英伟达582.16版本驱动高速下载，适合视频编辑、3D 渲染、专业软件 RTX 专业卡
驱动日期：20251223
支持系统：win10 /win11 /win Server 2025 /win Server 2022
下载链接：英伟达582.16版本驱动高速下载
支持的显卡：见文章最下方
Release notes for the Release 580 family of NVIDIA® Data Center GPU Drivers for Windows.
This edition of Release Notes describes the Release 580 family of NVIDIA® Data Center GPU Drivers for Windows. NVIDIA provides these notes to describe performance improvements, bug fixes and limitations in each documented version of the driver.
1. Version Highlights
This section provides highlights of the NVIDIA Data Center GPU R580 Driver (version 582.16 Windows).
For changes related to the 580 release of the NVIDIA display driver, review the file “NVIDIA_Changelog” available in the .run installer packages.
- Linux driver release date: 01/08/2026
- Windows driver release date: 01/08/2026
1.1. Software Versions
For this release, the software versions are as follows:
- CUDA Toolkit 13: 13.xNote that starting with CUDA 11, individual components of the toolkit are versioned independently. For a full list of the individual versioned components (for example, nvcc, CUDA libraries, and so on), see the CUDA Toolkit Release Notes.
- NVIDIA Data Center GPU Driver: 582.16 (Windows)
- NVFlash: 5.791
For more information on getting started with the NVIDIA Fabric Manager on NVSwitch-based systems (for example, NVIDIA HGX A100), refer to the Fabric Manager User Guide.
1.2. Fixed Issues
- Specific Linux kernel versions and configurations will take a long time to scrub large memory buffers without encountering scheduling points. The fix: explicit scheduling points are inserted into a buffer populating loop.
- Because FSP has logic to conditionally reset depending on the PLM, the source isolating it to remove FSP access will fix this issue. The fix CL therefore does that; the source isolates the RESET PLM so FSP cannot access it.
- Move the allocation of migrate_vma_state_t to use kvmalloc() which will fall back to a virtually contiguous allocation if a physically contiguous allocation from kmalloc() is not possible.Note that on 4K kernels, because the allocation is above UVM_KMALLOC_THRESHOLD so vmalloc() will be used. On 64K kernels kmalloc() will be used.
- Available multicast groups move some code to make sure we check the partition exists first before any other return code.
- Add a check before attempting to free the NULL pointer.
- Immediately cast each operand in get_global_id to size_t before the multiply/adds.
- Implemented the new functionality that reports the WAITING_FOR_CONN_RECOVERY state for connections that are blocking initialization.
- Improved/enhanced logging around connection/disconnected times.
- cuGetDeviceCount will fail and report cudaErrorNoDevice if GPU is disabled. The driver should return this error to hybrid_runtime but during the deinit flow for cudaErrorNoDevice, the driver frees the hybrid_runtime causing the hybrid_runtime return address to be invalidated on stack (stack corruption). Hybrid runtime is loaded by static loader via etlb and should be freed by static loader via etlb freeLibrary call.
- Set GPU health after the link state has been set to down for remove operations.
- Add proper synchronization for ISINK controller code.
- Enhanced GFM API SDK to include more detailed information of GPU, Switch, Partition, compute and switch nodes.
- Support operation of the fabric with a single switch tray failure/removal in the NVL domain. The GPUs will continue to operate with reduced bandwidth.
- SINK XIDs are only logged when power smoothing is enabled.
- Partitioning API calls are failed/reverted only when the call results in the partition becoming unhealthy.
- Do not advertise non-graphics MIG devices as physical GPUs when in MIG mode.
- Remove the old workaround in the NVIDIA driver to promote all spinlocks to raw spinlocks under PREEMPT_RT, which are no longer necessary.
- Introduced Multi-tenancy NVLink Partition Mode and Trusted NVLink Partition Mode.Multi-tenancy NVLink Partition Mode (default): GPU reset is needed when it is moved across partitions.Trusted NVLink Partition Mode: GPUs can move across partitions and start working without reset or service restart.The mode is controlled by a new FM config option MNNVL_NVLINK_PARTITION_MODE. The update to config option requires NMX-C restart to take effect.
- Remove WPR mappings from reserved memory mapping.
1.3. Known Issues
- This version of the GPU driver will fail to initialize on systems with Hopper GPUs subrevision = 3 and VBIOS versions older than 96.00.68.00.xx. Please ensure the system is using a VBIOS version 96.00.68.00.xx or newer before upgrading to this version of the driver.
