Adreno-GPU

Adreno (an anagram of Radeon, originating from AMD’s Imageon handheld graphics division acquired by Qualcomm in 2009) is Qualcomm’s family of integrated graphics processing units (GPUs) used across Snapdragon and Dragonwing SoCs. Adreno GPUs feature a unified shader model with a tile-based flex rendering architecture (FlexRender) that dynamically switches between immediate-mode and tile-based rendering, paired with Universal Bandwidth Compression (UBWC) and dedicated on-chip Graphics Memory (GMEM).

Adreno GPUs support OpenGL ES, OpenCL, Vulkan, and DirectX 11/12 on the proprietary driver stack, while the open-source Freedreno stack inside Mesa adds full desktop OpenGL 4.6 and Vulkan 1.3/1.4 support.

Adreno GPU Userspace Driver

Turnip (vulkan.freedreno) is widely used both on mainline Linux SBCs/laptops and inside Android containers/emulators (via the kgsl backend) for DirectX-to-Vulkan translation layers like DXVK and VKD3D-Proton.

Vulkan Support Timeline

Adreno GPU Kernel Driver

Qualcomm Adreno GPUs & Generations (600 Series and Upwards)

Adreno X Series (PC / Compute)

Designed for Qualcomm’s Oryon-based Snapdragon X PC and industrial compute platforms, featuring wide 128-bit to 192-bit LPDDR5X memory interfaces and full DirectX 12.1 / Vulkan 1.3+ support.

Note: In freedreno_devices.py, Adreno X1-85 is defined under CHIP.A7XX / a7xx_gen2 (internally Adreno 741) and Adreno X1-45 is defined under CHIP.A7XX / a7xx_gen2 (internally Adreno 726, derived from Adreno 725). Both are supported by Mesa Freedreno/Turnip starting from Mesa 24.2.

Adreno 800 Series (2024–2026)

Introduced in late 2024, the Adreno 800 (a8xx) generation transitions to a sliced architecture where shaders and dedicated L2/GMEM cache blocks are partitioned into independent slices, alongside UBWC v6 and Vulkan 1.3/1.4 support.

Adreno 700 Series (2021–2024)

Introduced with the Snapdragon 8 Gen 1, the Adreno 700 (a7xx) family doubled FP32 ALU density per shader block and introduced hardware-accelerated Ray Tracing starting with the Adreno 740.

Note: Based on Mesa’s freedreno_devices.py:

• Adreno 702 is marketed in the 700 series, but in freedreno_devices.py it is defined under CHIP.A6XX with [a6xx_base, a6xx_gen1_low] (alongside Adreno 610/612) rather than CHIP.A7XX, with a halved register file (reg_size_vec4 = 48) and stripped-down hardware blocks (# A702 cuts A LOT of things..).
• Within CHIP.A7XX, Mesa subdivides the family into a7xx_gen1 (Adreno 710, 720, 730), a7xx_gen2 (Adreno 725, 732, 740, and X1-85), and a7xx_gen3 (Adreno 735, 750). Hardware ray intersection (has_ray_intersection = True) is only present on Adreno 735, 740, and 750.

Adreno 600 Series (2018–2023)

Introduced in 2018 with the Snapdragon 845, the Adreno 600 (a6xx) architecture unified shader and compute pipelines with Direct3D 12_1, OpenGL ES 3.2, OpenCL 2.0/3.0, and Vulkan 1.1–1.3 support. It enjoys the most mature open-source support in Mesa’s Freedreno and Turnip drivers.

Note: Model numbers in the Adreno 600 series do not strictly reflect sub-architecture generation. In Mesa’s freedreno_devices.py, CHIP.A6XX is divided into five sub-generations:

• a6xx_gen1_low (reg_size_vec4 = 48): Adreno 610, 612, and 702
• a6xx_gen1 (reg_size_vec4 = 96): Adreno 615, 616, 618, 619, 620, and 630
• a6xx_gen2 (reg_size_vec4 = 96): Adreno 640 and 680
• a6xx_gen3 (reg_size_vec4 = 64): Adreno 613 and 650 (note that Adreno 613 is gen3 while 615–620 are gen1)
• a6xx_gen4 (reg_size_vec4 = 64, instr_cache_size = 127): Adreno 642L, 643, 644, 660, 663, and 690 (making Adreno 642L/643/644 architecturally newer gen4 parts than the gen2 Adreno 640 and gen3 Adreno 650)

Sources: Mesa — freedreno_devices.py, Wikipedia — Adreno, and Mesa 3D — Freedreno Driver Documentation