diff --git a/3rdparty/stdgpu/stdgpu.cmake b/3rdparty/stdgpu/stdgpu.cmake
index fb9a64e09fd..a725f228029 100644
--- a/3rdparty/stdgpu/stdgpu.cmake
+++ b/3rdparty/stdgpu/stdgpu.cmake
@@ -7,9 +7,8 @@ include(ExternalProject)
 ExternalProject_Add(
     ext_stdgpu
     PREFIX stdgpu
-    # Jun 20 2024. Later versions need CUDA 11.5 and an API update (stdgpu::pair)
-    URL https://github.com/stotko/stdgpu/archive/1b6a3319f1fbf180166e1bbc1d75f69ab622a0a0.tar.gz
-    URL_HASH SHA256=faa3bf9cbe49ef9cc09e2e07e60d10bbf3b896edb6089c920bebe0f850fd95e4
+    URL https://github.com/stotko/stdgpu/archive/2588168d226bd17229dbf58d821549580791089d.tar.gz
+    URL_HASH SHA256=86e50789bbe21c57f64358c6acbd4481d56c1e45ce9ba1fb5c5c8482c3973215
     DOWNLOAD_DIR "${OPEN3D_THIRD_PARTY_DOWNLOAD_DIR}/stdgpu"
     UPDATE_COMMAND ""
     CMAKE_ARGS
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 7b047cd0b82..971b4440e46 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -430,8 +430,8 @@ if(BUILD_CUDA_MODULE)
     endif()
     enable_language(CUDA)
     set(CMAKE_CUDA_STANDARD 17)
-    if (CMAKE_CUDA_COMPILER_ID STREQUAL "NVIDIA" AND CMAKE_CUDA_COMPILER_VERSION VERSION_LESS "10.1")
-        message(FATAL_ERROR "CUDA 10.0 and older are not supported. Please upgrade to CUDA 10.1 or newer.")
+    if (CMAKE_CUDA_COMPILER_ID STREQUAL "NVIDIA" AND CMAKE_CUDA_COMPILER_VERSION VERSION_LESS "11.5")
+        message(FATAL_ERROR "CUDA 11.4 and older are not supported. Please upgrade to CUDA 11.5 or newer.")
     endif()
 endif()
 
diff --git a/cmake/Open3DShowAndAbortOnWarning.cmake b/cmake/Open3DShowAndAbortOnWarning.cmake
index f1ac292fe38..c95bfa75ac0 100644
--- a/cmake/Open3DShowAndAbortOnWarning.cmake
+++ b/cmake/Open3DShowAndAbortOnWarning.cmake
@@ -33,28 +33,14 @@ function(open3d_show_and_abort_on_warning target)
         string(REPLACE ";" "," DISABLE_NVCC_WARNINGS "${DISABLE_NVCC_WARNINGS}")
 
         set(CUDA_FLAGS "--Werror cross-execution-space-call,deprecated-declarations")
-        if (CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL "10.2")
-            string(APPEND CUDA_FLAGS " --Werror all-warnings")
-        endif()
-        if (CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL "11.0")
-            string(APPEND CUDA_FLAGS " --Werror ext-lambda-captures-this")
-        endif()
+        string(APPEND CUDA_FLAGS " --Werror all-warnings")
+        string(APPEND CUDA_FLAGS " --Werror ext-lambda-captures-this")
         string(APPEND CUDA_FLAGS " --expt-relaxed-constexpr")
-        if (CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL "11.2")
-            string(APPEND CUDA_FLAGS " --diag-suppress ${DISABLE_NVCC_WARNINGS}")
-        else()
-            string(APPEND CUDA_FLAGS " -Xcudafe --diag_suppress=[${DISABLE_NVCC_WARNINGS}]")
-        endif()
+        string(APPEND CUDA_FLAGS " --diag-suppress ${DISABLE_NVCC_WARNINGS}")
 