- When upgrading from ClosedRM to OpenRM, nvidia-smi may fail.WorkaroundRun the following commands:sudo rpm -e nvidia-open-driver-G06-kmp-default –nodeps sudo zypper in nvidia-driver-G06-kmp-default sudo zypper install -y nvidia-open-570
- The default TCC mode in the NVIDIA driver does not support IOMMU-based isolation (necessary for Windows features such as DMA protection, kernel DMA guard, virtualization-based security, etc.). The impacted GPUs are NVIDIA L40, NVIDIA L40S, NVIDIA L20, NVIDIA L4, and NVIDIA RTX PRO 6000 Blackwell Server Edition.
- This GPU Driver release is compatible only with Data Center GPU Manager (DCGM) versions 4.3.x or newer. Earlier versions of DCGM are not compatible.
- On RHEL 10 x86_64 with kernel 6.12.0-55.29.1.el10_0.x86_64 the doca-ofed package does not include the ib_umad.ko module, causing Fabric Manager to fail at startup. This issue is not present on the older kernel 6.12.0-55.9.1.el10_0.x86_64.
- If your environment meets all of the following conditions:
- You are using an RPM based distribution.
- You are not using the online CUDA package repository.
- You manually installed nvidia-fabricmanager-devel-580.65.06-1.
2. Virtualization
To make use of GPU passthrough with virtual machines running Windows and Linux, the hardware platform must support the following features:
- A CPU with hardware-assisted instruction set virtualization: Intel VT-x or AMD-V.
- Platform support for I/O DMA remapping.
- On Intel platforms, the DMA remapper technology is called Intel VT-d.
- On AMD platforms, it is called AMD IOMMU.
Support for these features varies by processor family, product, and system, and should be verified at the manufacturer’s website.
The following hypervisors are supported for virtualization:
| Hypervisor | Notes |
|---|---|
| Citrix XenServer | Version 6.0 and later |
| VMware vSphere (ESX / ESXi) | Version 5.1 and later. |
| Red Hat KVM | Red Hat Enterprise Linux 9 with KVM |
| Microsoft Hyper-V | Windows Server 2019 Hyper-V Generation 2 |
Data Center products now support one display of up to 2560×1600 resolution.
The following GPUs are supported for device passthrough for virtualization:
| GPU Family | Boards Supported |
|---|---|
| NVIDIA Blackwell | NVIDIA HGX B300, NVIDIA RTX 6000D, NVIDIA H20BFX, NVIDIA RTX 6000 PRO, NVIDIA HGX GB200 NVL, NVIDIA HGX B200 |
| NVIDIA Grace Hopper | NVIDIA GH200 |
| NVIDIA Hopper | NVIDIA H100, NVIDIA H800 |
| NVIDIA Ada Lovelace | NVIDIA L40, L4, L2, L20 |
| NVIDIA Ampere GPU Architecture | NVIDIA A800, A100, A40, A30, A16, A10, A10G, A2, AX800 |
| NVIDIA Turing | NVIDIA T4, NVIDIA T4G |
| NVIDIA Volta | NVIDIA V100 |
| NVIDIA Pascal | Quadro: P2000, P4000, P5000, P6000, GP100Tesla: P100, P40, P4 |
| NVIDIA Maxwell | Quadro: K2200, M2000, M4000, M5000, M6000, M6000 24GBTesla: M60, M40, M6, M4 |
3. Hardware and Software Support
Support for these features varies by processor family, product, and system, and should be verified at the manufacturer’s website.
Coherent Driver-Based Memory Management (CDMM)
The R580 Driver introduces Coherent Driver-Based Memory Management (CDMM) for GB200 platforms. With CDMM, the driver manages GPU memory instead of the OS. CDMM avoids OS onlining of the GPU memory and the exposing of the GPU memory as a NUMA node to the OS. It is recommended that Kubernetes clusters enable CDMM to resolve potential memory over-reporting.
To set up the driver in CDMM mode, run the following commands and then reload the driver:
echo options nvidia NVreg_CoherentGPUMemoryMode=driver > /etc/modprobe.d/nvidia-openrm.conf
Note:
- If there is already a configuration file for the nvidia driver, please merge the options into a single options line.
- To remove the configuration, undo its addition to the configuration file
- To use GDRCopy with CDMM, please use version 2.5.1 or later of GDRCopy.
- GPU Direct Storage is not supported with CDMM.
