View source for Engineering:Turing (microarchitecture)

{{Short description|GPU microarchitecture by Nvidia}}
{{Infobox GPU microarchitecture
| name                   = Turing
| image                  = Nvidia@12nm@Turing@TU104@GeForce RTX 2080@S TAIWAN 1841A1 PKYN44.000 TU104-400-A1 DSCx3.jpg
| image_size             = 250px
| caption                = 
| alt                    = 
| launched               = {{Start date and age|2018|09|20}}
| discontinued           = 
| soldby                 = 
| designfirm             = [[Company:Nvidia|Nvidia]]
| manuf1                 = [[Company:TSMC|TSMC]]
| process                = TSMC 12FFC
| codename               = TU10x <br/> TU11x

<!------------------  Product Series  ------------------->
| products-desktop1      = [[GeForce 16 series|GeForce GTX 16 series]] <br/> [[GeForce 20 series|GeForce RTX 20 series]]
| products-hedt1         = [[Engineering:Quadro|Quadro RTX]]
| products-server1       = [[Engineering:Nvidia Tesla|Tesla T4]]

<!------------------  Supported Graphics APIs  ------------------->
| directx-version        = [[Software:DirectX#DirectX 12 Ultimate|DirectX 12 Ultimate (Feature Level 12 2)]]
| direct3d-version       = [[Software:Direct3D#Direct3D 12|Direct3D 12.0]]
| shadermodel-version    = [[High-Level Shader Language|Shader Model 6.7]]
| opencl-version         = [[OpenCL#OpenCL 3.0|OpenCL 3.0]]
| opengl-version         = [[Software:OpenGL#OpenGL 4.6|OpenGL 4.6]]
| opengles-version       = 
| cuda-version           = [[Software:CUDA#Version features and specifications|Compute Capability 7.5]]
| optix-version          = 
| mantle-api             = 
| vulkan-api             = [[Software:Vulkan#Vulkan 1.3|Vulkan 1.3]]

<!------------------  Supported Compute APIs  ------------------->
| opengl-compute-version = 
| cuda-compute-version   = 
| directcompute-version  = 

<!------------------  Specifications  ------------------->
| compute                = {{ubl |28.5{{nbsp}}TFLOPS (FP16) |14.2{{nbsp}}TFLOPS (FP32)}}<ref name="TuringWhitepaper">{{cite web |title=Nvidia Turing GPU Architecture: Graphics Reinvented |url=https://www.nvidia.com/content/dam/en-zz/Solutions/design-visualization/technologies/turing-architecture/NVIDIA-Turing-Architecture-Whitepaper.pdf |website=Nvidia |language=en-US |year=2018 |access-date=June 28, 2019}}</ref>
| slowest                = 
| slow-unit              = 
| fastest                = 
| fast-unit              = 
| shader-clock           = 
| l0-cache               = 
| l1-cache               = 96{{nbsp}}KB (per SM)
| l2-cache               = 2{{nbsp}}MB to 6{{nbsp}}MB
| l3-cache               = 
| memory-support         = GDDR6 <br/> [[High Bandwidth Memory#HBM 2|HBM2]]
| memory-clock           = 
| pcie-support           = [[Engineering:PCI Express#PCI Express 3.0|PCIe 3.0]]

<!------------------  Media  Engine ------------------->
| encode-codec           = {{hlist |[[Advanced Video Coding|H.264]] |[[High Efficiency Video Coding|H.265]]}}
| decode-codec           = {{hlist |[[Advanced Video Coding|H.264]] |[[High Efficiency Video Coding|H.265]]}}
| color-depth            = {{hlist |8-bit |10-bit}}
| encoders               = NVENC
| display-outputs        = {{ubl |[[DisplayPort#1.4a|DisplayPort 1.4a]] |[[Engineering:HDMI#Version 2.0|HDMI 2.0b]] |[[DisplayPort#USB-C|USB-C]]}}

