diff --git a/rpcs3/Emu/RSX/GL/GLFragmentProgram.cpp b/rpcs3/Emu/RSX/GL/GLFragmentProgram.cpp
index 723058ecc9..24559b4b2b 100644
--- a/rpcs3/Emu/RSX/GL/GLFragmentProgram.cpp
+++ b/rpcs3/Emu/RSX/GL/GLFragmentProgram.cpp
@@ -214,6 +214,7 @@ void GLFragmentDecompilerThread::insertGlobalFunctions(std::stringstream &OS)
 	m_shader_props.low_precision_tests = ::gl::get_driver_caps().vendor_NVIDIA;
 	m_shader_props.disable_early_discard = !::gl::get_driver_caps().vendor_NVIDIA;
 	m_shader_props.supports_native_fp16 = device_props.has_native_half_support;
+	m_shader_props.srgb_output_rounding = ::gl::get_driver_caps().vendor_NVIDIA;
 
 	glsl::insert_glsl_legacy_function(OS, m_shader_props);
 }
diff --git a/rpcs3/Emu/RSX/Program/GLSLCommon.cpp b/rpcs3/Emu/RSX/Program/GLSLCommon.cpp
index 1c44360be4..ef14c2bdd1 100644
--- a/rpcs3/Emu/RSX/Program/GLSLCommon.cpp
+++ b/rpcs3/Emu/RSX/Program/GLSLCommon.cpp
@@ -510,10 +510,10 @@ namespace glsl
 				OS <<
 				"		else if (srgb_convert)\n"
 				"		{\n"
-				"			" << reg0 << ".rgb = clamp16(linear_to_srgb(" << reg0 << ")).rgb;\n"
-				"			" << reg1 << ".rgb = clamp16(linear_to_srgb(" << reg1 << ")).rgb;\n"
-				"			" << reg2 << ".rgb = clamp16(linear_to_srgb(" << reg2 << ")).rgb;\n"
-				"			" << reg3 << ".rgb = clamp16(linear_to_srgb(" << reg3 << ")).rgb;\n"
+				"			" << reg0 << " = round_to_8bit(f16vec4(linear_to_srgb(" << reg0 << ").rgb, " << reg0 << ".a));\n"
+				"			" << reg1 << " = round_to_8bit(f16vec4(linear_to_srgb(" << reg1 << ").rgb, " << reg1 << ".a));\n"
+				"			" << reg2 << " = round_to_8bit(f16vec4(linear_to_srgb(" << reg2 << ").rgb, " << reg2 << ".a));\n"
+				"			" << reg3 << " = round_to_8bit(f16vec4(linear_to_srgb(" << reg3 << ").rgb, " << reg3 << ".a));\n"
 				"		}\n";
 			}
 			else
@@ -521,10 +521,10 @@ namespace glsl
 				OS <<
 				"		else if (srgb_convert)\n"
 				"		{\n"
-				"			" << reg0 << ".rgb = linear_to_srgb(" << reg0 << ").rgb;\n"
-				"			" << reg1 << ".rgb = linear_to_srgb(" << reg1 << ").rgb;\n"
-				"			" << reg2 << ".rgb = linear_to_srgb(" << reg2 << ").rgb;\n"
-				"			" << reg3 << ".rgb = linear_to_srgb(" << reg3 << ").rgb;\n"
+				"			" << reg0 << " = round_to_8bit(vec4(linear_to_srgb(" << reg0 << ").rgb, " << reg0 << ".a));\n"
+				"			" << reg1 << " = round_to_8bit(vec4(linear_to_srgb(" << reg1 << ").rgb, " << reg1 << ".a));\n"
+				"			" << reg2 << " = round_to_8bit(vec4(linear_to_srgb(" << reg2 << ").rgb, " << reg2 << ".a));\n"
+				"			" << reg3 << " = round_to_8bit(vec4(linear_to_srgb(" << reg3 << ").rgb, " << reg3 << ".a));\n"
 				"		}\n";
 			}
 		}
@@ -561,6 +561,20 @@ namespace glsl
 				OS << "#define _kill() discard\n\n";
 			}
 
+			if (!props.fp32_outputs)
+			{
+				OS << "// Workaround broken output rounding behavior\n";
+				if (props.srgb_output_rounding)
+				{
+					const auto scale = (props.supports_native_fp16) ? "float16_t(255.)" : "255.";
+					OS << "#define round_to_8bit(v4) (round(v4 * " << scale << ") / " << scale << ")\n\n";
+				}
+				else
+				{
+					OS << "#define round_to_8bit(v4) (v4)\n\n";
+				}
+			}
+
 			if (props.require_texture_ops)
 			{
 				OS <<
diff --git a/rpcs3/Emu/RSX/Program/GLSLTypes.h b/rpcs3/Emu/RSX/Program/GLSLTypes.h
index ac9a917f76..6e1461aafa 100644
--- a/rpcs3/Emu/RSX/Program/GLSLTypes.h
+++ b/rpcs3/Emu/RSX/Program/GLSLTypes.h
@@ -37,5 +37,6 @@ namespace glsl
 		bool low_precision_tests : 1;
 		bool disable_early_discard : 1;
 		bool supports_native_fp16 : 1;
+		bool srgb_output_rounding : 1;
 	};
 };
diff --git a/rpcs3/Emu/RSX/VK/VKFragmentProgram.cpp b/rpcs3/Emu/RSX/VK/VKFragmentProgram.cpp
index 748caf8222..e3ea50afd9 100644
--- a/rpcs3/Emu/RSX/VK/VKFragmentProgram.cpp
+++ b/rpcs3/Emu/RSX/VK/VKFragmentProgram.cpp
@@ -251,6 +251,7 @@ void VKFragmentDecompilerThread::insertGlobalFunctions(std::stringstream &OS)
 	m_shader_props.low_precision_tests = device_props.has_low_precision_rounding;
 	m_shader_props.disable_early_discard = vk::get_driver_vendor() != vk::driver_vendor::NVIDIA;
 	m_shader_props.supports_native_fp16 = device_props.has_native_half_support;
+	m_shader_props.srgb_output_rounding = vk::get_driver_vendor() == vk::driver_vendor::NVIDIA;
 
 	glsl::insert_glsl_legacy_function(OS, m_shader_props);
 }