Supported Operating Systems for NVIDIA Data Center GPUs
The Release 580 driver is supported on the following operating systems:
- Windows x86_64 operating systems:
- Microsoft Windows® Server 2025 24H2
- Microsoft Windows® Server 2022 21H2
- Microsoft Windows® 11 25H2
- Microsoft Windows® 11 24H2 – SV4
- Microsoft Windows® 11 23H2
- Microsoft Windows® 11 22H2 – SV2
- Microsoft Windows® 10 22H2
- The HGX platform also includes support for the Windows OS 64-bit distributions:
- Microsoft Windows® Server 2025
- Microsoft Windows® Server 2022
- Windows is supported only in shared NVSwitch virtualization configurations.
Supported Operating Systems and CPU Configurations for NVIDIA HGX B300
Windows 64-bit distributions:
- Microsoft Windows® Server 2025
- Microsoft Windows® Server 2022
Supported Operating Systems and CPU Configurations for NVIDIA RTX 6000D
Windows 64-bit distributions:
- Microsoft Windows® Server 2025
- Microsoft Windows® 11 24H2 – SV4
- Microsoft Windows® 11 23H2 – SV3
Supported Operating Systems and CPU Configurations for NVIDIA RTX Pro 6000 Blackwell Server Edition
The Release 580 driver is validated with NVIDIA RTX Pro 6000 Blackwell Server Edition on the following operating systems and CPU configurations:
- Windows 64-bit distributions:
- Microsoft Windows® Server 2025
- Microsoft Windows® 11 24H2 – SV4
- Microsoft Windows® 11 23H2 – SV3
- Microsoft Windows® 10 22H2
- Microsoft Windows® 10 21H2
API Support
This release supports the following APIs:
- NVIDIA® CUDA® 13.x for NVIDIA® Maxwell™, Pascal™, Volta™, Turing™, Hopper™, NVIDIA Ampere architecture, NVIDIA Ada Lovelace architecture, and NVIDIA Blackwell architecture GPUs
- OpenGL® 4.6
- Vulkan® 1.3
- DirectX 11
- DirectX 12 (Windows 10)
- Open Computing Language (OpenCL™ software) 3.0
Supported NVIDIA Data Center GPUs
The NVIDIA Data Center GPU driver package is designed for systems that have one or more Data Center GPU products installed. This release of the driver supports CUDA C/C++ applications and libraries that rely on the CUDA C Runtime and/or CUDA Driver API.
Attention: Release 470 was the last driver branch to support Data Center GPUs based on the NVIDIA Kepler architecture. This includes discontinued support for the following compute capabilities:
- sm_30 (NVIDIA Kepler)
- sm_32 (NVIDIA Kepler)
- sm_35 (NVIDIA Kepler)
- sm_37 (NVIDIA Kepler)
| NVIDIA Server Platforms | |
|---|---|
| Product | Architecture |
| NVIDIA HGX GB200 NVL | GB200 and NVLink |
| NVIDIA HGX B200 8-GPU | B200 and NVSwitch |
| NVIDIA HGX H20-3e 8-GPU | H20 and NVSwitch |
| NVIDIA HGX H20 8-GPU | H20 and NVSwitch |
| NVIDIA HGX H200 8-GPU | H200 and NVSwitch |
| NVIDIA HGX H100 8-GPU | H100 and NVSwitch |
| NVIDIA HGX H800 8-GPU | H800 and NVSwitch |
| NVIDIA HGX H100 4-GPU | H100 and NVLink |
| NVIDIA HGX A800 8-GPU | A800 and NVSwitch |
| NVIDIA HGX A100 8-GPU | A100 and NVSwitch |
| NVIDIA HGX A100 4-GPU | A100 and NVLink |
| NVIDIA HGX-2 | V100 and NVSwitch |
| Data Center H-Series Products | |
|---|---|
| Product | GPU Architecture |
| NVIDIA H100 PCIe | NVIDIA Hopper |
| NVIDIA H100 NVL | NVIDIA Hopper |
| NVIDIA H200 NVL | NVIDIA Hopper |
| NVIDIA H800 PCIe | NVIDIA Hopper |
| NVIDIA H800 NVL | NVIDIA Hopper |
| Data Center L-Series Products | |
|---|---|
| Product | GPU Architecture |
| NVIDIA L2 | NVIDIA Ada Lovelace |