<!------------------  History  ------------------->
| predecessor            = [[Engineering:Pascal (microarchitecture)|Pascal]]
| variant                = [[Engineering:Volta (microarchitecture)|Volta]] (professional)
| successor              = [[Engineering:Ampere (microarchitecture)|Ampere]]
}}
'''Turing''' is the codename for a [[Graphics processing unit|graphics processing unit]] (GPU) microarchitecture developed by [[Company:Nvidia|Nvidia]]. It is named after the prominent mathematician and computer scientist [[Biography:Alan Turing|Alan Turing]]. The architecture was first introduced in August 2018 at SIGGRAPH 2018 in the workstation-oriented [[Engineering:Nvidia Quadro#Quadro RTX|Quadro RTX]] cards,<ref>{{cite web |last=Smith |first=Ryan |date=August 13, 2018 |title=NVIDIA Reveals Next-Gen Turing GPU Architecture: NVIDIA Doubles-Down on Ray Tracing, GDDR6, & More |url=https://www.anandtech.com/show/13214/nvidia-reveals-next-gen-turing-gpu-architecture |website=AnandTech |language=en-US |access-date=April 9, 2023}} </ref> and one week later at Gamescom in consumer [[GeForce 20 series|GeForce RTX 20 series]] graphics cards.<ref>{{cite news |last=Smith |first=Ryan |date=August 20, 2018 |title=NVIDIA Announces the GeForce RTX 20 Series: RTX 2080 Ti & 2080 on Sept. 20th, RTX 2070 in October |url=https://www.anandtech.com/show/13249/nvidia-announces-geforce-rtx-20-series-rtx-2080-ti-2080-2070 |work=AnandTech |language=en-US |access-date=April 9, 2023}}</ref> Building on the preliminary work of its [[Supercomputer|HPC]]-exclusive predecessor, the Turing architecture introduces the first consumer products capable of real-time [[Ray tracing (graphics)|ray tracing]], a longstanding goal of the computer graphics industry. Key elements include dedicated [[Artificial intelligence|artificial intelligence]] processors ("Tensor cores") and dedicated ray tracing processors ("RT Cores"). Turing leverages [[DirectX Raytracing|DXR]], [[Software:OptiX|OptiX]], and [[Software:Vulkan (API)|Vulkan]] for access to ray-tracing. In February 2019, Nvidia released the [[GeForce 16 series]] of GPUs, which utilizes the new Turing design but lacks the RT and Tensor cores.

Turing is manufactured using [[Company:TSMC|TSMC]]'s 12 nm [[Engineering:FinFET|FinFET]] semiconductor fabrication process. The high-end TU102 GPU includes 18.6{{nbsp}}billion [[Engineering:Transistor count|transistors]] fabricated using this process.<ref name="TuringWhitepaper"/> Turing also uses GDDR6 memory from [[Company:Samsung Electronics|Samsung Electronics]], and previously [[Company:Micron Technology|Micron Technology]].

== Details ==
[[File:Nvidia@12nm@Turing@TU104@GeForce RTX 2080@S TAIWAN 1841A1 PKYN44.000 TU104-400-A1 DSCx7 poly@5xExt.jpg|thumb|Die shot of the TU104 GPU used in RTX 2080 cards]]
[[File:Nvidia@12nm@Turing@TU106@GeForce RTX 2060@S TAIWAN 1844A1 PM4F79.000 TU106-200A-KA-A1 DSCx5 poly@5xExt.jpg|thumb|Die shot of the TU106 GPU used in RTX 2060 cards]]
[[File:Nvidia@12nm@Turing@TU116@GeForce GTX 1660@S TAIWAN 1941A1 PPHV26.000 TU116-300-A1 DSCx5 poly@5xExt.jpg|thumb|Die shot of the TU116 GPU used in GTX 1660 cards]]
The Turing microarchitecture combines multiple types of specialized [[Multi-core processor|processor core]], and enables an implementation of limited real-time ray tracing.<ref>{{Cite news |last=Warren |first=Tom |date=August 20, 2018 |title=Nvidia announces RTX 2000 GPU series with '6 times more performance' and ray-tracing |url=https://www.theverge.com/2018/8/20/17758724/nvidia-geforce-rtx-2080-specs-pricing-release-date-features |work=The Verge |language=en-US |access-date=August 20, 2018}}</ref> This is accelerated by the use of new RT (ray-tracing) cores, which are designed to process [[Quadtree|quadtree]]s and spherical hierarchies, and speed up collision tests with individual triangles.