         # Host compiler flags
         if (MSVC)
             set(CUDA_DISABLE_MSVC_WARNINGS ${DISABLE_MSVC_WARNINGS})
-            if (CMAKE_CUDA_COMPILER_VERSION VERSION_LESS "11.2")
-                list(APPEND CUDA_DISABLE_MSVC_WARNINGS $<$<CONFIG:Debug>:/wd4700>) # uninitialized local variable used (thrust)
-            endif()
-            if (CMAKE_CUDA_COMPILER_VERSION VERSION_LESS "11.1")
-                list(APPEND CUDA_DISABLE_MSVC_WARNINGS /wd4515) # namespace uses itself (thrust)
-            endif()
             string(REPLACE ";" "," CUDA_DISABLE_MSVC_WARNINGS "${CUDA_DISABLE_MSVC_WARNINGS}")
 
             string(APPEND CUDA_FLAGS " -Xcompiler /W4,/WX,${CUDA_DISABLE_MSVC_WARNINGS}")
diff --git a/cpp/open3d/core/hashmap/CUDA/StdGPUHashBackend.h b/cpp/open3d/core/hashmap/CUDA/StdGPUHashBackend.h
index 823e19f9901..d6707a6ce17 100644
--- a/cpp/open3d/core/hashmap/CUDA/StdGPUHashBackend.h
+++ b/cpp/open3d/core/hashmap/CUDA/StdGPUHashBackend.h
@@ -8,6 +8,7 @@
 #pragma once
 
 #include <stdgpu/memory.h>
+#include <stdgpu/utility.h>
 #include <thrust/device_vector.h>
 #include <thrust/transform.h>
 
@@ -90,7 +91,7 @@ class StdGPUAllocator {
 // accessible in raw CUDA kernels.
 template <typename Key>
 using InternalStdGPUHashBackendAllocator =
-        StdGPUAllocator<thrust::pair<const Key, buf_index_t>>;
+        StdGPUAllocator<stdgpu::pair<const Key, buf_index_t>>;
 
 template <typename Key, typename Hash, typename Eq>
 using InternalStdGPUHashBackend =
@@ -252,7 +253,7 @@ void StdGPUHashBackend<Key, Hash, Eq>::Erase(const void* input_keys,
 template <typename Key>
 struct ValueExtractor {
     OPEN3D_HOST_DEVICE buf_index_t
-    operator()(const thrust::pair<Key, buf_index_t>& x) const {
+    operator()(const stdgpu::pair<Key, buf_index_t>& x) const {
         return x.second;
     }
 };
diff --git a/docs/compilation.rst b/docs/compilation.rst
index 270c928b327..fa663c0d72d 100644
--- a/docs/compilation.rst
+++ b/docs/compilation.rst
@@ -25,8 +25,8 @@ System requirements
   * macOS: Install with Homebrew: ``brew install cmake``
   * Windows: Download from: `CMake download page <https://cmake.org/download/>`_
 
-* CUDA 10.1+ (optional): Open3D supports GPU acceleration of an increasing number
-  of operations through CUDA on Linux. We recommend using CUDA 11.0 for the
+* CUDA 11.5+ (optional): Open3D supports GPU acceleration of an increasing number
+  of operations through CUDA on Linux. We recommend using CUDA 12+ for the
   best compatibility with recent GPUs and optional external dependencies such
   as Tensorflow or PyTorch. Please see the `official documentation
   <https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html>`_ to
@@ -250,7 +250,7 @@ pipelines from Open3D-ML in the python package, set ``BUNDLE_OPEN3D_ML=ON`` and
 Open3D-ML from GitHub during the build with
 ``OPEN3D_ML_ROOT=https://github.com/isl-org/Open3D-ML.git``.
 
-.. warning:: Compiling PyTorch ops with CUDA 11 and PyTorch < 1.9 may have
+.. warning:: Compiling PyTorch ops with PyTorch < 1.9 may have
     stability issues. See `Open3D issue #3324 <https://github.com/isl-org/Open3D/issues/3324>`_
     and `PyTorch issue #52663 <https://github.com/pytorch/pytorch/issues/52663>`_
     for more information on this problem. Official PyTorch wheels 1.9 and later