<!DOCTYPE html
PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml" lang="en-us" xml:lang="en-us">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8"></meta>
<meta http-equiv="X-UA-Compatible" content="IE=edge"></meta>
<meta name="copyright" content="(C) Copyright 2005"></meta>
<meta name="DC.rights.owner" content="(C) Copyright 2005"></meta>
<meta name="DC.Type" content="cuda_reference"></meta>
<meta name="DC.Title" content="Difference between the driver and runtime APIs"></meta>
<meta name="DC.Format" content="XHTML"></meta>
<meta name="DC.Identifier" content="driver-vs-runtime-api"></meta>
<link rel="stylesheet" type="text/css" href="../common/formatting/commonltr.css"></link>
<link rel="stylesheet" type="text/css" href="../common/formatting/site.css"></link>
<title>CUDA Driver API :: CUDA Toolkit Documentation</title>
<!--[if lt IE 9]>
<script src="../common/formatting/html5shiv-printshiv.min.js"></script>
<![endif]-->
<script type="text/javascript" charset="utf-8" src="//assets.adobedtm.com/b92787824f2e0e9b68dc2e993f9bd995339fe417/satelliteLib-7ba51e58dc61bcb0e9311aadd02a0108ab24cc6c.js"></script>
<script type="text/javascript" charset="utf-8" src="../common/formatting/jquery.min.js"></script>
<script type="text/javascript" charset="utf-8" src="../common/formatting/jquery.ba-hashchange.min.js"></script>
<script type="text/javascript" charset="utf-8" src="../common/formatting/jquery.scrollintoview.min.js"></script>
<script type="text/javascript" src="../search/htmlFileList.js"></script>
<script type="text/javascript" src="../search/htmlFileInfoList.js"></script>
<script type="text/javascript" src="../search/nwSearchFnt.min.js"></script>
<script type="text/javascript" src="../search/stemmers/en_stemmer.min.js"></script>
<script type="text/javascript" src="../search/index-1.js"></script>
<script type="text/javascript" src="../search/index-2.js"></script>
<script type="text/javascript" src="../search/index-3.js"></script>
<link rel="canonical" href="http://docs.nvidia.com/cuda/cuda-driver-api/index.html"></link>
<link rel="stylesheet" type="text/css" href="../common/formatting/qwcode.highlight.css"></link>
</head>
<body>
<header id="header"><span id="company">NVIDIA</span><span id="site-title">CUDA Toolkit Documentation</span><form id="search" method="get" action="search">
<input type="text" name="search-text"></input><fieldset id="search-location">
<legend>Search In:</legend>
<label><input type="radio" name="search-type" value="site"></input>Entire Site</label>
<label><input type="radio" name="search-type" value="document"></input>Just This Document</label></fieldset>
<button type="reset">clear search</button>
<button id="submit" type="submit">search</button></form>
</header>
<div id="site-content">
<nav id="site-nav">
<div class="category closed"><a href="../index.html" title="The root of the site.">CUDA Toolkit
v10.1.168</a></div>
<div class="category"><a href="index.html" title="CUDA Driver API">CUDA Driver API</a></div>
<ul>
<li>
<div class="section-link"><a href="driver-vs-runtime-api.html#driver-vs-runtime-api">1. Difference between the driver and runtime APIs </a></div>
</li>
<li>
<div class="section-link"><a href="api-sync-behavior.html#api-sync-behavior">2. API synchronization behavior </a></div>
</li>
<li>
<div class="section-link"><a href="stream-sync-behavior.html#stream-sync-behavior">3. Stream synchronization behavior</a></div>
</li>
<li>
