GPU Power Ratings Explained: TGP, TBP, TDP, Board Power, Power Limits, Connectors, and PSU Sizing

TGP, TBP, board power, GPU power, and TDP are often discussed as though they were interchangeable wattage labels. They are not. The scope of a published power figure depends on the vendor, product class, generation, and document that defines it. The safest interpretation is therefore the one attached to the exact product specification rather than a generic expansion of the acronym.

Current examples make the terminology difference visible. NVIDIA lists the GeForce RTX 5090 Founders Edition at 575 W Total Graphics Power. AMD lists the Radeon RX 9070 XT at 304 W Typical Board Power. Intel lists the Arc B580 at 190 W TBP. Those numbers describe vendor-defined product specifications; they do not establish that all three vendors calculate or apply the ratings identically.

Intel provides an unusually explicit terminology breakdown for Arc. Its January 2026 support documentation describes TGP for Arc notebook products as GPU power plus losses plus graphics-memory power, while desktop add-in-card TBP additionally includes non-graphics board loads such as fans and LEDs. Intel also states that software-reported GPU Power is only a subset of TGP or TBP on the covered Arc products.

That distinction is useful beyond Intel as a reading discipline, but Intel’s exact formula should not be projected onto AMD or NVIDIA products unless those vendors document the same scope. A monitoring overlay that reports “GPU power” also should not automatically be compared one-to-one with a board-level specification when the telemetry source measures a narrower domain.

TDP is used in computing as a thermal-design term, but its precise meaning is vendor- and product-specific. For a discrete graphics card, the useful planning number is normally the power specification the card vendor actually publishes for that exact model, such as NVIDIA Total Graphics Power, AMD Typical Board Power, or Intel TBP, rather than substituting an unrelated TDP figure from a chip database.

This also matters when comparing a GPU chip with a complete add-in board. Memory, voltage regulation, fans, lighting, and other board components can consume power outside the graphics silicon itself. A silicon-only or telemetry-only number therefore may answer a different question from the complete card’s board-power specification.

A GPU power specification is not a prediction that a complete PC will draw exactly that wattage from the wall. Whole-system input also includes the CPU, motherboard, memory, storage, cooling, peripherals powered by the system, and PSU conversion losses. Workload, frame rate, voltage/frequency behavior, thermals, firmware, and idle state can all change actual consumption over time.

Likewise, a card rated for a particular board or graphics power does not promise that every sample and workload will sit at that number continuously. Power-management controllers operate against product limits and workload demand. Short-duration behavior and measurement interval also matter, so a slow software graph, board telemetry, and instrumented rail measurement are not necessarily observing the same quantity.

A configurable power limit is part of the GPU’s control policy. It can constrain how much power the hardware is permitted to use within the vendor and board design, but it should not be assumed to redefine every published board-power term. Intel, for example, separately defines PL1 and PL2 as GPU SoC power limits on covered Arc products and notes that memory inclusion can depend on board design.

Board-partner cards can also differ from a reference design in clocks, firmware limits, cooling, connectors, and published power requirements. NVIDIA explicitly warns that add-in-card specifications can vary by manufacturer. When sizing or troubleshooting a specific retail card, use that exact board partner’s specification rather than silently substituting the reference-card value.

A supplementary power connector describes a supported delivery interface; the card’s power specification describes the product’s power behavior or design target under the vendor’s definition. One does not replace the other. A connector or cable with a high nominal capability does not mean the card continuously consumes that amount, and a GPU’s published board power is not by itself a complete description of how power reaches the card.

For example, NVIDIA lists the RTX 5090 Founders Edition with either an adapter fed by PCIe 8-pin cables or a PCIe Gen 5 cable while separately publishing its Total Graphics Power and Required System Power. Intel lists the Arc B580 reference card with one 8-pin connector while separately specifying 190 W TBP and a 600 W minimum PSU. Keep connector compatibility, board power, and system-PSU sizing as three related but distinct checks.

Do not turn GPU board power into a PSU recommendation by applying a universal percentage or multiplier. GPU vendors publish system-level recommendations with assumptions that can differ. NVIDIA’s RTX 5090 page, for example, states that its Required System Power figure is based on a particular Ryzen 9 9950X configuration and warns that requirements can differ with system configuration. AMD separately publishes a Minimum PSU Recommendation alongside Typical Board Power for the RX 9070 XT.

Start with the exact GPU or board-partner requirement, then account for the actual CPU and the rest of the system, connector/cable requirements, PSU documentation, and any intentional overclocking or future upgrade. A 300 W-class graphics-card specification does not imply that a 300 W PSU is sufficient, while a connector marked for substantially more power does not prove that the rest of the PSU is suitable.

First, identify the exact card, not just the GPU family. Second, record the vendor’s exact label and definition: TGP, Typical Board Power, TBP, GPU Power, or another term. Third, separate that specification from monitoring telemetry and from any adjustable firmware power limit. Fourth, check the exact card’s connector requirements and the vendor or board partner’s whole-system PSU guidance.

If two cards use different vendor terminology, compare the documented scopes before comparing the wattages. If the documentation does not define enough to make a precise apples-to-apples statement, keep the uncertainty explicit. That is more useful than inventing a conversion between TGP, TBP, and TDP that the vendors themselves do not specify.