diff --git a/Detectors/ITSMFT/ITS/tracking/GPU/cuda/TimeFrameGPU.cu b/Detectors/ITSMFT/ITS/tracking/GPU/cuda/TimeFrameGPU.cu
index d3c56ead7de6c..ff9ef6c80b9a3 100644
--- a/Detectors/ITSMFT/ITS/tracking/GPU/cuda/TimeFrameGPU.cu
+++ b/Detectors/ITSMFT/ITS/tracking/GPU/cuda/TimeFrameGPU.cu
@@ -418,8 +418,8 @@ void TimeFrameGPU<nLayers>::initDevice(IndexTableUtils* utils,
   // mVerticesInChunks.resize(mGpuParams.nTimeFrameChunks);
   // mNVerticesInChunks.resize(mGpuParams.nTimeFrameChunks);
   // mLabelsInChunks.resize(mGpuParams.nTimeFrameChunks);
-  // LOGP(debug, "Size of fixed part is: {} MB", GpuTimeFrameChunk<nLayers>::computeFixedSizeBytes(mGpuParams) / MB);
-  // LOGP(debug, "Size of scaling part is: {} MB", GpuTimeFrameChunk<nLayers>::computeScalingSizeBytes(GpuTimeFrameChunk<nLayers>::computeRofPerChunk(mGpuParams, mAvailMemGB), mGpuParams) / MB);
+  // LOGP(info, "Size of fixed part is: {} MB", GpuTimeFrameChunk<nLayers>::computeFixedSizeBytes(mGpuParams) / MB);
+  // LOGP(info, "Size of scaling part is: {} MB", GpuTimeFrameChunk<nLayers>::computeScalingSizeBytes(GpuTimeFrameChunk<nLayers>::computeRofPerChunk(mGpuParams, mAvailMemGB), mGpuParams) / MB);
   // LOGP(info, "Allocating {} chunks of {} rofs capacity each.", mGpuParams.nTimeFrameChunks, mGpuParams.nROFsPerChunk);
 
   // for (int iChunk{0}; iChunk < mMemChunks.size(); ++iChunk) {
@@ -571,7 +571,7 @@ template <int nLayers>
 void TimeFrameGPU<nLayers>::downloadTrackITSExtDevice(std::vector<CellSeed>& seeds)
 {
   LOGP(debug, "gpu-transfer: downloading {} tracks, for {} MB.", mTrackITSExt.size(), mTrackITSExt.size() * sizeof(o2::its::TrackITSExt) / MB);
-  checkGPUError(cudaMemcpyAsync(mTrackITSExt.data(), mTrackITSExtDevice, mTrackITSExt.size() * sizeof(o2::its::TrackITSExt), cudaMemcpyDeviceToHost, mGpuStreams[0].get()));
+  checkGPUError(cudaMemcpyAsync(mTrackITSExt.data(), mTrackITSExtDevice, seeds.size() * sizeof(o2::its::TrackITSExt), cudaMemcpyDeviceToHost, mGpuStreams[0].get()));
   checkGPUError(cudaHostUnregister(mTrackITSExt.data()));
   checkGPUError(cudaHostUnregister(seeds.data()));
   discardResult(cudaDeviceSynchronize());
diff --git a/Detectors/ITSMFT/ITS/tracking/GPU/cuda/TrackerTraitsGPU.cxx b/Detectors/ITSMFT/ITS/tracking/GPU/cuda/TrackerTraitsGPU.cxx
index 13a5b9f1ce9c3..7f1d6812bc6cd 100644
--- a/Detectors/ITSMFT/ITS/tracking/GPU/cuda/TrackerTraitsGPU.cxx
+++ b/Detectors/ITSMFT/ITS/tracking/GPU/cuda/TrackerTraitsGPU.cxx
@@ -380,7 +380,7 @@ void TrackerTraitsGPU<nLayers>::findRoads(const int iteration)
       for (int iLayer{startLayer - 1}; iLayer > 0 && level > 2; --iLayer) {
         lastCellSeed.swap(updatedCellSeed);
         lastCellId.swap(updatedCellId);
-        updatedCellSeed.clear();
+        std::vector<CellSeed>().swap(updatedCellSeed); /// tame the memory peaks
         updatedCellId.clear();
         processNeighbours(iLayer, --level, lastCellSeed, lastCellId, updatedCellSeed, updatedCellId);
       }