diff --git a/src/graphics/shader/shader.cpp b/src/graphics/shader/shader.cpp
index b78d6e742..80b111818 100644
--- a/src/graphics/shader/shader.cpp
+++ b/src/graphics/shader/shader.cpp
@@ -583,16 +583,16 @@ ShaderParams PrepareProgram(const HW::VertexShaderInfo& regs, const HW::Context&
 		     mesh.input_primitive, sh.m_vgtGsOutPrimType, sh.m_vgtGsMaxVertOut,
 		     group.primitive_group_size, group.vertex_group_size, mesh.max_vertices);
 	}
-	mesh.max_primitives = mesh.fast_launch ? mesh.max_vertices :
-	                      native_ngg ? mesh.InputPrimitiveCount(mesh.max_vertices) :
-	                      group.primitive_group_size * (sh.m_vgtGsMaxVertOut - 2u);
 	mesh.primitives_per_group = std::min({static_cast<uint32_t>(group.primitive_group_size),
 	                                      mesh.InputPrimitiveCount(group.vertex_group_size),
-	                                      native_ngg ? mesh.max_primitives :
+	                                      native_ngg ? mesh.InputPrimitiveCount(mesh.max_vertices) :
 	                                          mesh.max_vertices / sh.m_vgtGsMaxVertOut});
 	EXIT_IF(mesh.primitives_per_group == 0u);
 	mesh.vertices_per_group = mesh.InputVertexCount(mesh.primitives_per_group);
-	const auto threads = native_ngg && !mesh.fast_launch ? mesh.vertices_per_group : mesh.max_vertices;
+	mesh.max_primitives = mesh.primitives_per_group * (sh.m_geNggSubgrpCntl & 0x1ffu);
+	const auto threads = std::max(native_ngg && !mesh.fast_launch ? mesh.vertices_per_group
+	                                                            : mesh.max_vertices,
+	                              mesh.max_primitives);
 	mesh.threads_num[0] = ((threads + mesh.wave_size - 1u) / mesh.wave_size) * mesh.wave_size;
 	mesh.threads_num[1] = mesh.threads_num[2] = 1u;
 	return params;
diff --git a/tests/shaderCfgTests.cpp b/tests/shaderCfgTests.cpp
index 8f351f584..50d326b80 100644
--- a/tests/shaderCfgTests.cpp
+++ b/tests/shaderCfgTests.cpp
@@ -11108,12 +11108,13 @@ void TestMergedShaderUserDataSnapshot() {
   ShaderMapUserData(regs.gs_regs.data_addr, mapped);
   HW::Context context;
   context.SetShaderStages(0x20);
-  context.SetMaxOutputPerSubgroup(252);
+  context.SetMaxOutputPerSubgroup(192);
   context.SetGsMaxVertOut(18);
+  context.SetNggSubgrpCntl(16);
   context.SetGsOutPrimType(2);
   HW::UserConfig user_config;
   user_config.SetPrimitiveType(Prospero::PrimitiveType::kTriList);
-  user_config.SetGeControl({14, 14});
+  user_config.SetGeControl({62, 192});
   ShaderVertexInputInfo first_input{};
   const auto first = PrepareProgram(regs, context, user_config, first_input);
   regs.gs_regs.user_data_addr = reinterpret_cast<uint64_t>(second_table.data());
@@ -11126,8 +11127,10 @@ void TestMergedShaderUserDataSnapshot() {
         "merged shader parameters did not snapshot ordinary user SGPRs at s8");
   Check(first.hash == second.hash &&
             MakeStageStaticKey(first_input) == MakeStageStaticKey(second_input) &&
-            first_input.mesh.scratch_size_dwords == 7,
-        "merged shader identity or maximum scratch size changed with dynamic user data");
+            first_input.mesh.scratch_size_dwords == 7 &&
+            first_input.mesh.primitives_per_group == 10 &&
+            first_input.mesh.max_primitives == 160 && first_input.mesh.threads_num[0] == 192,
+        "merged shader lost its identity, scratch size or amplified subgroup bounds");
   CompileOptions options{};
   options.stage = ShaderType::Mesh;
   options.user_data_base = 0;
@@ -11190,6 +11193,7 @@ void TestMergedShaderUserDataSnapshot() {
   regs.gs_regs.rsrc2.lds_size = 64;
   regs.gs_regs.rsrc1.gs_vgpr_component_count = 1;
   context.SetGsMaxVertOut(0);
+  context.SetNggSubgrpCntl(1);
   for (const uint32_t wave_size : {32u, 64u}) {
     context.SetShaderStages(wave_size == 32 ? 0x00400000u : 0u);
     for (const uint16_t capacity : {64u, 128u, 256u}) {
@@ -11215,18 +11219,31 @@ void TestMergedShaderUserDataSnapshot() {
   context.SetShaderStages(0x20);
   context.SetMaxOutputPerSubgroup(256);
   context.SetGsMaxVertOut(8);
+  context.SetNggSubgrpCntl(6);
   user_config.SetPrimitiveType(Prospero::PrimitiveType::kTriFan);
   user_config.SetGeControl({32, 32});
   ShaderVertexInputInfo fan_input{};
   PrepareProgram(regs, context, user_config, fan_input);
   Check(fan_input.mesh.primitives_per_group == 30 && fan_input.mesh.vertices_per_group == 32 &&
-            fan_input.mesh.max_vertices == 256 && fan_input.mesh.max_primitives == 192,
+            fan_input.mesh.max_vertices == 256 && fan_input.mesh.max_primitives == 180,
         "captured triangle-fan GS configuration lost its subgroup assembly limits");
-  context.SetShaderStages(0x00482030u);
+  context.SetShaderStages(0x00402030u);
   context.SetMaxOutputPerSubgroup(32);
   context.SetGsMaxVertOut(32);
   user_config.SetPrimitiveType(Prospero::PrimitiveType::kPointList);
   user_config.SetGeControl({1, 1});
+  for (const uint32_t amplification : {32u, 48u}) {
+    context.SetNggSubgrpCntl(amplification);
+    ShaderVertexInputInfo input{};
+    PrepareProgram(regs, context, user_config, input);
+    Check(!input.mesh.fast_launch && input.mesh.primitives_per_group == 1 &&
+              input.mesh.vertices_per_group == 1 && input.mesh.max_vertices == 32 &&
+              input.mesh.max_primitives == amplification &&
+              input.mesh.threads_num[0] == (amplification == 32u ? 32u : 64u),
+          "native NGG output primitives were derived from a vertex strip or omitted from subgroup size");
+  }
+  context.SetShaderStages(0x00482030u);
+  context.SetNggSubgrpCntl(32);
   regs.gs_regs.rsrc1.gs_vgpr_component_count = 0;
   regs.gs_regs.rsrc2.es_vgpr_component_count = 1;
   ShaderVertexInputInfo fast_input{};