diff --git a/src/amd/common/ac_nir.c b/src/amd/common/ac_nir.c
index 27636d41ac3..f96f72ea339 100644
--- a/src/amd/common/ac_nir.c
+++ b/src/amd/common/ac_nir.c
@@ -898,7 +898,8 @@ ac_nir_accum_ior(nir_builder *b, nir_def *accum_result, nir_def *new_term)
 bool
 ac_nir_gs_shader_query(nir_builder *b,
                        bool has_gen_prim_query,
-                       bool has_pipeline_stats_query,
+                       bool has_gs_invocations_query,
+                       bool has_gs_primitives_query,
                        unsigned num_vertices_per_primitive,
                        unsigned wave_size,
                        nir_def *vertex_count[4],
@@ -913,7 +914,7 @@ ac_nir_gs_shader_query(nir_builder *b,
       any_query_enabled = ac_nir_accum_ior(b, any_query_enabled, prim_gen_query_enabled);
    }
 
-   if (has_pipeline_stats_query) {
+   if (has_gs_invocations_query || has_gs_primitives_query) {
       pipeline_query_enabled = nir_load_pipeline_stat_query_enabled_amd(b);
       any_query_enabled = ac_nir_accum_ior(b, any_query_enabled, pipeline_query_enabled);
    }
@@ -959,7 +960,7 @@ ac_nir_gs_shader_query(nir_builder *b,
    /* Store the query result to query result using an atomic add. */
    nir_if *if_first_lane = nir_push_if(b, nir_elect(b, 1));
    {
-      if (has_pipeline_stats_query) {
+      if (has_gs_invocations_query || has_gs_primitives_query) {
          nir_if *if_pipeline_query = nir_push_if(b, pipeline_query_enabled);
          {
             nir_def *count = NULL;
@@ -974,10 +975,11 @@ ac_nir_gs_shader_query(nir_builder *b,
                }
             }
 
-            if (count)
+            if (has_gs_primitives_query && count)
                nir_atomic_add_gs_emit_prim_count_amd(b, count);
 
-            nir_atomic_add_shader_invocation_count_amd(b, num_active_threads);
+            if (has_gs_invocations_query)
+               nir_atomic_add_shader_invocation_count_amd(b, num_active_threads);
          }
          nir_pop_if(b, if_pipeline_query);
       }
@@ -1237,6 +1239,7 @@ ac_nir_lower_legacy_gs(nir_shader *nir,
    bool progress = ac_nir_gs_shader_query(b,
                                           has_gen_prim_query,
                                           has_pipeline_stats_query,
+                                          has_pipeline_stats_query,
                                           num_vertices_per_primitive,
                                           64,
                                           s.vertex_count,
diff --git a/src/amd/common/ac_nir.h b/src/amd/common/ac_nir.h
index 8699d30529d..011fd544ec6 100644
--- a/src/amd/common/ac_nir.h
+++ b/src/amd/common/ac_nir.h
@@ -166,6 +166,8 @@ typedef struct {
    bool disable_streamout;
    bool has_gen_prim_query;
    bool has_xfb_prim_query;
+   bool has_gs_invocations_query;
+   bool has_gs_primitives_query;
    bool kill_pointsize;
    bool kill_layer;
    bool force_vrs;
@@ -268,7 +270,8 @@ ac_nir_lower_legacy_vs(nir_shader *nir,
 bool
 ac_nir_gs_shader_query(nir_builder *b,
                        bool has_gen_prim_query,
-                       bool has_pipeline_stats_query,
+                       bool has_gs_invocations_query,
+                       bool has_gs_primitives_query,
                        unsigned num_vertices_per_primitive,
                        unsigned wave_size,
                        nir_def *vertex_count[4],
diff --git a/src/amd/common/ac_nir_lower_ngg.c b/src/amd/common/ac_nir_lower_ngg.c
index 74646f1c2bd..8e0e316009f 100644
--- a/src/amd/common/ac_nir_lower_ngg.c
+++ b/src/amd/common/ac_nir_lower_ngg.c
@@ -3573,7 +3573,8 @@ ac_nir_lower_ngg_gs(nir_shader *shader, const ac_nir_lower_ngg_options *options)
    b->cursor = nir_after_cf_list(&if_gs_thread->then_list);
    ac_nir_gs_shader_query(b,
                           state.options->has_gen_prim_query,
-                          state.options->gfx_level < GFX11,
+                          state.options->has_gs_invocations_query,
+                          state.options->has_gs_primitives_query,
                           state.num_vertices_per_primitive,
                           state.options->wave_size,
                           state.vertex_count,
diff --git a/src/amd/vulkan/radv_shader.c b/src/amd/vulkan/radv_shader.c
index db32f0a4224..821dde34bfa 100644
--- a/src/amd/vulkan/radv_shader.c
+++ b/src/amd/vulkan/radv_shader.c
@@ -856,6 +856,8 @@ radv_lower_ngg(struct radv_device *device, struct radv_shader_stage *ngg_stage,
    options.disable_streamout = !device->physical_device->use_ngg_streamout;
    options.has_gen_prim_query = info->has_prim_query;
    options.has_xfb_prim_query = info->has_xfb_query;
+   options.has_gs_invocations_query = device->physical_device->rad_info.gfx_level < GFX11;
+   options.has_gs_primitives_query = device->physical_device->rad_info.gfx_level < GFX11;
    options.force_vrs = info->force_vrs_per_vertex;
 
