the unit's dimensions and the airflow around it. Laptops, all-in-one machines, and branded prebuilt systems may use non-standard or non-replaceable power supplies, and the assumption that any desktop can accept a standard ATX unit should be verified for the specific machine before a card is ordered. A card that cannot be fed because the machine cannot accept a standard supply is a purchasing error, not a power supply fault.

Multi-graphics-card systems and sustained workstation loads are outside this article's scope. The single-card formula here describes a single gaming or general-purpose card; a multi-card or continuously fully loaded workstation system needs a different budget built for its own sustained draw, and forcing the single-card arithmetic onto it understates the requirement.

Common Misconceptions, Corrected

TDP equals power draw is incorrect: TDP is a thermal figure, and the budget uses board power and processor maximum power. Box wattage equals 12V capability is incorrect: only the label's combined +12V figure answers that question. An 80 PLUS Gold label equals quality and capability is incorrect: it certifies efficiency only. Buying the largest unit available has no downside is incorrect: price, size, and low-load efficiency are real costs, even though sizing exactly to the edge is worse. An adapter can solve any connector problem is incorrect: adapters do not create 12V capacity or ATX 3.x compliance, and a missing native connector is a signal that the supply does not fit. A small processor TDP means processor power can be ignored is incorrect: maximum turbo power can approach double the base figure. The manufacturer's recommended wattage can safely be undercut is incorrect: it is normally a whole-system figure that already includes margin.

Frequently Asked Questions

Is the TDP figure on the card's page enough to size my power supply?

No. TDP is a thermal-design figure describing heat to be dissipated, not the board's electrical draw. Use the card manufacturer's official board-power figure, TGP or TBP, for the exact model, and add the processor's maximum turbo power or PPT plus an allowance for the rest of the system.

My power supply is rated 750W. Does that mean it delivers 750W to the graphics card?

No. The 750W rating is the total across all voltage rails. The graphics card and processor draw almost entirely from 12V, so the governing figure is the combined +12V wattage on the unit's own label, calculated as 12 multiplied by the +12V amps. Check that figure, not the box total.

What does ATX 3.0 add that an older power supply lacks?

ATX 3.0 specifies a short-duration power excursion requirement, for example 200% of rated power for 100 microseconds on units above 450W with a 12VHPWR connector. ATX 2.x units were never required to meet that requirement, so an older unit with adequate rated watts still carries no guarantee of riding through the same class of spike. That is why ATX 3.0 or 3.1 is favoured for newer high-end cards.

Can I use an adapter if my power supply lacks the connector my new card needs?

A missing native connector should be treated as a sign that the power supply does not fit the upgrade. An adapter does not increase the supply's 12V capacity and does not turn an ATX 2.x unit into an ATX 3.x unit. The connector check is part of the triple check precisely because watts alone cannot substitute for it.

Should I buy the biggest power supply available to be safe?

Not necessarily. Oversizing costs money and space, and a unit running far below its rating much of the time may not operate at its best efficiency. The recommended approach is the planning convention in this article: size so that heavy load sits at roughly 50 to 80 percent of rated capacity, with headroom for transients and a future upgrade, rather than sizing exactly to the edge or far beyond it.

Does a Gold efficiency rating guarantee the unit can drive a high-end card?

No. 80 PLUS certifies efficiency at specified loads, not build quality, not 12V capability, and not connector provision. A Gold unit with insufficient combined 12V wattage or the wrong connectors is still the wrong unit for the card, whatever its efficiency tier.

My processor's TDP is small. Can I leave it out of the budget?

No. Processor maximum turbo power or PPT can approach double the base figure on high-end models, and that difference is large enough to change the power supply size. Always include the processor's maximum figure for the exact installed model, verified against the manufacturer's specification.

For inspection enquiries, HK GPU Fix is at Dali Plaza, Room 1209, Shek Mun, telephone 2186 8010, WhatsApp 6558 6806. Inspection follows the published three-tier fee range of HK$400 to HK$2,800, based on a quote after inspection.