Graphics Driver Issues

Graphics Driver Issues

This page quickly summarizes graphics driver quirks and issues present in GPU drivers by different manufacturers.

» Intel (Proprietary)
» AMD (Proprietary)
» NVIDIA (Proprietary)
» AMD (Mesa)
» Intel (Mesa)
» Hall of Fame (List of fixed issues)

Intel (Proprietary)

OpenGL

• GLSL compiler is very broken
  » Workaround implemented: No
  » Reported: No
  » Note: Intel proprietary OpenGL drivers are unsupported by RPCS3 due to their poor state.

• There are likely other issues as well...

Vulkan

• No known problems at this time

AMD (Proprietary)

OpenGL

• glMultiDrawArrays is broken
  » Workaround implemented: Yes
  » Reported: Yes, on 2018-04-25. No replies.
  » Note: This seems to be undefined behavior in the spec. The gl_VertexID propagation behavior across MultiDrawArrays is not defined, i.e whether it should continue incrementing of reset to 0 on a new instance. However, due to the comparison with a loop of [first, count] and implied functional equivalence, this may be interpreted as a bug. Either way, RPCS3 provides a workaround using index lookups.

Vulkan

• Primitive restart is 'broken', this seems to be a GCN hardware bug as it affects both proprietary and open-source drivers
  » Workaround implemented: Yes (automatic)
  » Note: Confirmed, see eae8f49f@mesa3d/mesa

• MSAA with sample shading disabled does not work on NAVI when using Vulkan
  » Workaround implemented: Yes (automatic)
  » Note: Forcefully enabling sampleRateShading for all pipelines and calculating the minimum ratio works around this problem

NVIDIA (Proprietary)

OpenGL

• Floating point precision issues with Nan/Inf values
  » Workaround implemented: Yes (automatic)
  » Note: This is not a driver bug as it can be considered implementation-defined behavior.

Vulkan

• Floating point precision issues with Nan/Inf values
  » Workaround implemented: Yes (automatic)
  » Note: This is not a driver bug as it can be considered implementation-defined behavior.

AMD (Mesa)

OpenGL

• No known problems at this time

Vulkan

• Primitive restart is 'broken', this seems to be a GCN hardware bug as it affects both proprietary and open-source drivers
  » Workaround implemented: Yes (automatic)
  » Note: Confirmed, see eae8f49f@mesa3d/mesa

• LLVM 8 generates broken code on Mesa
  » Workaround implemented: No, use newer versions of Mesa that are based on LLVM 9 or above
  » Reported: Yes, on 2019-06-22. Quick reply.
  » Note: This is caused by a dependency (LLVM), and not a bug on the driver's implementation.
  

Intel (Mesa)

OpenGL

• i965 Driver: Not supported. Games won't render properly with it, it's very broken [1], [2], [3].
• iris Driver: No known problems at this time.

Vulkan

• Diagonal corruption line
  » Workaround implemented: Yes, use Strict Rendering Mode
  » Reported: Yes, on 2020-03-22. Quickly replied.
  

• Missing shaderStorageImageMultisample Vulkan feature support on Intel iGPUs (RPCS3 crashes if MSAA is enabled)
  » Workaround implemented: None, disable Anti-Aliasing
  » Reported: Yes, on 2020-08-03. No solution given from the driver developers, feature unsupported in hardware.
  

---

Hall of Fame

List of reported graphics driver bugs that are now fixed. We'd like to thank the developers who fixed these!

• CRITICAL: OpenGL no longer renders any game on any AMD GPU! (Windows)
  » Affected drivers: Radeon Software Adrenalin 19.12.1, 19.12.2, 19.12.3, 20.1.1, 20.1.2, 20.1.3
  » Workaround used: None, not possible
  » Fixed on: Radeon Software Adrenalin 20.1.4

• High: AMD RADV driver crash when using ACO due to unusual control flow (Mesa)
  » Affected drivers: Mesa 20.0.2 and earlier
  » Workaround used: None, using RADV with LLVM instead of ACO
  » Fixed on: Mesa 20.0.3

• High: Incorrectly rendered Vulkan shaders due to broken SPIR-V optimization on NVIDIA
  » Affected drivers: NVIDIA Windows 452.06 and earlier, NVIDIA Linux 450.56.06 and earlier
  » Workaround used: None
  » Fixed on: NVIDIA Windows 452.28, NVIDIA Linux 450.56.11

• High: Unexpectedly slower performance (-25%) when replacing VK_FORMAT_B8G8R8A8_UNORM with VK_FORMAT_R8G8B8A8_UNORM on NVIDIA
  » Affected drivers: NVIDIA Windows 457.30 and earlier, NVIDIA Linux untested
  » Workaround used: None
  » Fixed on: NVIDIA Windows 457.51

• Normal: OpenGL's glGetTexImage/glGetTextureImage/glGetTextureImageEXT do not work with immutable textures if pack/unpack byteswap is requested on any AMD GPU (Windows)
  » Affected drivers: Radeon Software Adrenalin 19.4.2 and earlier
  » Workaround used: Manually byteswap texel data (now removed)
  » Fixed on: Radeon Software Adrenalin 19.4.3

• Normal: Unnecessary implicit subpass dependencies caused performance regression on AMD RADV (Mesa)
  » Affected drivers: Mesa 20.0.0, 20.0.1
  » Workaround used: Explicitely injecting null subpass dependencies (now removed)
  » Fixed on: Mesa 20.0.2

• Normal: vkGetQueryPoolResults returns incorrect values when VK_QUERY_RESULT_PARTIAL_BIT is set on NVIDIA (Windows)
  » Affected drivers: NVIDIA GeForce 442.59 and earlier
  » Workaround used: Keeping NVIDIA driver from using partial results (now removed)
  » Fixed on: ~NVIDIA GeForce 442.74

• Normal: Missing shaderStorageImageMultisample Vulkan feature support on GCN1 GPUs (RPCS3 crashes if MSAA is enabled)
  » Affected drivers: Mesa 20.0.8 and older
  » Workaround used: None (disable MSAA on affected drivers)
  » Fixed on: Mesa 20.1.0

• Normal: Vertex explosion when using ACO shader compiler on AMD RADV (Mesa)
  » Affected drivers: Mesa 20.2.1 and older
  » Workaround used: None, not possible. Using the older shader compiler instead of ACO, which was the default prior to Mesa 20.2
  » Fixed on: Mesa 20.2.2