<div class="section-link"><a href="graphs-thread-safety.html#graphs-thread-safety">4. Graph object thread safety</a></div>
</li>
<li>
<div class="section-link"><a href="modules.html#modules">5. Modules</a></div>
<ul>
<li>
<div class="section-link"><a href="group__CUDA__TYPES.html#group__CUDA__TYPES">5.1. Data types used by CUDA driver</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__ERROR.html#group__CUDA__ERROR">5.2. Error Handling</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__INITIALIZE.html#group__CUDA__INITIALIZE">5.3. Initialization</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__VERSION.html#group__CUDA__VERSION">5.4. Version Management</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__DEVICE.html#group__CUDA__DEVICE">5.5. Device Management</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__DEVICE__DEPRECATED.html#group__CUDA__DEVICE__DEPRECATED">5.6. Device Management [DEPRECATED]</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__PRIMARY__CTX.html#group__CUDA__PRIMARY__CTX">5.7. Primary Context Management</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__CTX.html#group__CUDA__CTX">5.8. Context Management</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__CTX__DEPRECATED.html#group__CUDA__CTX__DEPRECATED">5.9. Context Management [DEPRECATED]</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__MODULE.html#group__CUDA__MODULE">5.10. Module Management</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__MEM.html#group__CUDA__MEM">5.11. Memory Management</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__UNIFIED.html#group__CUDA__UNIFIED">5.12. Unified Addressing</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__STREAM.html#group__CUDA__STREAM">5.13. Stream Management</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__EVENT.html#group__CUDA__EVENT">5.14. Event Management</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__EXTRES__INTEROP.html#group__CUDA__EXTRES__INTEROP">5.15. External Resource Interoperability</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__MEMOP.html#group__CUDA__MEMOP">5.16. Stream memory operations</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__EXEC.html#group__CUDA__EXEC">5.17. Execution Control</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__EXEC__DEPRECATED.html#group__CUDA__EXEC__DEPRECATED">5.18. Execution Control [DEPRECATED]</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__GRAPH.html#group__CUDA__GRAPH">5.19. Graph Management</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__OCCUPANCY.html#group__CUDA__OCCUPANCY">5.20. Occupancy</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__TEXREF__DEPRECATED.html#group__CUDA__TEXREF__DEPRECATED">5.21. Texture Reference Management [DEPRECATED]</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__SURFREF__DEPRECATED.html#group__CUDA__SURFREF__DEPRECATED">5.22. Surface Reference Management [DEPRECATED]</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__TEXOBJECT.html#group__CUDA__TEXOBJECT">5.23. Texture Object Management</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__SURFOBJECT.html#group__CUDA__SURFOBJECT">5.24. Surface Object Management</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__PEER__ACCESS.html#group__CUDA__PEER__ACCESS">5.25. Peer Context Memory Access</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__GRAPHICS.html#group__CUDA__GRAPHICS">5.26. Graphics Interoperability</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__PROFILER.html#group__CUDA__PROFILER">5.27. Profiler Control</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__GL.html#group__CUDA__GL">5.28. OpenGL Interoperability</a></div>
