Integrated GPU

On this page 7

An integrated GPU (iGPU) is a graphics processing unit built into the same chip or package as the CPU, which uses the computer’s shared system memory instead of dedicated video memory. It renders the desktop, decodes and encodes video, drives displays and runs games and compute tasks within a much smaller power budget than a separate graphics card.

In computer hardware and graphics, the integrated GPU is the counterpart of the discrete GPU, which sits on its own board with its own VRAM. Integrated graphics power most laptops, office desktops, tablets and phones, and on modern system-on-chip designs they have become capable enough for mainstream gaming, creative work and machine learning inference.

How an integrated GPU works

An integrated GPU contains the same basic building blocks as any GPU: shader or compute units for parallel arithmetic, texture units, raster and output stages and fixed-function media engines for video. The difference lies in what surrounds it: it shares the die or package with CPU cores and connects to the same memory controller.

Because the CPU and GPU address the same physical memory, data does not need to be copied across a PCI Express link. Designs with a unified memory architecture let both processors work on the same buffers directly, which cuts latency and power for mixed workloads.

AspectIntegrated GPUEffect
MemoryShared system RAM, part reserved or allocated dynamicallyNo separate VRAM cost, competes with the CPU for bandwidth
Power budgetShares package power, often a few watts to about 30 WLong battery life, limited peak performance
CoolingSame cooler as the CPUSustained load can reduce CPU boost
Media engineHardware decode and encode of H.264, HEVC, AV1Efficient playback and streaming
Display outputsRouted through the motherboard or USB-CMultiple monitors without an extra card
Key characteristics of integrated graphics

Memory bandwidth as the main constraint

The most important limit of an integrated GPU is memory bandwidth. A GPU needs a constant stream of textures and geometry, and shared DDR or LPDDR memory provides far less bandwidth than the dedicated GDDR or HBM on a graphics card.

ConfigurationApproximate bandwidthTypical device
Dual-channel DDR4-3200About 51 GB/sOlder desktops and laptops
Dual-channel DDR5-5600About 90 GB/sCurrent desktops
128-bit LPDDR5XAbout 120 to 135 GB/sThin laptops
Wide unified memory, 256 to 512 bitsAbout 200 to 800 GB/sHigh-end system-on-chip laptops
Discrete midrange GDDR6About 250 to 500 GB/sSeparate graphics cards
Approximate memory bandwidth by configuration

In practice, memory configuration strongly affects iGPU performance. A laptop with a single memory module runs in single-channel mode and can lose a large share of graphics performance compared with an identical model using two modules.

Integrated versus discrete graphics

The choice between integrated and discrete graphics is a trade-off between efficiency, cost and size on one side and peak performance on the other. Many laptops include both and switch dynamically, using the iGPU for the desktop and the discrete GPU for demanding applications.

CriterionIntegrated GPUDiscrete GPU
Peak performanceLow to moderateModerate to very high
Power consumptionVery lowRoughly 50 to over 400 W
MemoryShared system RAMDedicated VRAM, 8 to 32 GB or more
CostIncluded with the processorSeparate component
UpgradeabilityFixed with the CPUCard can be replaced
Large model memoryCan use most of a large unified poolLimited by VRAM size
Integrated GPU compared with a discrete GPU

Large unified memory pools give integrated designs one notable advantage: a machine with 64 GB or more of unified memory can load AI models that would not fit into the VRAM of many consumer graphics cards, even though it computes them more slowly.

What integrated graphics handle well

Integrated GPUs are fully adequate for office work, web browsing, video playback and multiple displays. Recent designs also deliver playable frame rates in many games at 1080p with reduced settings, especially with upscaling technologies.

WorkloadSuitabilityNote
Office and browsingExcellentHardware acceleration keeps the CPU free
4K and 8K video playbackExcellentDepends on codec support in the media engine
Esports and older gamesGoodOften 60 fps or more at 1080p
Recent demanding gamesLimitedLow settings, upscaling required
Professional 3D and large training jobsPoorBandwidth and compute limits
Suitability of integrated graphics by workload

Common pitfalls and misconceptions

  • Reported shared memory is not dedicated VRAM; the system lends RAM to the GPU and both compete for bandwidth.
  • Single-channel memory cripples integrated graphics, so matched pairs of modules matter.
  • The same GPU name can perform differently depending on power limits and cooling in each laptop.
  • Outdated drivers cause many game crashes and missing features.
  • Some desktop processors have no iGPU, so a separate card is needed even for basic display output.

A plain-language summary

An integrated GPU is the graphics part that lives inside the main processor instead of on a separate card. It is like a kitchen built into a studio apartment rather than a separate restaurant kitchen: compact, efficient and enough for everyday meals.

For browsing, video, office work and lighter games it does the job well while saving power. For heavy 3D work or the newest games, a separate graphics card is still much stronger.

An example from practice

A school district buys 400 laptops for students and teachers. The models chosen use processors with integrated graphics and 16 GB of dual-channel memory, with no discrete GPU. The machines run video calls, presentation software, browser-based learning tools and short video editing projects.

Because there is no separate graphics card, each laptop costs less, weighs less and lasts a full school day on one charge. The hardware media engine handles video lessons without heating the device. Only the media classroom receives a few desktops with discrete GPUs for 3D animation, where the integrated graphics would be too slow.