diff --git a/.github/workflows/ci.yaml b/.github/workflows/ci.yaml
index 5a9dd80d05..6d5447492d 100644
--- a/.github/workflows/ci.yaml
+++ b/.github/workflows/ci.yaml
@@ -101,9 +101,6 @@ jobs:
           - name: build-ubuntu-clang12
             env: {CXX: clang++-12, INSTALL_EXTRA: clang-12 libomp-12-dev}
             add-llvm-repo: true
-          - name: build-ubuntu-icpc
-            env: {CXX: icpc}
-            intel: true
           - name: build-ubuntu-icpx
             env: {CXX: icpx}
             intel: true
diff --git a/include/llama/BlobAllocators.hpp b/include/llama/BlobAllocators.hpp
index 8690b1ed93..b365b60fa8 100644
--- a/include/llama/BlobAllocators.hpp
+++ b/include/llama/BlobAllocators.hpp
@@ -10,9 +10,6 @@
 #include <cstddef>
 #include <memory>
 #include <vector>
-#ifdef __INTEL_COMPILER
-#    include <aligned_new>
-#endif
 #if defined(_LIBCPP_VERSION) && _LIBCPP_VERSION < 11000
 #    include <boost/shared_ptr.hpp>
 #endif
diff --git a/include/llama/Tuple.hpp b/include/llama/Tuple.hpp
index 45db85875c..f2c1835e96 100644
--- a/include/llama/Tuple.hpp
+++ b/include/llama/Tuple.hpp
@@ -30,8 +30,6 @@ namespace llama
         {
         }
 
-        // icpc fails to compile the treemap tests with this ctor
-#ifndef __INTEL_COMPILER
         /// Construct a tuple from forwarded values of potentially different types as the tuple stores.
         // SFINAE away this ctor if tuple elements cannot be constructed from ctor arguments
         template<
@@ -46,7 +44,6 @@ namespace llama
             , rest(std::forward<Ts>(restArgs)...)
         {
         }
-#endif
 
         FirstElement first; ///< the first element (if existing)
 #ifndef __NVCC__
diff --git a/include/llama/macros.hpp b/include/llama/macros.hpp
index c870c653c6..b4f38b6fca 100644
--- a/include/llama/macros.hpp
+++ b/include/llama/macros.hpp
@@ -3,7 +3,11 @@
 
 #pragma once
 
-#if defined(__INTEL_COMPILER) || defined(__INTEL_LLVM_COMPILER)
+#ifdef __INTEL_COMPILER
+#    error LLAMA has stopped supporting the Intel Classic Compiler after Intel announced its planned deprecation and replacement by the Intel LLVM-based compiler. Please migrate to Intel's LLVM-based compiler.
+#endif
+
+#if defined(__INTEL_LLVM_COMPILER)
 #    define LLAMA_INDEPENDENT_DATA _Pragma("ivdep")
 #elif defined(__clang__)
 #    define LLAMA_INDEPENDENT_DATA _Pragma("clang loop vectorize(assume_safety) interleave(assume_safety)")
@@ -28,7 +32,7 @@
 #        define LLAMA_FN_HOST_ACC_INLINE __host__ __device__ __forceinline__
 #    elif defined(__GNUC__) || defined(__clang__)
 #        define LLAMA_FN_HOST_ACC_INLINE inline __attribute__((always_inline))
-#    elif defined(_MSC_VER) || defined(__INTEL_COMPILER) || defined(__INTEL_LLVM_COMPILER)
+#    elif defined(_MSC_VER) || defined(__INTEL_LLVM_COMPILER)
 #        define LLAMA_FN_HOST_ACC_INLINE __forceinline
 #    else
 /// Some offloading parallelization language extensions such a CUDA, OpenACC or OpenMP 4.5 need to specify whether a
@@ -44,7 +48,7 @@
 #ifndef LLAMA_LAMBDA_INLINE_WITH_SPECIFIERS
 #    if defined(__clang__) || defined(__INTEL_LLVM_COMPILER)
 #        define LLAMA_LAMBDA_INLINE_WITH_SPECIFIERS(...) __attribute__((always_inline)) __VA_ARGS__
-#    elif defined(__GNUC__) || defined(__INTEL_COMPILER) || (defined(__NVCC__) && !defined(_MSC_VER))
+#    elif defined(__GNUC__) || (defined(__NVCC__) && !defined(_MSC_VER))
 #        define LLAMA_LAMBDA_INLINE_WITH_SPECIFIERS(...) __VA_ARGS__ __attribute__((always_inline))
 #    elif defined(_MSC_VER)
 #        define LLAMA_LAMBDA_INLINE_WITH_SPECIFIERS(...)                                                              \
@@ -66,9 +70,7 @@
 #    define LLAMA_SUPPRESS_HOST_DEVICE_WARNING
 #endif
 
-#if defined(__INTEL_COMPILER) /*|| defined(__INTEL_LLVM_COMPILER)*/
-#    define LLAMA_FORCE_INLINE_RECURSIVE _Pragma("forceinline recursive")
-#elif defined(_MSC_VER)
+#if defined(_MSC_VER)
 #    define LLAMA_FORCE_INLINE_RECURSIVE __pragma(inline_depth(255))
 #else
 /// Forces the compiler to recursively inline the call hiearchy started by the subsequent function call.