Features in Turing:
* [[Software:CUDA|CUDA]] cores (SM, Streaming Multiprocessor)
** Compute Capability 7.5
** traditional [[Engineering:Rasterisation|rasterized]] shaders and compute
** concurrent execution of integer and floating point operations (inherited from Volta)
* Ray-tracing (RT) cores
** [[Bounding volume hierarchy|bounding volume hierarchy]] acceleration<ref name=anandarch>{{cite web |last=Oh |first=Nate |date=September 14, 2018 |title=The NVIDIA Turing GPU Architecture Deep Dive: Prelude to GeForce RTX |url=https://www.anandtech.com/show/13282/nvidia-turing-architecture-deep-dive/ |website=AnandTech |language=en-US |access-date=April 9, 2023}}</ref>
** shadows, [[Ambient occlusion|ambient occlusion]], lighting, reflections
* Tensor (AI) cores
** artificial intelligence
** large [[Matrix (mathematics)|matrix]] operations
** Deep Learning Super Sampling (DLSS)
* Memory controller with GDDR6/HBM2 support
*[[DisplayPort]] 1.4a with Display Stream Compression (DSC) 1.2
* PureVideo Feature Set J hardware video decoding
* GPU Boost 4
* [[Engineering:NVLink|NVLink]] Bridge with [[VRAM]] stacking pooling memory from multiple cards
* [[Engineering:VirtualLink|VirtualLink]] [[Virtual reality|VR]]
* NVENC hardware encoding

The GDDR6 memory is produced by [[Company:Samsung Electronics|Samsung Electronics]] for the Quadro RTX series.<ref>{{cite news |last1=Mujtaba |first1=Hassan |date=August 14, 2018 |title=Samsung GDDR6 Memory Powers NVIDIA's Turing GPU Based Quadro RTX Cards |url=https://wccftech.com/nvidia-quadro-rtx-turing-gpu-samsung-gddr6-memory/ |work=Wccftech |language=en-US |access-date=April 9, 2023}}</ref> The RTX 20 series initially launched with [[Company:Micron Technology|Micron]] memory chips, before switching to Samsung chips by November 2018.<ref>{{cite news |last1=Maislinger |first1=Florian |date=November 21, 2018 |title=Faulty RTX 2080 Ti: Nvidia switches from Micron to Samsung for GDDR6 memory |url=https://www.pcbuildersclub.com/en/2018/11/faulty-rtx-2080-ti-nvidia-switches-from-micron-to-samsung-for-gddr6-memory/ |work=PC Builder's Club |access-date=July 15, 2019}}</ref>

=== Rasterization ===
Nvidia reported rasterization (CUDA) performance gains for existing titles of approximately 30–50% over the previous generation.<ref name=Keynote>{{cite web |title=#BeForTheGame |url=https://www.twitch.tv/videos/299680425 |website=Twitch |language=en-US}}</ref><ref name=fisher>{{cite web |last=Fisher |first=Jeff |date=August 20, 2018 |title=GeForce RTX Propels PC Gaming's Golden Age with Real-Time Ray Tracing |url=https://blogs.nvidia.com/blog/2018/08/20/geforce-rtx-real-time-ray-tracing/ |website=Nvidia |language=en-US |access-date=April 9, 2023}}</ref>

=== Ray-tracing ===
The ray-tracing performed by the RT cores can be used to produce reflections, refractions and shadows, replacing traditional raster techniques such as [[Cube mapping|cube maps]] and [[Depth map|depth map]]s. Instead of replacing rasterization entirely, however, the information gathered from ray-tracing can be used to augment the shading with information that is much more photo-realistic, especially in regards to off-camera action. Nvidia said the ray-tracing performance increased about 8 times over the previous consumer architecture, Pascal.