<ul>
<li>
<div class="section-link"><a href="group__CUDA__GL__DEPRECATED.html#group__CUDA__GL__DEPRECATED">5.28.1. OpenGL Interoperability [DEPRECATED]</a></div>
</li>
</ul>
</li>
<li>
<div class="section-link"><a href="group__CUDA__D3D9.html#group__CUDA__D3D9">5.29. Direct3D 9 Interoperability</a></div>
<ul>
<li>
<div class="section-link"><a href="group__CUDA__D3D9__DEPRECATED.html#group__CUDA__D3D9__DEPRECATED">5.29.1. Direct3D 9 Interoperability [DEPRECATED]</a></div>
</li>
</ul>
</li>
<li>
<div class="section-link"><a href="group__CUDA__D3D10.html#group__CUDA__D3D10">5.30. Direct3D 10 Interoperability</a></div>
<ul>
<li>
<div class="section-link"><a href="group__CUDA__D3D10__DEPRECATED.html#group__CUDA__D3D10__DEPRECATED">5.30.1. Direct3D 10 Interoperability [DEPRECATED]</a></div>
</li>
</ul>
</li>
<li>
<div class="section-link"><a href="group__CUDA__D3D11.html#group__CUDA__D3D11">5.31. Direct3D 11 Interoperability</a></div>
<ul>
<li>
<div class="section-link"><a href="group__CUDA__D3D11__DEPRECATED.html#group__CUDA__D3D11__DEPRECATED">5.31.1. Direct3D 11 Interoperability [DEPRECATED]</a></div>
</li>
</ul>
</li>
<li>
<div class="section-link"><a href="group__CUDA__VDPAU.html#group__CUDA__VDPAU">5.32. VDPAU Interoperability</a></div>
</li>
<li>
<div class="section-link"><a href="group__CUDA__EGL.html#group__CUDA__EGL">5.33. EGL Interoperability</a></div>
</li>
</ul>
</li>
<li>
<div class="section-link"><a href="annotated.html#annotated">6. Data Structures</a></div>
<ul>
<li>
<div class="section-link"><a href="structCUDA__ARRAY3D__DESCRIPTOR.html#structCUDA__ARRAY3D__DESCRIPTOR">6.1. CUDA_ARRAY3D_DESCRIPTOR</a></div>
</li>
<li>
<div class="section-link"><a href="structCUDA__ARRAY__DESCRIPTOR.html#structCUDA__ARRAY__DESCRIPTOR">6.2. CUDA_ARRAY_DESCRIPTOR</a></div>
</li>
<li>
<div class="section-link"><a href="structCUDA__EXTERNAL__MEMORY__BUFFER__DESC.html#structCUDA__EXTERNAL__MEMORY__BUFFER__DESC">6.3. CUDA_EXTERNAL_MEMORY_BUFFER_DESC</a></div>
</li>
<li>
<div class="section-link"><a href="structCUDA__EXTERNAL__MEMORY__HANDLE__DESC.html#structCUDA__EXTERNAL__MEMORY__HANDLE__DESC">6.4. CUDA_EXTERNAL_MEMORY_HANDLE_DESC</a></div>
</li>
<li>
<div class="section-link"><a href="structCUDA__EXTERNAL__MEMORY__MIPMAPPED__ARRAY__DESC.html#structCUDA__EXTERNAL__MEMORY__MIPMAPPED__ARRAY__DESC">6.5. CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC</a></div>
</li>
<li>
<div class="section-link"><a href="structCUDA__EXTERNAL__SEMAPHORE__HANDLE__DESC.html#structCUDA__EXTERNAL__SEMAPHORE__HANDLE__DESC">6.6. CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC</a></div>
</li>
<li>
<div class="section-link"><a href="structCUDA__EXTERNAL__SEMAPHORE__SIGNAL__PARAMS.html#structCUDA__EXTERNAL__SEMAPHORE__SIGNAL__PARAMS">6.7. CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS</a></div>
</li>
<li>
<div class="section-link"><a href="structCUDA__EXTERNAL__SEMAPHORE__WAIT__PARAMS.html#structCUDA__EXTERNAL__SEMAPHORE__WAIT__PARAMS">6.8. CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS</a></div>
</li>
<li>
<div class="section-link"><a href="structCUDA__HOST__NODE__PARAMS.html#structCUDA__HOST__NODE__PARAMS">6.9. CUDA_HOST_NODE_PARAMS</a></div>
</li>
<li>
