diff --git a/src/nnue/trainer/trainer_affine_transform.h b/src/nnue/trainer/trainer_affine_transform.h
index 3179aeb0..449a0a11 100644
--- a/src/nnue/trainer/trainer_affine_transform.h
+++ b/src/nnue/trainer/trainer_affine_transform.h
@@ -353,18 +353,18 @@ namespace Eval::NNUE {
         LayerType* const target_layer_;
 
         // parameter
-        LearnFloatType biases_[kOutputDimensions];
-        LearnFloatType weights_[kOutputDimensions * kInputDimensions];
+        alignas(kCacheLineSize) LearnFloatType biases_[kOutputDimensions];
+        alignas(kCacheLineSize) LearnFloatType weights_[kOutputDimensions * kInputDimensions];
 
         // Buffer used for updating parameters
-        LearnFloatType biases_diff_[kOutputDimensions];
-        LearnFloatType weights_diff_[kOutputDimensions * kInputDimensions];
+        alignas(kCacheLineSize) LearnFloatType biases_diff_[kOutputDimensions];
+        alignas(kCacheLineSize) LearnFloatType weights_diff_[kOutputDimensions * kInputDimensions];
 
         // Forward propagation buffer
-        std::vector<LearnFloatType> output_;
+        std::vector<LearnFloatType, CacheLineAlignedAllocator<LearnFloatType>> output_;
 
         // buffer for back propagation
-        std::vector<LearnFloatType> gradients_;
+        std::vector<LearnFloatType, CacheLineAlignedAllocator<LearnFloatType>> gradients_;
 
         // hyper parameter
         LearnFloatType momentum_;
diff --git a/src/nnue/trainer/trainer_clipped_relu.h b/src/nnue/trainer/trainer_clipped_relu.h
index 57e9bac4..5f2ff065 100644
--- a/src/nnue/trainer/trainer_clipped_relu.h
+++ b/src/nnue/trainer/trainer_clipped_relu.h
@@ -149,10 +149,10 @@ namespace Eval::NNUE {
         LayerType* const target_layer_;
 
         // Forward propagation buffer
-        std::vector<LearnFloatType> output_;
+        std::vector<LearnFloatType, CacheLineAlignedAllocator<LearnFloatType>> output_;
 
         // buffer for back propagation
-        std::vector<LearnFloatType> gradients_;
+        std::vector<LearnFloatType, CacheLineAlignedAllocator<LearnFloatType>> gradients_;
 
         // Health check statistics
         LearnFloatType min_activations_[kOutputDimensions];
diff --git a/src/nnue/trainer/trainer_feature_transformer.h b/src/nnue/trainer/trainer_feature_transformer.h
index 97b19c46..9f0648d2 100644
--- a/src/nnue/trainer/trainer_feature_transformer.h
+++ b/src/nnue/trainer/trainer_feature_transformer.h
@@ -421,11 +421,11 @@ namespace Eval::NNUE {
             LearnFloatType weights_[kHalfDimensions * kInputDimensions];
 
         // Buffer used for updating parameters
-        LearnFloatType biases_diff_[kHalfDimensions];
-        std::vector<LearnFloatType> gradients_;
+        alignas(kCacheLineSize) LearnFloatType biases_diff_[kHalfDimensions];
+        std::vector<LearnFloatType, CacheLineAlignedAllocator<LearnFloatType>> gradients_;
 
         // Forward propagation buffer
-        std::vector<LearnFloatType> output_;
+        std::vector<LearnFloatType, CacheLineAlignedAllocator<LearnFloatType>> output_;
 
         // Features that appeared in the training data
         std::bitset<kInputDimensions> observed_features;
@@ -437,8 +437,8 @@ namespace Eval::NNUE {
         // Health check statistics
         LearnFloatType min_pre_activation_;
         LearnFloatType max_pre_activation_;
-        LearnFloatType min_activations_[kHalfDimensions];
-        LearnFloatType max_activations_[kHalfDimensions];
+        alignas(kCacheLineSize) LearnFloatType min_activations_[kHalfDimensions];
+        alignas(kCacheLineSize) LearnFloatType max_activations_[kHalfDimensions];
     };
 
 }  // namespace Eval::NNUE
diff --git a/src/nnue/trainer/trainer_input_slice.h b/src/nnue/trainer/trainer_input_slice.h
index 43968776..9b8e5e13 100644
--- a/src/nnue/trainer/trainer_input_slice.h
+++ b/src/nnue/trainer/trainer_input_slice.h
@@ -163,7 +163,7 @@ namespace Eval::NNUE {
         const LearnFloatType* output_;
 
         // buffer for back propagation
-        std::vector<LearnFloatType> gradients_;
+        std::vector<LearnFloatType, CacheLineAlignedAllocator<LearnFloatType>> gradients_;
     };
 
     // Learning: Input layer
@@ -256,10 +256,10 @@ namespace Eval::NNUE {
         const std::shared_ptr<SharedInputTrainer> shared_input_trainer_;
 
         // Forward propagation buffer
-        std::vector<LearnFloatType> output_;
+        std::vector<LearnFloatType, CacheLineAlignedAllocator<LearnFloatType>> output_;
 
         // buffer for back propagation
-        std::vector<LearnFloatType> gradients_;
+        std::vector<LearnFloatType, CacheLineAlignedAllocator<LearnFloatType>> gradients_;
     };
 
 }  // namespace Eval::NNUE
diff --git a/src/nnue/trainer/trainer_sum.h b/src/nnue/trainer/trainer_sum.h
index c2e40b1c..b35420d6 100644
--- a/src/nnue/trainer/trainer_sum.h
+++ b/src/nnue/trainer/trainer_sum.h
@@ -184,7 +184,7 @@ namespace Eval::NNUE {
         LayerType* const target_layer_;
 
         // Forward propagation buffer
-        std::vector<LearnFloatType> output_;
+        std::vector<LearnFloatType, CacheLineAlignedAllocator<LearnFloatType>> output_;
     };
 
 }  // namespace Eval::NNUE