    if (nir->info.stage == MESA_SHADER_VERTEX || nir->info.stage == MESA_SHADER_TESS_EVAL) {
diff --git a/src/gallium/drivers/radeonsi/si_nir_lower_abi.c b/src/gallium/drivers/radeonsi/si_nir_lower_abi.c
index 8a99e9682b9..cf1f846842f 100644
--- a/src/gallium/drivers/radeonsi/si_nir_lower_abi.c
+++ b/src/gallium/drivers/radeonsi/si_nir_lower_abi.c
@@ -470,12 +470,17 @@ static bool lower_intrinsic(nir_builder *b, nir_instr *instr, struct lower_abi_s
    }
    case nir_intrinsic_atomic_add_gs_emit_prim_count_amd:
    case nir_intrinsic_atomic_add_shader_invocation_count_amd: {
-      nir_def *buf =
-         si_nir_load_internal_binding(b, args, SI_GS_QUERY_EMULATED_COUNTERS_BUF, 4);
-
       enum pipe_statistics_query_index index =
          intrin->intrinsic == nir_intrinsic_atomic_add_gs_emit_prim_count_amd ?
          PIPE_STAT_QUERY_GS_PRIMITIVES : PIPE_STAT_QUERY_GS_INVOCATIONS;
+
+      /* GFX11 only needs to emulate PIPE_STAT_QUERY_GS_PRIMITIVES because GS culls,
+       * which makes the pipeline statistic incorrect.
+       */
+      assert(sel->screen->info.gfx_level < GFX11 || index == PIPE_STAT_QUERY_GS_PRIMITIVES);
+
+      nir_def *buf =
+         si_nir_load_internal_binding(b, args, SI_GS_QUERY_EMULATED_COUNTERS_BUF, 4);
       unsigned offset = si_query_pipestat_end_dw_offset(sel->screen, index) * 4;
 
       nir_def *count = intrin->src[0].ssa;
diff --git a/src/gallium/drivers/radeonsi/si_query.c b/src/gallium/drivers/radeonsi/si_query.c
index 93736254ca9..55e768d7c15 100644
--- a/src/gallium/drivers/radeonsi/si_query.c
+++ b/src/gallium/drivers/radeonsi/si_query.c
@@ -693,6 +693,12 @@ static struct pipe_query *si_query_hw_create(struct si_screen *sscreen, unsigned
       if ((index == PIPE_STAT_QUERY_GS_PRIMITIVES || index == PIPE_STAT_QUERY_GS_INVOCATIONS) &&
           sscreen->use_ngg && (sscreen->info.gfx_level >= GFX10 && sscreen->info.gfx_level <= GFX10_3))
          query->flags |= SI_QUERY_EMULATE_GS_COUNTERS;
+
+      /* GFX11 only emulates PIPE_STAT_QUERY_GS_PRIMITIVES because the shader culls,
+       * which makes the statistic incorrect.
+       */
+      if (sscreen->info.gfx_level >= GFX11 && index == PIPE_STAT_QUERY_GS_PRIMITIVES)
+         query->flags |= SI_QUERY_EMULATE_GS_COUNTERS;
       break;
    default:
       assert(0);
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 0645de43256..4fa430fdbbf 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -1976,6 +1976,8 @@ static void si_lower_ngg(struct si_shader *shader, nir_shader *nir)
       options.gs_out_vtx_bytes = sel->info.gsvs_vertex_size;
       options.has_gen_prim_query = options.has_xfb_prim_query =
          sel->screen->info.gfx_level >= GFX11;
+      options.has_gs_invocations_query = sel->screen->info.gfx_level < GFX11;
+      options.has_gs_primitives_query = true;
 
       /* For monolithic ES/GS to add vscnt wait when GS export pos0. */
       if (key->ge.part.gs.es)