NVIDIA Published Driver 470.57.02, Open-Sourced RTXMU, and Added Linux Support to RTX SDK

NVIDIA has published the first stable release of the new NVIDIA proprietary driver branch 470.57.02. The driver is available for Linux (ARM, x86_64), FreeBSD (x86_64) and Solaris (x86_64).

Main innovations:

  • Added support for new GPUs: GeForce RTX 3070 Ti, GeForce RTX 3080 Ti, T4G, A100 80GB PCIe, A16, PG506-243, PG506-242, CMP 90HX, CMP 70HX, A100-PG506-207, A100-PG506-217, CMP 50HX.
  • Added initial support for OpenGL and Vulkan hardware acceleration for X11 applications running in Wayland environments using the Xwayland DDX component. According to the tests performed, when using the NVIDIA 470 driver branch, the performance of OpenGL and Vulkan in X applications launched using XWayland is almost the same as running under a regular X server.
  • Implemented the ability to use NVIDIA NGX technology in Wine and the Proton package developed by Valve to run Windows games on Linux. Including Wine and Proton, you can now run games that support DLSS technology, which allows you to use the Tensor cores of NVIDIA video cards to realistically scale images using machine learning methods to increase resolution without losing quality.

    The nvngx.dll library is included to use the NGX functionality in Windows applications launched with Wine. On the Wine side and stable releases of Proton, NGX support has not yet been implemented, but the inclusion of changes to support this functionality has already begun in the Proton Experimental branch.

  • The limit on the number of concurrently running OpenGL contexts has been removed and is now limited only by the amount of available memory.
  • Added support for PRIME technology to offload rendering operations to other GPUs (PRIME Display Offload) in configurations where the source and destination GPUs are handled by the NVIDIA driver, and when the source GPU is handled by the AMDGPU driver.
  • Added support for new Vulkan extensions: VK_EXT_global_priority (VK_QUEUE_GLOBAL_PRIORITY_REALTIME_EXT, allows asynchronous reprojection in SteamVR), VK_EXT_global_priority_query, VK_EXT_provoking_vertex, VK_EXT_extended_dynamic_state2, VK_EXT_color_write_enable, VK_EXT_vertex_input_ dynamic_state, VK_EXT_ycbcr_2plane_444_formats, VK_NV_inherited_viewport_scissor.
  • Using global Vulkan properties other than VK_QUEUE_GLOBAL_PRIORITY_MEDIUM_EXT now requires root access or CAP_SYS_NICE privileges.
  • A new nvidia-peermem.ko kernel module has been added to allow RDMA to be used to directly access third-party devices such as Mellanox InfiniBand HCA (Host Channel Adapters) to NVIDIA GPU memory without copying data to system memory.
  • By default, SLI initialization is enabled when using GPUs with different video memory sizes.
  • nvidia-settings and NV-CONTROL provide cooler control tools by default for boards that support software cooler control.
  • The gsp.bin firmware is included, which is used to move the initialization and control of the GPU to the side of the GPU System Processor (GSP) chip.

At the same time, at the Game Developers Conference, NVIDIA announced the open source code of the RTXMU (RTX Memory Utility) SDK tool under the MIT license, which allows using BLAS (bottom level acceleration structures) buffer compaction and suballocation to significantly reduce video memory consumption. Compaction makes it possible to reduce the overall memory consumption of BLAS by 50%, and suballocation improves buffer storage efficiency by combining several small buffers into 64 KB or 4 MB pages.

NVIDIA Published Driver 470.57.02, Open-Sourced RTXMU, and Added Linux Support to RTX SDK

NVIDIA has also open-sourced the NVRHI (NVIDIA Rendering Hardware Interface) library and the Donut framework under an MIT license. NVRHI is an abstract layer that runs on top of various graphics APIs (Direct3D 11, Direct3D 12, Vulkan 1.2) on Windows and Linux. Donut provides a set of prebuilt components and render stages for prototyping real-time rendering systems.

In addition, NVIDIA has provided support for Linux and ARM architecture in the SDK: DLSS (Deep Learning Super Sampling, realistic image scaling using machine learning methods), RTXDI (RTX Direct Illumination, dynamic lighting), RTXGI (RTX Global Illumination, recreating light reflections ), NRD (NVIDIA Optix AI-Acceleration Denoiser, using machine learning to accelerate realistic image rendering).

Source: opennet.ru

Add a comment