Since the acquisition of the ATI by AMD, the graphics core of ATI products can be found in later AMD CPUs and APUs. Later AMD retiring the ATI brand in favor of the AMD name in 2010.

R100 to R500 series were based on the fixed pipelines architecture, while later GPU families were based on unified shader architectures; HD2000 to HD6000 series were based on the Terascale architecture, while HD7000 and later GPU families were based on the newer Graphics Core Next architecture.

GPU families based on the Graphics Core Next architecture support AMD's Mantle.

Generations

Fixed Pipeline based

R100 series (7xxx)

The very first Radeon GPU family which is based on the Charisma Engine architecture. They were made in 180 nm process (except RV200, which is made in the smaller 150 nm.) It contains ATI's Video Immersion video acceleration technology, which also available on R200 and R300 series GPUs.

The R100 introduces HyperZ technology, which increase memory bandwidth and fill-rate efficiency. The HyperZ technology is also available on any other Radeon GPU families, including the Graphics Core Next-based GPU families.

All GPU in this family supports DirectX 7.0 and OpenGL 1.3 (T&L only).

R200 series (8xxx/9xxx)

The second generation Radeon GPU family which is based on the Charisma Engine II architecture. It has improved version of Hyper Z (known as HyperZ II) and It is capable of dual display output through ATI's HydraVision (Later known as AMD Eyefinity).

It supports ATI's TruForm hardware tessellation, but it was never been accepted into DirectX or OpenGL specifications.

All model supports DirectX 8.1, Shader Model 1.4 and OpenGL 1.3. All models were made in 150 nm process.

R300 series (9xxx/X3xx/X5xx/X6xx/X10xx)

Also made in 150 nm process (with RV350 and RV360 made in 130 nm and RV370 made in 110 nm), this GPU family supports DirectX 9.0, Shader Model 2.0 and OpenGL 2.0.

The high-end cards has wider 256-bit of memory bus width. Some cards in the GPU family also supports PCI-Express interface.

R400 series (X7xx/X8xx)

Made in both 110 nm and 130 nm process, this GPU family supports DirectX 9.0b, OpenGL 2.0 and Shader Model 2.0b.

R500 series (X1xxx)

Made in the smaller 90 nm and 80 nm process, this GPU family supports DirecX 9.0C, OpenGL 2.0 and Shader Model 3.0.

TeraScale based

HD 2000 and HD 3000 series (R600)

Made in the smaller 65 nm process (with some made in 80 nm process), it is the first Radeon GPU family to be based on the TeraScale, ATI's first unified shader model architecture. It features the 5-way VLIW designs.

Video Immersion, the video acceleration engine that ATI introduce since the very first Radeon GPU familiy, is replaced by the Unified Video Decoder (UVD) SIP core for decoding videos.

The Radeon HD 3870 X2 (R680) is the very first dual-GPU graphics card for the Radeon GPU family.

The GPU family supports DirectX 10.0, OpenGL 3.3 and Shader Model 4.0.

HD 4000 series (R700)

Made in 55 nm and 40 nm process, it is the die-shrunken TeraScale-based GPU family. This is the first Radeon GPU family to utilize GDDR5 memory.

Those supports DirectX 10.1, OpenGL 3.3, OpenCL 1.0 and Shader Model 4.1.

HD 5000 series (Evergreen)

Made in the 40 nm process, it is the GPU family based on the 2nd-generation of TeraScale architecture.

The HD 5450 to HD 5770 desktop GPUs are based on the 1st-generation TeraScale, while others are based on the 2nd-generation TeraScale Architecture. On Mobile GPUs, only the Mobility Radeon HD 530v to HD 5165 are 1st-generation TeraScale-based.

Most of the GPUs in this GPU family supports DirectX 11.0, OpenGL 4.3 (4.4 with later driver support), OpenCL 1.1 and Shader Model 5.0.

HD 6000 series (Northern Islands)

Still made in 40 nm process, it is the GPU family which introduces GPUs based on the 3rd-generation TeraScale architecture. The improved architecture replaces the 5-way VLIW designs with the 4-way VLIW designs.

Only the HD 69xx desktop GPUs are 3rd-generation TeraScale-based GPUs, others are based on the 2nd-generation TeraScale.

As with HD 5000 series, it supports DirectX 11.0, OpenGL 4.3 (4.4 with later driver support), OpenCL 1.2 and Shader Model 5.0.

Graphics Core Next based

HD 7000 series (Southern Islands)

Made in the smaller 28 nm for GCN-based GPUs and 40 nm for TeraScale-based GPUs, it served as the introduction of the Graphics Core Next architecture.

HD 7700 to HD 7900 desktop graphic cards (and HD 7730M to HD 7970M for its mobile counterparts) are GCN-based while others (including some integrated graphics for AMD APUs) were TeraScale-based. For GCN-based GPUs, AMD also added the Video Codec Engine (VCE) in addition of Unified Video Decoder for video acceleration. Only the Bonaire XT-based HD 7790 desktop graphics card is based on the second-generation GCN architecture.

Only the GCN-based GPUs supports Direct3D 12.0 (feature level 11.1 on all but Bonaire XT-based GPU which has feature level 12_0 support) , Shader Model 6.0, OpenCL 2.0, OpenGL 4.5 and Vulkan 1.0. This GPU family does not support Mantle unlike the later Rx 200 series.