| NVIDIA L20 | NVIDIA Ada Lovelace |
| NVIDIA L40 | NVIDIA Ada Lovelace |
| NVIDIA L40S | NVIDIA Ada Lovelace |
| NVIDIA L4 | NVIDIA Ada Lovelace |
| RTX-Series / T-Series Products | |
|---|---|
| Product | GPU Architecture |
| NVIDIA RTX PRO 6000 Blackwell Server Edition | NVIDIA Blackwell |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | NVIDIA Blackwell |
| NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | NVIDIA Blackwell |
| NVIDIA RTX PRO 5000 Blackwell | NVIDIA Blackwell |
| NVIDIA RTX PRO 4500 Blackwell | NVIDIA Blackwell |
| NVIDIA RTX PRO 4000 Blackwell | NVIDIA Blackwell |
| NVIDIA RTX PRO 2000 Blackwell | NVIDIA Blackwell |
| NVIDIA RTX PRO 4000 SFF Blackwell | NVIDIA Blackwell |
| NVIDIA RTX 6000 Ada Generation | NVIDIA Ada Lovelace |
| NVIDIA RTX 5880 Ada Generation | NVIDIA Ada Lovelace |
| NVIDIA RTX 5000 Ada Generation | NVIDIA Ada Lovelace |
| NVIDIA RTX 4500 Ada Generation | NVIDIA Ada Lovelace |
| NVIDIA RTX 4000 Ada Generation | NVIDIA Ada Lovelace |
| NVIDIA RTX 4000 SFF Ada Generation | NVIDIA Ada Lovelace |
| NVIDIA RTX 2000 Ada Generation | NVIDIA Ada Lovelace |
| NVIDIA RTX 2000E Ada Generation | NVIDIA Ada Lovelace |
| NVIDIA RTX A6000 | NVIDIA Ampere architecture |
| NVIDIA RTX A5500 | NVIDIA Ampere architecture |
| NVIDIA RTX A5000 | NVIDIA Ampere architecture |
| NVIDIA RTX A4500 | NVIDIA Ampere architecture |
| NVIDIA RTX A4000H | NVIDIA Ampere architecture |
| NVIDIA RTX A4000 | NVIDIA Ampere architecture |
| NVIDIA RTX A2000 12GB | NVIDIA Ampere architecture |
| NVIDIA RTX A2000 | NVIDIA Ampere architecture |
| NVIDIA RTX A1000 | NVIDIA Ampere architecture |
| NVIDIA RTX A400 | NVIDIA Ampere architecture |
| NVIDIA RTX A800 40GB Active | NVIDIA Ampere architecture |
| Quadro RTX 8000 | NVIDIA Turing |
| Quadro RTX 6000 | NVIDIA Turing |
| Quadro RTX A6000 | NVIDIA Turing |
| Quadro RTX 5000 | NVIDIA Turing |
| Quadro RTX A5000 | NVIDIA Turing |
| Quadro RTX 4000 | NVIDIA Turing |
| Quadro RTX A4000 | NVIDIA Turing |
| NVIDIA T1000 8GB | NVIDIA Turing |
| NVIDIA T600 | NVIDIA Turing |
| NVIDIA T400 4GB | NVIDIA Turing |
| NVIDIA T400 | NVIDIA Turing |
| NVIDIA T400E | NVIDIA Turing |
| Data Center A-Series Products | |
|---|---|
| Product | GPU Architecture |
| NVIDIA A2 | NVIDIA Ampere architecture |
| NVIDIA A800, AX800 | NVIDIA Ampere architecture |
| NVIDIA A100X | NVIDIA Ampere architecture |
| NVIDIA A100NVIDIA A100 80 GB PCIe | NVIDIA Ampere architecture |
| NVIDIA A40 | NVIDIA Ampere architecture |
| NVIDIA A30, A30X | NVIDIA Ampere architecture |
| NVIDIA A16 | NVIDIA Ampere architecture |
| NVIDIA A10, A10M, A10G | NVIDIA Ampere architecture |
| Data Center T-Series Products | |
|---|---|
| Product | GPU Architecture |
| NVIDIA T4, T4G | NVIDIA Turing |
| Data Center V-Series Products | |
|---|---|
| Product | GPU Architecture |
| NVIDIA V100 | Volta |
| Data Center P-Series Products | |
|---|---|
| Product | GPU Architecture |
| NVIDIA Tesla P100 | NVIDIA Pascal |
| NVIDIA Tesla P40 | NVIDIA Pascal |
| NVIDIA Tesla P4 | NVIDIA Pascal |
| Data Center M-Class Products | |
|---|---|
| Product | GPU Architecture |
| NVIDIA Tesla M60 | Maxwell |
| NVIDIA Tesla M40 24 GB | Maxwell |
| NVIDIA Tesla M40 | Maxwell |
| NVIDIA Tesla M6 | Maxwell |
| NVIDIA Tesla M4 | Maxwell |
💡 遇到驱动问题？推荐阅读： 显卡驱动装不上怎么办？点击查看完整解决方案 →
