diff --git a/GPU/GPUTracking/DataCompression/GPUTPCDecompression.cxx b/GPU/GPUTracking/DataCompression/GPUTPCDecompression.cxx
index fd0c929dd2ba7..7e7ee86623099 100644
--- a/GPU/GPUTracking/DataCompression/GPUTPCDecompression.cxx
+++ b/GPU/GPUTracking/DataCompression/GPUTPCDecompression.cxx
@@ -18,6 +18,7 @@
 #include "GPUO2DataTypes.h"
 #include "GPUMemorySizeScalers.h"
 #include "GPULogging.h"
+#include <algorithm>
 
 using namespace o2::gpu;
 
@@ -116,5 +117,10 @@ void GPUTPCDecompression::RegisterMemoryAllocation()
 
 void GPUTPCDecompression::SetMaxData(const GPUTrackingInOutPointers& io)
 {
-  mMaxNativeClustersPerBuffer = mRec->GetProcessingSettings().tpcMaxAttachedClustersPerSectorRow;
+  uint32_t maxAttachedClsMargin1 = *std::max_element(mInputGPU.nSliceRowClusters, mInputGPU.nSliceRowClusters + mInputGPU.nSliceRows);
+  float clsRatio1 = (mInputGPU.nUnattachedClusters > 0 ? float(mInputGPU.nAttachedClusters) / float(mInputGPU.nUnattachedClusters) : 1.0f) * 1.5f;
+  maxAttachedClsMargin1 *= clsRatio1;
+  uint32_t maxAttachedClsMargin2 = mInputGPU.nAttachedClusters / mInputGPU.nSliceRows * 3.5;                                             // mean #attached cls per SectorRow multiplied by 3.5 (tuned)
+  mMaxNativeClustersPerBuffer = std::max({maxAttachedClsMargin1, maxAttachedClsMargin2, 1000u});                                         // take biggest margin, 1000 clusters minimum
+  mMaxNativeClustersPerBuffer = std::min(mMaxNativeClustersPerBuffer, mRec->GetProcessingSettings().tpcMaxAttachedClustersPerSectorRow); // upperbound given by configurable param
 }