diff --git a/wgpu-hal/src/auxil/mod.rs b/wgpu-hal/src/auxil/mod.rs
index bf4470f8e..6a0e99a74 100644
--- a/wgpu-hal/src/auxil/mod.rs
+++ b/wgpu-hal/src/auxil/mod.rs
@@ -7,6 +7,9 @@ pub mod db {
         pub const DEVICE_KABY_LAKE_MASK: u32 = 0x5900;
         pub const DEVICE_SKY_LAKE_MASK: u32 = 0x1900;
     }
+    pub mod nvidia {
+        pub const VENDOR: u32 = 0x10DE;
+    }
 }
 
 pub fn map_naga_stage(stage: naga::ShaderStage) -> wgt::ShaderStages {
diff --git a/wgpu-hal/src/vulkan/adapter.rs b/wgpu-hal/src/vulkan/adapter.rs
index 45e84c6a3..0262dd61e 100644
--- a/wgpu-hal/src/vulkan/adapter.rs
+++ b/wgpu-hal/src/vulkan/adapter.rs
@@ -697,6 +697,7 @@ impl super::Instance {
                     .contains(vk::FormatFeatureFlags::DEPTH_STENCIL_ATTACHMENT)
             },
             non_coherent_map_mask: phd_capabilities.properties.limits.non_coherent_atom_size - 1,
+            can_present: true,
         };
 
         let capabilities = crate::Capabilities {
@@ -1018,6 +1019,10 @@ impl crate::Adapter<super::Api> for super::Adapter {
         &self,
         surface: &super::Surface,
     ) -> Option<crate::SurfaceCapabilities> {
+        if !self.private_caps.can_present {
+            return None;
+        }
+
         let queue_family_index = 0; //TODO
         match surface.functor.get_physical_device_surface_support(
             self.raw,
diff --git a/wgpu-hal/src/vulkan/instance.rs b/wgpu-hal/src/vulkan/instance.rs
index 9aaed7a95..0ac44cbc1 100644
--- a/wgpu-hal/src/vulkan/instance.rs
+++ b/wgpu-hal/src/vulkan/instance.rs
@@ -598,10 +598,33 @@ impl crate::Instance<super::Api> for super::Instance {
             }
         };
 
-        raw_devices
+        let mut exposed_adapters = raw_devices
             .into_iter()
             .flat_map(|device| self.expose_adapter(device))
-            .collect()
+            .collect::<Vec<_>>();
+
+        // detect if it's an Intel + NVidia configuration
+        if cfg!(target_os = "linux") {
+            use crate::auxil::db;
+            let has_nvidia_dgpu = exposed_adapters.iter().any(|exposed| {
+                exposed.info.device_type == wgt::DeviceType::DiscreteGpu
+                    && exposed.info.vendor == db::nvidia::VENDOR as usize
+            });
+            if has_nvidia_dgpu {
+                for exposed in exposed_adapters.iter_mut() {
+                    if exposed.info.device_type == wgt::DeviceType::IntegratedGpu
+                        && exposed.info.vendor == db::intel::VENDOR as usize
+                    {
+                        // See https://gitlab.freedesktop.org/mesa/mesa/-/issues/4688
+                        log::warn!("Disabling presentation on '{}' (id {:?}) because of an Nvidia dGPU (on Linux)",
+                            exposed.info.name, exposed.adapter.raw);
+                        exposed.adapter.private_caps.can_present = false;
+                    }
+                }
+            }
+        }
+
+        exposed_adapters
     }
 }
 
diff --git a/wgpu-hal/src/vulkan/mod.rs b/wgpu-hal/src/vulkan/mod.rs
index ddb350c60..8a8b732c5 100644
--- a/wgpu-hal/src/vulkan/mod.rs
+++ b/wgpu-hal/src/vulkan/mod.rs
@@ -158,6 +158,8 @@ struct PrivateCapabilities {
     timeline_semaphores: bool,
     texture_d24: bool,
     texture_d24_s8: bool,
+    /// Ability to present contents to any screen. Only needed to work around broken platform configurations.
+    can_present: bool,
     non_coherent_map_mask: wgt::BufferAddress,
 }
 
diff --git a/wgpu-types/src/lib.rs b/wgpu-types/src/lib.rs
index c58764390..5647ec294 100644
--- a/wgpu-types/src/lib.rs
+++ b/wgpu-types/src/lib.rs
@@ -809,6 +809,8 @@ pub enum DeviceType {
     Cpu,
 }
 
+//TODO: convert `vendor` and `device` to `u32`
+
 /// Information about an adapter.
 #[derive(Clone, Debug, PartialEq)]
 #[cfg_attr(feature = "trace", derive(serde::Serialize))]