diff --git a/Project.toml b/Project.toml
index 3f73c968f13..effb24d3251 100644
--- a/Project.toml
+++ b/Project.toml
@@ -57,8 +57,10 @@ AMDGPU = "21141c5a-9bdb-4563-92ae-f87d6854732e"
 CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
 Convex = "f65535da-76fb-5f13-bab9-19810c17039a"
 ECOS = "e2685f51-7e38-5353-a97d-a921fd2c8199"
+IntelITT = "c9b2f978-7543-4802-ae44-75068f23ee64"
 Makie = "ee78f7c6-11fb-53f2-987a-cfe4a2b5a57a"
 NLsolve = "2774e3e8-f4cf-5e23-947b-6d7e65073b56"
+NVTX = "5da4648a-3479-48b8-97b9-01cb529c0a1f"
 Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"
 SparseConnectivityTracer = "9f842d2f-2579-4b1d-911e-f412cf18a3f5"
 
@@ -66,8 +68,10 @@ SparseConnectivityTracer = "9f842d2f-2579-4b1d-911e-f412cf18a3f5"
 TrixiAMDGPUExt = "AMDGPU"
 TrixiCUDAExt = "CUDA"
 TrixiConvexECOSExt = ["Convex", "ECOS"]
+TrixiIntelITTExt = "IntelITT"
 TrixiMakieExt = "Makie"
 TrixiNLsolveExt = "NLsolve"
+TrixiNVTXExt = ["NVTX", "CUDA"]
 TrixiPlotsExt = "Plots"
 TrixiSparseConnectivityTracerExt = "SparseConnectivityTracer"
 
@@ -89,6 +93,7 @@ EllipsisNotation = "1.0"
 FillArrays = "1.9"
 ForwardDiff = "0.10.38, 1"
 HDF5 = "0.17"
+IntelITT = "0.2"
 KernelAbstractions = "0.9.38"
 LinearAlgebra = "1"
 LinearMaps = "2.7, 3.0"
@@ -97,6 +102,7 @@ MPI = "0.20.23"
 Makie = "0.22, 0.23, 0.24"
 MuladdMacro = "0.2.4"
 NLsolve = "4.5.1"
+NVTX = "1.0.0"
 Octavian = "0.3.28"
 OffsetArrays = "1.13"
 P4est = "0.4.12"
diff --git a/benchmark/CUDA/Project.toml b/benchmark/CUDA/Project.toml
index 2e9f130fe6c..1bdbed2467b 100644
--- a/benchmark/CUDA/Project.toml
+++ b/benchmark/CUDA/Project.toml
@@ -1,6 +1,7 @@
 [deps]
 CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
 JSON = "682c06a0-de6a-54ab-a142-c8b1cf79cde6"
+NVTX = "5da4648a-3479-48b8-97b9-01cb529c0a1f"
 OrdinaryDiffEqLowStorageRK = "b0944070-b475-4768-8dec-fb6eb410534d"
 TimerOutputs = "a759f4b9-e2f1-59dc-863e-4aeb61b1ea8f"
 Trixi = "a7f1ee26-1774-49b1-8366-f1abc58fbfcb"
@@ -11,6 +12,7 @@ Trixi = {path = "../.."}
 [compat]
 CUDA = "5.8.2"
 JSON = "1.4.0"
+NVTX = "1"
 OrdinaryDiffEqLowStorageRK = "1.12.0"
 TimerOutputs = "0.5.25"
 Trixi = "0.16"
diff --git a/benchmark/CUDA/run.jl b/benchmark/CUDA/run.jl
index b9d02246c9b..89907cb9e9b 100644
--- a/benchmark/CUDA/run.jl
+++ b/benchmark/CUDA/run.jl
@@ -1,5 +1,6 @@
 using Trixi
 using CUDA
+using NVTX # Load to get tracing support for Trixi
 using TimerOutputs
 using JSON
 