<div class="section-link"><a href="structCUDA__KERNEL__NODE__PARAMS.html#structCUDA__KERNEL__NODE__PARAMS">6.10. CUDA_KERNEL_NODE_PARAMS</a></div>
</li>
<li>
<div class="section-link"><a href="structCUDA__LAUNCH__PARAMS.html#structCUDA__LAUNCH__PARAMS">6.11. CUDA_LAUNCH_PARAMS</a></div>
</li>
<li>
<div class="section-link"><a href="structCUDA__MEMCPY2D.html#structCUDA__MEMCPY2D">6.12. CUDA_MEMCPY2D</a></div>
</li>
<li>
<div class="section-link"><a href="structCUDA__MEMCPY3D.html#structCUDA__MEMCPY3D">6.13. CUDA_MEMCPY3D</a></div>
</li>
<li>
<div class="section-link"><a href="structCUDA__MEMCPY3D__PEER.html#structCUDA__MEMCPY3D__PEER">6.14. CUDA_MEMCPY3D_PEER</a></div>
</li>
<li>
<div class="section-link"><a href="structCUDA__MEMSET__NODE__PARAMS.html#structCUDA__MEMSET__NODE__PARAMS">6.15. CUDA_MEMSET_NODE_PARAMS</a></div>
</li>
<li>
<div class="section-link"><a href="structCUDA__POINTER__ATTRIBUTE__P2P__TOKENS.html#structCUDA__POINTER__ATTRIBUTE__P2P__TOKENS">6.16. CUDA_POINTER_ATTRIBUTE_P2P_TOKENS</a></div>
</li>
<li>
<div class="section-link"><a href="structCUDA__RESOURCE__DESC.html#structCUDA__RESOURCE__DESC">6.17. CUDA_RESOURCE_DESC</a></div>
</li>
<li>
<div class="section-link"><a href="structCUDA__RESOURCE__VIEW__DESC.html#structCUDA__RESOURCE__VIEW__DESC">6.18. CUDA_RESOURCE_VIEW_DESC</a></div>
</li>
<li>
<div class="section-link"><a href="structCUDA__TEXTURE__DESC.html#structCUDA__TEXTURE__DESC">6.19. CUDA_TEXTURE_DESC</a></div>
</li>
<li>
<div class="section-link"><a href="structCUdevprop.html#structCUdevprop">6.20. CUdevprop</a></div>
</li>
<li>
<div class="section-link"><a href="structCUeglFrame.html#structCUeglFrame">6.21. CUeglFrame</a></div>
</li>
<li>
<div class="section-link"><a href="structCUipcEventHandle.html#structCUipcEventHandle">6.22. CUipcEventHandle</a></div>
</li>
<li>
<div class="section-link"><a href="structCUipcMemHandle.html#structCUipcMemHandle">6.23. CUipcMemHandle</a></div>
</li>
<li>
<div class="section-link"><a href="unionCUstreamBatchMemOpParams.html#unionCUstreamBatchMemOpParams">6.24. CUstreamBatchMemOpParams</a></div>
</li>
</ul>
</li>
<li>
<div class="section-link"><a href="functions.html#functions">7. Data Fields</a></div>
</li>
<li>
<div class="section-link"><a href="deprecated.html#deprecated">8. Deprecated List</a></div>
</li>
<li>
<div class="section-link"><a href="notices-header.html#notices-header">Notices</a></div>
<ul></ul>
</li>
</ul>
</nav>
<div id="resize-nav"></div>
<nav id="search-results">
<h2>Search Results</h2>
<ol></ol>
</nav>
<div id="contents-container">
<div id="breadcrumbs-container">
<div id="breadcrumbs"><a href="index.html" shape="rect">< Previous</a> | <a href="api-sync-behavior.html" shape="rect">Next ></a></div>
<div id="release-info">CUDA Driver API
(<a href="../../pdf/CUDA_Driver_API.pdf">PDF</a>)
-
v10.1.168
(<a href="https://developer.nvidia.com/cuda-toolkit-archive">older</a>)
-
Last updated April 24, 2019
-
<a href="mailto:CUDAIssues@nvidia.com?subject=CUDA Toolkit Documentation Feedback: CUDA Driver API">Send Feedback</a></div>
</div>
<article id="contents">
<div class="topic nested1" id="driver-vs-runtime-api"><a name="driver-vs-runtime-api" shape="rect">
<!-- --></a><h2 class="topictitle2">1. Difference between the driver and runtime APIs </h2>
<div class="body refbody">
<div class="section">
<p class="p">The driver and runtime APIs are very similar and can for the most part be used interchangeably. However, there are some key
differences worth noting between the two.