=== Tensor cores ===
Generation of the final image is further accelerated by the Tensor cores, which are used to fill in the blanks in a partially rendered image, a technique known as de-noising. The Tensor cores perform the result of [[Deep learning|deep learning]] to codify how to, for example, increase the resolution of images generated by a specific application or game. In the Tensor cores' primary usage, a problem to be solved is analyzed on a supercomputer, which is taught by example what results are desired, and the supercomputer determines a method to use to achieve those results, which is then done with the consumer's Tensor cores. These methods are delivered via [[Device driver|driver updates]] to consumers.<ref name=Keynote/> The supercomputer uses a large number of Tensor cores itself.

== Turing dies ==
{| class="wikitable" style="font-size:100%; text-align:center;"
|+Comparison of Turing dies
! Die
! [https://www.techpowerup.com/gpu-specs/nvidia-tu102.g813 TU102]
! [https://www.techpowerup.com/gpu-specs/nvidia-tu104.g854 TU104]
! [https://www.techpowerup.com/gpu-specs/nvidia-tu106.g875 TU106]
! [https://www.techpowerup.com/gpu-specs/nvidia-tu116.g902 TU116]
! [https://www.techpowerup.com/gpu-specs/nvidia-tu117.g881 TU117]
|-
! Die size
| 754 mm<sup>2</sup>
| 545 mm<sup>2</sup>
| 445 mm<sup>2</sup>
| 284 mm<sup>2</sup>
| 200 mm<sup>2</sup>
|-
! Transistors
| 18.6B
| 13.6B
| 10.8B
| 6.6B
| 4.7B
|-
! Transistor density
| 24.7 MTr/mm<sup>2</sup>
| 25.0 MTr/mm<sup>2</sup>
| 24.3 MTr/mm<sup>2</sup>
| 23.2 MTr/mm<sup>2</sup>
| 23.5 MTr/mm<sup>2</sup>
|-
! Graphics processing<br />clusters (GPC)
| 6
| 6
| 3
| 3
| 2
|-
! Streaming<br />multiprocessors (SM)
| 72
| 48
| 36
| 24
| 16
|-
! [[Unified shader model|CUDA cores]]
| 4608
| 3072
| 2304
| 1536
| 1024
|-
! [[Texture mapping unit]]s
| 288
| 192
| 144
| 96
| 64
|-
! [[Render output unit]]s
| 96
| 64
| 64
| 48
| 32
|-
! [[Deep learning super sampling#Architecture|Tensor cores]]
| 576
| 384
| 288
| colspan="2" rowspan="2" {{N/a}}
|-
! [[Nvidia RTX|RT cores]]
| 72
| 48
| 36
|-
! rowspan="2" | L1 [[Cache (computing)#Examples of hardware caches|cache]]
| 6.75&nbsp;MB
| 4.5&nbsp;MB
| 3.375&nbsp;MB
| 2.25&nbsp;MB
| 1.5&nbsp;MB
|-
| colspan="5" | 96&nbsp;KB per SM
|-
! L2 cache
| 6&nbsp;MB
| 4&nbsp;MB
| 4&nbsp;MB
| 1.5&nbsp;MB
| 1&nbsp;MB
|-
! Max [[Thermal design power|TDP]]
| 280&nbsp;W
| 250&nbsp;W
| 185&nbsp;W
| 125&nbsp;W
| 75&nbsp;W
|}

== Development ==
{{Main|Nvidia RTX}}
Turing's [[Software development kit|development platform]] is called RTX. RTX ray-tracing features can be accessed using [[Company:Microsoft|Microsoft]]'s [[DirectX Raytracing|DXR]], [[Software:OptiX|OptiX]], as well using [[Software:Vulkan (API)|Vulkan]] extensions (the last one being also available on Linux drivers).<ref name=nvidiartx>{{cite web |title=NVIDIA RTX platform |url=https://developer.nvidia.com/rtx |website=Nvidia |language=en-US |date=July 20, 2018 |access-date=April 9, 2023}}</ref> It includes access to AI-accelerated features through NGX. The Mesh Shader, Shading Rate Image functionalities are accessible using DX12, Vulkan and [[Software:OpenGL|OpenGL]] extensions on Windows and Linux platforms.<ref>{{cite web |title=Turing Extensions for Vulkan and OpenGL |url=https://developer.nvidia.com/vulkan-turing |website=Nvidia |language=en-US |date=September 11, 2018 |access-date=April 9, 2023}}</ref>