HD 8000 series

The Radeon HD 8000 series is a OEM-only GPU family, consists of previous GPU family based on either GCN or TeraScale.

Rx 200 series (Volcanic Islands)

The Rx 200 series serves as the introduction of the 2nd-generation GCN-based GPUs (R7 260, R9 290, R9 290X, R9 295X2 and R9 M280X)as well as 3rd generation GCN-based GPUs (R9 285 and R9 M295X)

Only R5 240 to R9 295X2 for desktop, all mobile GPUs and IGPs for Steamroller, Jaguar and Puma-based APUs are Graphics Core Next-Based. Others GPUs are TeraScale-based.

Only those that were based on the 2nd-generation and 3rd-generation GCN architectures supports AMD's TrueAudio audio processing technology.

In addition, the Radeon R9 295X2, the dual-GPU graphics cards, is the first graphics card to utilizes liquid cooling design in its reference design. (This is due to being a single Hawaii GPU (R9 290 and R9 290X) is already running hot)

All models supports DirectX 12.0 (Feature level 11_0 on 1st-gen GCN GPUs and 12_0 on 2nd-gen and 3rd-gen GCN GPUs), Shader Model 5.0 (6.0 on GCN-based GPUs), OpenCL 1.2 (2.0 for 2nd-gen and 3rd-gen GCN GPUs), OpenGL 4.5 and Vulkan 1.0. This GPU family also support Mantle. Only the 3rd-gen GCN GPU supports hardware-based color compression.

Rx 300 series (Volcanic Islands)

The Rx 300 series is the GCN-based Radeon graphics card line-up released in 2015.

It consists of rebrands of earlier GCN-based AMD Radeon products, with the exceptions of the Fiji-based graphics cards: the R9 Fury, E9 Fury X and R9 Nano. They are the first to introduce the first-generation High-Bandwidth Memory (HBM) module, a type of vertically-stacked memory that has the bandwidth of 512 GB/s.

All models supports DirectX 12.0 (Feature level 11_0 on most GPUs and 12_0 on 2nd-gen and 3rd-gen GCN GPUs), Shader Model 6.0, OpenCL 1.2 (2.0 on 2nd-gen and 3rd-gen GCN GPUs), OpenGL 4.5 and Vulkan 1.0. This GPU family also support Mantle. Only the 3rd-gen GCN GPUs supports hardware-based color compression.

Radeon 400 series

The Radeon 400 series is the GCN-based Radeon graphics card line-up released in 2016. It serves as the introduction of the 4th-generation GCN-based GPUs and the first die-strink (from 28 nm to 14 nm) since the 1st-gen GCN GPUs.

The 4th-generation GCN-based GPUs were designed to be more power-efficient than the earlier generations. It also features the updated hardware schedulers, a new primitive discard accelerator, a new display controller and an updated UVD (which can decodes HEVC at 4K resolutions at 60 frames per second). The new iteration of the GCN architecture also supports DisplayPort 1.3 HBR and HDMI 2.0b outputs.

All models supports DirectX 12.0 (Feature level 11_0 on most GPUs and 12_0 on 2nd-gen, 3rd-gen and 4th-gen GCN GPUs), Shader Model 6.0, OpenCL 1.2 (2.0 on 2nd-gen, 3rd-gen and 4th-gen GCN GPUs), OpenGL 4.5 and Vulkan 1.0. Only the 3rd-gen GCN GPUs and later supports hardware-based color compression.

Others

Fliper (seen on Nintendo GameCube)

Fliper is a customized GPU made for Nintendo's GameCube video game console. It contains 4 pixel pipelines and 1 texture unit and has 3 MB of video memory on chip.

Xenos (seen on Microsoft Xbox 360)

Xenos is the customized GPU made for Microsoft's Xbox 360. It is based on the R520 core. It has 10 MB of eDRAM framebuffer.

It is connected to the unified GDDR3 memory with 128-bit bus width. It only supports DirectX 9.0c.

Hollywood (seen on Nintendo Wii)

Hollywood is the customized GPU made for Nintendo's Wii. The GPU is actually a multi-chip module package, consists of two dies with one for graphics processing and one for audio processing.

Latte (seen on Nintendo Wii U)

Latte is a customized GPU made for Nintendo's Wii U. This GPU is based on the TeraScale architecture. Like Hollywood, it is a multi-chip module.

It is connected to the unified 2 GB of DDR3 memory with 32 MB of eDRAM memory as secondary.

Liverpool (seen on Sony PlayStation 4)

Liverpool is a customized GPU core seen in the custom APU of Sony's PlayStation 4. It is based on the GCN architecture. It contains 18 Compute units with total of 1,152 cores (64 cores per CU).

It is connected to the unified 8 GB of GDDR5 memory with 256-bit bus width.

Unlike ordinary graphics cards, it has an additional dedicated 20 GB/s bus that bypasses L1 and L2 GPU cache for direct system memory access.

It supports OpenGL 4.2 and PlayStation Shader Language for PS4.

Durango (seen on Microsoft Xbox One)

Durango is a customized GPU core seen in the custom APU of Microsoft's Xbox One. Like Liverpool, it is based on the GCN architecture, with 12 compute units (total of 768 cores).

It is connected to the 8 GB of unified DDR3 memory, and it has 32 MB of eSRAM in the memory subsystem in addition of the slower DDR3 memory.

The GPU only supports DirectX 11.x, a superset of DirectX 11.2 for Xbox One.