diff --git a/CHANGELOG.md b/CHANGELOG.md
index 4e2a40d5..6a1dabf9 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -17,6 +17,7 @@ and this project adheres to [Semantic Versioning](http://semver.org/spec/v2.0.0.
 - Make guided-alignment faster via sparse memory layout, add alignment points for EOS, remove losses other than ce.
 - Changed minimal C++ standard to C++-17
 - Faster LSH top-k search on CPU
+- Updated intgemm to the latest upstream version
 
 ## [1.11.0] - 2022-02-08
 
diff --git a/VERSION b/VERSION
index 3d461ead..f5f1545d 100644
--- a/VERSION
+++ b/VERSION
@@ -1 +1 @@
-v1.11.3
+v1.11.4
diff --git a/src/3rd_party/intgemm b/src/3rd_party/intgemm
index 8abde25b..a05a2e51 160000
--- a/src/3rd_party/intgemm
+++ b/src/3rd_party/intgemm
@@ -1 +1 @@
-Subproject commit 8abde25b13c3ab210c0dec8e23f4944e3953812d
+Subproject commit a05a2e51ab524bcee954a39ee72005193f3adf7c
diff --git a/src/tensors/cpu/expression_graph_packable.h b/src/tensors/cpu/expression_graph_packable.h
index f5a9cad9..1a233372 100644
--- a/src/tensors/cpu/expression_graph_packable.h
+++ b/src/tensors/cpu/expression_graph_packable.h
@@ -172,19 +172,19 @@ public:
           // Hardware-specific conversions which allow to implement memory-mapping and avoid conversion at runtime
           cpu::integer::passOrAbort(gemmElementType); // Check if the hardware supports the GEMM type
           if(isSsse3(gemmElementType)) {
-            intgemm::ssse3::Kernels8::PrepareBTransposed(tmp->data(), /*input*/
+            intgemm::SSSE3::Kernels8::PrepareBTransposed(tmp->data(), /*input*/
                                                     paramMat->data<int8_t>(), /*output*/
                                                     quantMult, /*Quant Mult*/
                                                     rows(val),
                                                     cols(val));
           } else if(isAvx2(gemmElementType)) {
-            intgemm::avx2::Kernels8::PrepareBTransposed(tmp->data(), /*input*/
+            intgemm::AVX2::Kernels8::PrepareBTransposed(tmp->data(), /*input*/
                                                    paramMat->data<int8_t>(), /*output*/
                                                    quantMult, /*Quant Mult*/
                                                    rows(val),
                                                    cols(val));
           } else if(isAvx512(gemmElementType)) {
-            intgemm::avx512bw::Kernels8::PrepareBTransposed(tmp->data(), /*input*/
+            intgemm::AVX512BW::Kernels8::PrepareBTransposed(tmp->data(), /*input*/
                                                      paramMat->data<int8_t>(), /*output*/
                                                      quantMult, /*Quant Mult*/
                                                      rows(val),
@@ -206,19 +206,19 @@ public:
            // Hardware-specific conversions which allow to implement memory-mapping and avoid conversion at runtime
            cpu::integer::passOrAbort(gemmElementType); // Check if the hardware supports the GEMM type
           if(isSse2(gemmElementType)) {
-            intgemm::sse2::Kernels16::PrepareBTransposed(tmp->data(), /*input*/
+            intgemm::SSE2::Kernels16::PrepareBTransposed(tmp->data(), /*input*/
                                                     paramMat->data<int16_t>(), /*output*/
                                                     quantMult, /*Quant Mult*/
                                                     rows(val),
                                                     cols(val));
           } else if(isAvx2(gemmElementType)) {
-            intgemm::avx2::Kernels16::PrepareBTransposed(tmp->data(), /*input*/
+            intgemm::AVX2::Kernels16::PrepareBTransposed(tmp->data(), /*input*/
                                                     paramMat->data<int16_t>(), /*output*/
                                                     quantMult, /*Quant Mult*/
                                                     rows(val),
                                                     cols(val));
           } else if(isAvx512(gemmElementType)) {
-            intgemm::avx512bw::Kernels16::PrepareBTransposed(tmp->data(), /*input*/
+            intgemm::AVX512BW::Kernels16::PrepareBTransposed(tmp->data(), /*input*/
                                                       paramMat->data<int16_t>(), /*output*/
                                                       quantMult, /*Quant Mult*/
                                                       rows(val),
diff --git a/src/tensors/cpu/integer_common.h b/src/tensors/cpu/integer_common.h
index cb372a74..f4e632b5 100644
--- a/src/tensors/cpu/integer_common.h
+++ b/src/tensors/cpu/integer_common.h
@@ -11,21 +11,21 @@
 namespace intgemm {
   struct Int8;
   struct Int16;
-  namespace ssse3 {
+  namespace SSSE3 {
     struct Kernels8;
   }
-  namespace sse2 {
+  namespace SSE2 {
     struct Kernels16;
   }
-  namespace avx2 {
+  namespace AVX2 {
     struct Kernels8;
     struct Kernels16;
   }
-  namespace avx512bw {
+  namespace AVX512BW {
     struct Kernels8;
     struct Kernels16;
   }
-  namespace avx512vnni {
+  namespace AVX512VNNI {
     struct Kernels8;
   }
 }
@@ -57,22 +57,22 @@ template <> struct intgemm_<Type::intgemm8> {
 };
 
 template <> struct intgemm_<Type::intgemm8ssse3> {
-  using width = intgemm::ssse3::Kernels8;
+  using width = intgemm::SSSE3::Kernels8;
   using type = int8_t;
 };
 
 template <> struct intgemm_<Type::intgemm8avx2> {
-  using width = intgemm::avx2::Kernels8;
+  using width = intgemm::AVX2::Kernels8;
   using type = int8_t;
 };
 
 template <> struct intgemm_<Type::intgemm8avx512> {
-  using width = intgemm::avx512bw::Kernels8;
+  using width = intgemm::AVX512BW::Kernels8;
   using type = int8_t;
 };
 
 template <> struct intgemm_<Type::intgemm8avx512vnni> {
-  using width = intgemm::avx512vnni::Kernels8;
+  using width = intgemm::AVX512VNNI::Kernels8;
   using type = int8_t;
 };
 
@@ -82,17 +82,17 @@ template <> struct intgemm_<Type::intgemm16> {
 };
 
 template <> struct intgemm_<Type::intgemm16sse2> {
-  using width = intgemm::sse2::Kernels16;
+  using width = intgemm::SSE2::Kernels16;
   using type = int16_t;
 };
 
 template <> struct intgemm_<Type::intgemm16avx2> {
-  using width = intgemm::avx2::Kernels16;
+  using width = intgemm::AVX2::Kernels16;
   using type = int16_t;
 };
 
 template <> struct intgemm_<Type::intgemm16avx512> {
-  using width = intgemm::avx512bw::Kernels16;
+  using width = intgemm::AVX512BW::Kernels16;
   using type = int16_t;
 };
 
@@ -220,4 +220,4 @@ void prepareAndTransposeB(io::Item& item, const char * input) {
 
 } //integer
 } //cpu
-} //marian
\ No newline at end of file
+} //marian