Windows 10 October 2018 update includes the public release of DirectX Raytracing.<ref>{{Cite web |last=Pelletier |first=Sean |date=October 2, 2018 |title=Windows 10 October 2018 Update a Catalyst for Ray-Traced Games |url=https://blogs.nvidia.com/blog/2018/10/02/real-time-ray-tracing-rtx-windows-10-october-update/ |website=Nvidia |language=en-US |access-date=April 9, 2023}}</ref><ref>{{Cite web |last=van Rhyn |first=Jacques |date=October 2, 2018 |title=DirectX Raytracing and the Windows 10 October 2018 Update |url=https://blogs.msdn.microsoft.com/directx/2018/10/02/directx-raytracing-and-the-windows-10-october-2018-update/ |website=Microsoft |language=en-US |access-date=April 9, 2023}}</ref>

== Products using Turing ==
* [[Engineering:List of Nvidia graphics processing units#GeForce MX series|GeForce MX series]]
** GeForce MX450 (Mobile)
** GeForce MX550 (Mobile)
* [[GeForce 16 series]]
** GeForce GTX 1630
** GeForce GTX 1650 (Mobile)
** GeForce GTX 1650
** GeForce GTX 1650 Super
** GeForce GTX 1650 Ti (Mobile)
** GeForce GTX 1660
** GeForce GTX 1660 Super
** GeForce GTX 1660 Ti (Mobile)
** GeForce GTX 1660 Ti
* [[GeForce 20 series]]
** GeForce RTX 2060 (Mobile)
** GeForce RTX 2060
** GeForce RTX 2060 Super
** GeForce RTX 2070 (Mobile)
** GeForce RTX 2070
** GeForce RTX 2070 Super (Mobile)
** GeForce RTX 2070 Super
** GeForce RTX 2080 (Mobile)
** GeForce RTX 2080
** GeForce RTX 2080 Super (Mobile)
** GeForce RTX 2080 Super
** GeForce RTX 2080 Ti
** Titan RTX
* [[Engineering:Quadro|Nvidia Quadro]]
** Quadro RTX 3000 (Mobile)
** Quadro RTX 4000 (Mobile)
** Quadro RTX 4000
** Quadro RTX 5000 (Mobile)
** Quadro RTX 5000
** Quadro RTX 6000 (Mobile)
** Quadro RTX 6000
** Quadro RTX 8000
** Quadro T1000 (Mobile)
** Quadro T2000 (Mobile)
** T400
** T400 4GB
** T500 (Mobile)
** T600 (Mobile)
** T600
** T1000
** T1000 8GB
** T1200 (Mobile)
* [[Engineering:Nvidia Tesla|Nvidia Tesla]]
** Tesla T4

==See also==
* [[Engineering:List of eponyms of Nvidia GPU microarchitectures|List of eponyms of Nvidia GPU microarchitectures]]
*[[Engineering:List of Nvidia graphics processing units|List of Nvidia graphics processing units]]
*[[Engineering:Volta (microarchitecture)|Volta (microarchitecture)]]

==References==
{{reflist}}

==External links==
*[https://www.nvidia.com/content/dam/en-zz/Solutions/design-visualization/technologies/turing-architecture/NVIDIA-Turing-Architecture-Whitepaper.pdf Nvidia Turing GPU Architecture Whitepaper]
*[https://www.nvidia.com/en-us/geforce/turing/ Nvidia page about Turing]
*[https://blogs.nvidia.com/blog/2018/03/19/whats-difference-between-ray-tracing-rasterization/ Nvidia blog about raytracing vs. rasterization]
*[https://devblogs.nvidia.com/nvidia-turing-architecture-in-depth/ NVIDIA Turing Architecture In-Depth]
*[https://blogs.msdn.microsoft.com/directx/2018/03/19/announcing-microsoft-directx-raytracing/ Microsoft developer blog on DirectX Raytracing]

{{Nvidia}}

[[Category:Graphics microarchitectures|Nvidia Turing]]

{{Sourceattribution|Turing (microarchitecture)}}