diff --git a/docs/src/performance.md b/docs/src/performance.md
index 3d4ff8fdfce..505663735ae 100644
--- a/docs/src/performance.md
+++ b/docs/src/performance.md
@@ -288,3 +288,44 @@ requires. It can thus be seen as a proxy for "energy used" and, as an extension,
     timing result, you need to set the analysis interval such that the
     `AnalysisCallback` is invoked at least once during the course of the simulation and
     discard the first PID value.
+
+## Tracing support for profilers
+
+Trixi supports tracing profiler integration through [ittapi](https://github.com/intel/ittapi) for Intel VTune and [NVTX](https://github.com/NVIDIA/NVTX) for [NVIDIA Nsight Systems](https://developer.nvidia.com/nsight-systems).
+
+!!! note "Extensions"
+    Tracing support is implemented through extensions and requires trigger packages to be loaded.
+
+Tracing support is only available for regions that are instrumented with `@trixi_timeit_ext`.
+
+### Using Intel VTune
+
+We can use Intel VTune to profile CPU code. For more information see the [Julia documentation](https://docs.julialang.org/en/v1/manual/profile/#External-Profiling) and the [IntelITT.jl](https://github.com/JuliaPerf/IntelITT.jl) package.
+
+!!! note "Trigger package"
+    ```julia
+    using IntelITT
+    ```
+
+To get the most out of Intel VTune we recommend passing the environment flag `ENABLE_JITPROFILING=1` to Julia, which will allow you to symbolize JIT compiled call frames.
+
+!!! note "Usage of `juliaup`"
+    Sometime `juliaup` can make it harder for a profiler to attach to the right process. You can use `Base.julia_cmd()` in the REPL to obtain the path to the actual Julia binary you will be running.
+
+
+### NVIDIA Nsight Systems
+
+We can use NVIDIA Nsight Systems to trace GPU.
+
+We recommend reading the CUDA.jl documentation on using [Nsight Systems](https://cuda.juliagpu.org/stable/development/profiling/#NVIDIA-Nsight-Systems)
+
+!!! note "Trigger package"
+    ```julia
+    using CUDA
+    using NVTX
+    ```
+
+You can also just use `CUDA.@profile` (see [Integrated Profiler](https://cuda.juliagpu.org/stable/development/profiling/#Integrated-profiler)) to obtain profiler results that include the NVTX ranges.
+
+#### Known limitation
+Nsight Systems can also be used for CPU and in particular MPI codes. The Trixi extension will only be enabled when GPU backend is being used. 
diff --git a/ext/TrixiIntelITTExt.jl b/ext/TrixiIntelITTExt.jl
new file mode 100644
index 00000000000..3ce1158ed6e
--- /dev/null
+++ b/ext/TrixiIntelITTExt.jl
@@ -0,0 +1,28 @@
+module TrixiIntelITTExt
+
+using Trixi: CPU
+import Trixi: trixi_range_active, trixi_range_start, trixi_range_end
+
+import IntelITT
+
+const domain = Ref{IntelITT.Domain}()
+function __init__()
+    domain[] = IntelITT.Domain("Trixi")
+end
+
+function trixi_range_active(::Union{Nothing, CPU})
+    return IntelITT.isactive()
+end
+
+function trixi_range_start(::Union{Nothing, CPU}, label)
+    task = IntelITT.Task(domain[], label)
+    IntelITT.start(task)
+    return task
+end
+
+function trixi_range_end(::Union{Nothing, CPU}, id)
+    IntelITT.stop(id)
+    return nothing
+end
+
+end # module
diff --git a/ext/TrixiNVTXExt.jl b/ext/TrixiNVTXExt.jl
new file mode 100644
index 00000000000..69c79c7a3d0
--- /dev/null
+++ b/ext/TrixiNVTXExt.jl
@@ -0,0 +1,25 @@
+module TrixiNVTXExt
+
+using NVTX
+using CUDA: CUDABackend
+import Trixi: trixi_range_active, trixi_range_start, trixi_range_end
+
+# One can also use Nsight Systems and thus NVTX for CPU code
+
+const domain = NVTX.Domain("Trixi")
+const color = 0xff40e0d0 # turquoise
+
+function trixi_range_active(::CUDABackend)
+    return NVTX.isactive()
+end
+
+function trixi_range_start(::CUDABackend, label)
+    return NVTX.range_start(NVTX.init!(domain); message = label, color = color)
+end
+
+function trixi_range_end(::CUDABackend, id)
+    NVTX.range_end(id)
+    return nothing
+end
+
+end # module
diff --git a/src/auxiliary/auxiliary.jl b/src/auxiliary/auxiliary.jl
index 28e4c47d339..887b8e10746 100644
--- a/src/auxiliary/auxiliary.jl
+++ b/src/auxiliary/auxiliary.jl
@@ -82,6 +82,35 @@ end
     return ncalls_first
 end
 
+# TODO: move to KernelAbstractions
+"""
+    trixi_range_active(backend)
+
+Returns `true` if the given `backend` supports range annotations and a profiler is active, `false` otherwise.
+"""
+function trixi_range_active(backend::Any)
+    return false
+end
+
+"""
+    trixi_range_start(backend, label)
+
+Starts a range annotation for the given `backend` with the specified `label`.
+Returns a handle to the started range, which should be passed to `trixi_range_end` to end the range annotation.
+"""
+function trixi_range_start(backend::Any, label)
+    return nothing
+end
+
+"""
+    trixi_range_end(backend, id)
+
+Ends a range annotation for the given `backend` with the specified `id`.
+"""
+function trixi_range_end(backend::Any, id)
+    return nothing
+end
+
 """
     @trixi_timeit_ext backend timer() "some label" expression
 
@@ -93,10 +122,17 @@ See also [`@trixi_timeit`](@ref).
 """
 macro trixi_timeit_ext(backend, timer_output, label, expr)
     expr = quote
+        local active = $trixi_range_active($(esc(backend)))
+        if active
+            id = $trixi_range_start($(esc(backend)), $(esc(label)))
+        end
         local val = $(esc(expr))
         if $(esc(backend)) !== nothing && $(TrixiBase).timeit_debug_enabled()
             $(KernelAbstractions.synchronize)($(esc(backend)))
         end
+        if active
+            $trixi_range_end($(esc(backend)), id)
+        end
         val
     end
     return :(@trixi_timeit($(esc(timer_output)), $(esc(label)), $(expr)))