New details on turing progress

Nvidia's Turing architecture is one of the biggest jumps in graphics in 20 years, we review the biggest innovations offered by this modern GPU architecture for gaming.

The most interesting news from Turing

CUDA 10: CUDA 10 includes support for Turing GPUs, performance-optimized libraries, a new asynchronous task graphics programming model, CUDA interoperability and improved graphics API, and new development tools. CUDA 10 also provides all the necessary components, to create applications for Nvidia's most powerful server platforms, for high performance computing (HPC) and AI workloads, both on-site (DGX-2) and in the cloud (HGX-2).

We recommend reading our post on What is Nvidia Scanner and how it works

TensorRT 5 - Release Candidate: TensorRT 5 offers inference performance up to 40 times faster than CPUs through new optimizations, APIs and support for GPU Turing. Optimizes mixed precision inference dramatically in applications such as recommenders, neural machine translation, speech, and natural language processing.

cuDNN 7.3: Deep learning frameworks using cuDNN 7.3 can take advantage of new features and performance of Turing architectures to deliver faster training performance.

NCCL 2.3: Deep learning frameworks using NCCL 2.3 and later, can take advantage of the new features and performance of the Volta and Turing architecture to offer high-performance and efficient multi-node, multi-GPU deep learning scaling. New features include improved low-latency algorithms for small message sizes, and more precise control of when to use GPU Direct P2P and RDMA.

CUTLASS 1.1: Allows developers to use the Turing Tensor Cores for high performance matrix multiplication in CUDA C ++. New features include support for CUDA 10, and new distortion matrix functions, to access Turing's sub-byte capabilities, to enable deep learning research with ultra-low precision.

VRWorks Graphics 3.0 - VRWorks graphics features are targeted at game and application developers and bring a new level of visual fidelity, performance and responsiveness to virtual reality. This version combined with Turing-based GPUs brings many new technologies including variable rate shading and multi-view rendering.

Nsight Compute 1.0: is a next generation tool that provides interactive CUDA API and kernel profiling. This version of Nsight Compute offers fast data collection of detailed performance metrics, and API debugging through a user interface and command line tool.

Nsight Systems 2018.2 - is a low-performance analytics tool that is designed to provide the insights developers need to optimize their software, such as identifying CPU and GPU bottlenecks. Updates in Nsight Systems 2018.2 include CUDA 10 support, command line interface enhancements to cover new usage scenarios, and various compatibility and usability improvements.

Nsight Graphics 2018.5 - is a standalone developer tool that allows you to debug, create profiles, and export charts created with popular graphical APIs. Version 2018.5 makes GPU Trace publicly available, adds support for Direct3D 12 DXR and Vulkan Ray Tracing extensions, extends the pixel history feature to cover DirectX 12, and completes support for Windows RS3 DirectX 12 SDK.

Nsight VSE 6.0 - is a GPU application development environment that allows you to create, debug, profile, and track a wide range of applications. Updates to Nsight VSE 6.0 include graph debugging with ray tracing and debugging support, and enhanced compute analytics with CUDA 10 support.

Techpowerup font