diff --git a/src/gallium/frontends/mediafoundation/codecapi.cpp b/src/gallium/frontends/mediafoundation/codecapi.cpp
index ae572e551af..683d48b457a 100644
--- a/src/gallium/frontends/mediafoundation/codecapi.cpp
+++ b/src/gallium/frontends/mediafoundation/codecapi.cpp
@@ -1824,6 +1824,14 @@ CDX12EncHMFT::SetValue( const GUID *Api, VARIANT *Value )
       {
          CHECKHR_GOTO(E_INVALIDARG, done);
       }
+
+      if ((Value->ulVal > 0) && (!m_EncoderCapabilities.m_HWSupportSlicedFences.bits.supported))
+      {
+         MFE_ERROR("[dx12 hmft 0x%p] User tried to set CODECAPI_AVEncSliceGenerationMode: %d, but this encoder does NOT support sliced fence generation.",
+                   this, Value->ulVal);
+         CHECKHR_GOTO(E_INVALIDARG, done);
+      }
+
       debug_printf("[dx12 hmft 0x%p] SET CODECAPI_AVEncSliceGenerationMode - %u\n", this, Value->ulVal);
       m_uiSliceGenerationMode = Value->ulVal;
       m_bSliceGenerationModeSet = TRUE;
diff --git a/src/gallium/frontends/mediafoundation/encode.cpp b/src/gallium/frontends/mediafoundation/encode.cpp
index 7e5ef77fc0c..93a686f9edd 100644
--- a/src/gallium/frontends/mediafoundation/encode.cpp
+++ b/src/gallium/frontends/mediafoundation/encode.cpp
@@ -554,7 +554,7 @@ CDX12EncHMFT::PrepareForEncode( IMFSample *pSample, LPDX12EncodeContext *ppDX12E
 #endif
 
       pDX12EncodeContext->sliceNotificationMode = D3D12_VIDEO_ENCODER_COMPRESSED_BITSTREAM_NOTIFICATION_MODE_FULL_FRAME;
-      if( m_EncoderCapabilities.m_HWSupportSlicedFences.bits.supported && ( num_output_buffers > 1 ) )
+      if( m_bSliceGenerationModeSet && (m_uiSliceGenerationMode > 0) && ( num_output_buffers > 1 ) /* IHV driver requires > 1 slices */ )
       {
          pDX12EncodeContext->sliceNotificationMode = D3D12_VIDEO_ENCODER_COMPRESSED_BITSTREAM_NOTIFICATION_MODE_SUBREGIONS;
          if( m_EncoderCapabilities.m_HWSupportSlicedFences.bits.multiple_buffers_required )
diff --git a/src/gallium/frontends/mediafoundation/mftransform.cpp b/src/gallium/frontends/mediafoundation/mftransform.cpp
index e2723cc7f54..dd0e5fc1a2d 100644
--- a/src/gallium/frontends/mediafoundation/mftransform.cpp
+++ b/src/gallium/frontends/mediafoundation/mftransform.cpp
@@ -1443,11 +1443,19 @@ CDX12EncHMFT::xThreadProc( void *pCtx )
                   return result;
                };
 
-               // If slice generation mode is explicitly set to 1 (1 slice per output sample) and auto mode is off
-               // emit multiple output samples, one per slice
-               if (pThis->m_bSliceGenerationModeSet &&
-                  (pThis->m_uiSliceGenerationMode == 1) &&
-                  (!pDX12EncodeContext->IsSliceAutoModeEnabled())) // We cannot know if the last slice is actually the last one on time to set the last MFSample properties
+               //
+               // We can assume pLastSliceFence is signaled after the last pSliceFences[slice_idx] is signaled
+               // but there may still be a race condition there in between these last two signals
+               //
+               // If we are using PIPE_SLICE_AUTO_MODE, we need to wait on pLastSliceFence to know when all actual slices are done
+               // and what pSliceFences[] after that point are unused and must not be waited on.
+               //
+               // When emitting the MFSamples asynchronously for each slice, we need to mark MFSample_LastSlice
+               // on the last actual slice. In auto mode, we only know which one is the last actual slice after pLastSliceFence is signaled
+               // and by that time, it is too late to mark MFSample_LastSlice on the last slice only (as it may have been already emitted),
+               // so in PIPE_SLICE_AUTO_MODE, we gather all completed slices here and emit them together after pLastSliceFence is signaled.
+               //
+               if (!pDX12EncodeContext->IsSliceAutoModeEnabled())
                {
                   std::vector<struct codec_unit_location_t> codec_unit_metadata;
                   codec_unit_metadata.reserve( 16 );