</p>
</div>
<div class="section" id="driver-vs-runtime-api__complexity-vs-control"><a name="driver-vs-runtime-api__complexity-vs-control" shape="rect">
<!-- --></a><h2 class="title sectiontitle">Complexity vs. control</h2>
<p class="p">The runtime API eases device code management by providing implicit initialization, context management, and module management.
This leads to simpler code, but it also lacks the level of control that the driver API has.
</p>
<p class="p">In comparison, the driver API offers more fine-grained control, especially over contexts and module loading. Kernel launches
are much more complex to implement, as the execution configuration and kernel parameters must be specified with explicit function
calls. However, unlike the runtime, where all the kernels are automatically loaded during initialization and stay loaded for
as long as the program runs, with the driver API it is possible to only keep the modules that are currently needed loaded,
or even dynamically reload modules. The driver API is also language-independent as it only deals with cubin objects.
</p>
</div>
<div class="section" id="driver-vs-runtime-api__context-management"><a name="driver-vs-runtime-api__context-management" shape="rect">
<!-- --></a><h2 class="title sectiontitle">Context management</h2>
<p class="p">Context management can be done through the driver API, but is not exposed in the runtime API. Instead, the runtime API decides
itself which context to use for a thread: if a context has been made current to the calling thread through the driver API,
the runtime will use that, but if there is no such context, it uses a "primary context." Primary contexts are created as needed,
one per device per process, are reference-counted, and are then destroyed when there are no more references to them. Within
one process, all users of the runtime API will share the primary context, unless a context has been made current to each thread.
The context that the runtime uses, i.e, either the current context or primary context, can be synchronized with <samp class="ph codeph">cudaDeviceSynchronize()</samp>, and destroyed with <samp class="ph codeph">cudaDeviceReset()</samp>.
</p>
<p class="p">Using the runtime API with primary contexts has its tradeoffs, however. It can cause trouble for users writing plug-ins for
larger software packages, for example, because if all plug-ins run in the same process, they will all share a context but
will likely have no way to communicate with each other. So, if one of them calls <samp class="ph codeph">cudaDeviceReset()</samp> after finishing all its CUDA work, the other plug-ins will fail because the context they were using was destroyed without
their knowledge. To avoid this issue, CUDA clients can use the driver API to create and set the current context, and then
use the runtime API to work with it. However, contexts may consume significant resources, such as device memory, extra host
threads, and performance costs of context switching on the device. This runtime-driver context sharing is important when using
the driver API in conjunction with libraries built on the runtime API, such as cuBLAS or cuFFT.
</p>
</div>
</div>
</div>
<hr id="contents-end"></hr>
</article>
</div>
</div>
<script language="JavaScript" type="text/javascript" charset="utf-8" src="../common/formatting/common.min.js"></script>
<script language="JavaScript" type="text/javascript" charset="utf-8" src="../common/scripts/google-analytics/google-analytics-write.js"></script>
<script language="JavaScript" type="text/javascript" charset="utf-8" src="../common/scripts/google-analytics/google-analytics-tracker.js"></script>
<script type="text/javascript">var switchTo5x=true;</script><script type="text/javascript" src="http://w.sharethis.com/button/buttons.js"></script><script type="text/javascript">stLight.options({publisher: "998dc202-a267-4d8e-bce9-14debadb8d92", doNotHash: false, doNotCopy: false, hashAddressBar: false});</script><script type="text/javascript">_satellite.pageBottom();</script></body>
</html>
distrib
>
Mageia
>
7
>
x86_64
>
media
>
nonfree-updates
>
by-pkgid
>
b86a85131cc739c1c53d0b55840a4328
>
files
>
1368
