diff --git a/qrenderdoc/Code/CaptureContext.cpp b/qrenderdoc/Code/CaptureContext.cpp
index a3dcff75e..c0c66e850 100644
--- a/qrenderdoc/Code/CaptureContext.cpp
+++ b/qrenderdoc/Code/CaptureContext.cpp
@@ -168,6 +168,11 @@ void CaptureContext::LoadCapture(const rdcstr &captureFile, const rdcstr &origFi
   ANALYTIC_SET(CaptureFeatures.MultiGPU, m_APIProps.MultiGPU);
   ANALYTIC_SET(CaptureFeatures.D3D12Bundle, m_APIProps.D3D12Bundle);
 
+  if(m_APIProps.vendor != GPUVendor::Unknown)
+  {
+    ANALYTIC_ADDUNIQ(GPUVendors, ToQStr(m_APIProps.vendor));
+  }
+
   m_MainWindow->setProgress(-1.0f);
 
   if(m_CaptureLoaded)
diff --git a/renderdoc/api/replay/data_types.h b/renderdoc/api/replay/data_types.h
index aeb7bde09..5b00b0822 100644
--- a/renderdoc/api/replay/data_types.h
+++ b/renderdoc/api/replay/data_types.h
@@ -1074,6 +1074,9 @@ different to the above, and lets the UI make decisions e.g. to flip rendering of
 )");
   GraphicsAPI localRenderer = GraphicsAPI::D3D11;
 
+  DOCUMENT("The :class:`GPUVendor` of the active GPU being used.");
+  GPUVendor vendor = GPUVendor::Unknown;
+
   DOCUMENT(R"(``True`` if the capture was loaded successfully but running in a degraded mode - e.g.
 with software rendering, or with some functionality disabled due to lack of support.
 )");
diff --git a/renderdoc/api/replay/renderdoc_tostr.inl b/renderdoc/api/replay/renderdoc_tostr.inl
index 0e02a287f..0333b93ee 100644
--- a/renderdoc/api/replay/renderdoc_tostr.inl
+++ b/renderdoc/api/replay/renderdoc_tostr.inl
@@ -769,6 +769,25 @@ std::string DoStringise(const DebugOverlay &el)
   END_ENUM_STRINGISE();
 }
 
+template <>
+std::string DoStringise(const GPUVendor &el)
+{
+  BEGIN_ENUM_STRINGISE(GPUVendor)
+  {
+    STRINGISE_ENUM_CLASS(Unknown);
+    STRINGISE_ENUM_CLASS(ARM);
+    STRINGISE_ENUM_CLASS(AMD);
+    STRINGISE_ENUM_CLASS(Broadcom);
+    STRINGISE_ENUM_CLASS(Imagination);
+    STRINGISE_ENUM_CLASS(Intel);
+    STRINGISE_ENUM_CLASS(nVidia);
+    STRINGISE_ENUM_CLASS(Qualcomm);
+    STRINGISE_ENUM_CLASS(Verisilicon);
+    STRINGISE_ENUM_CLASS(Software);
+  }
+  END_ENUM_STRINGISE();
+}
+
 template <>
 std::string DoStringise(const GraphicsAPI &el)
 {
diff --git a/renderdoc/api/replay/replay_enums.h b/renderdoc/api/replay/replay_enums.h
index 78a32f954..172afd6d5 100644
--- a/renderdoc/api/replay/replay_enums.h
+++ b/renderdoc/api/replay/replay_enums.h
@@ -1207,6 +1207,79 @@ enum class QualityHint : uint32_t
 
 DECLARE_REFLECTION_ENUM(QualityHint);
 
+DOCUMENT(R"(Identifies a GPU vendor.
+
+.. data:: Unknown
+
+  A GPU from an unknown vendor
+
+.. data:: ARM
+
+  An ARM GPU
+
+.. data:: AMD
+
+  An AMD GPU
+
+.. data:: Broadcom
+
+  A Broadcom GPU
+
+.. data:: Imagination
+
+  An Imagination GPU
+
+.. data:: Intel
+
+  An Intel GPU
+
+.. data:: nVidia
+
+  An nVidia GPU
+
+.. data:: Qualcomm
+
+  A Qualcomm  GPU
+
+.. data:: Verisilicon
+
+  A Verisilicon or Vivante GPU
+
+.. data:: Software
+
+  A software-rendering emulated GPU
+)");
+enum class GPUVendor : uint32_t
+{
+  Unknown,
+  ARM,
+  AMD,
+  Broadcom,
+  Imagination,
+  Intel,
+  nVidia,
+  Qualcomm,
+  Verisilicon,
+  Software,
+};
+
+DECLARE_REFLECTION_ENUM(GPUVendor);
+
+constexpr GPUVendor GPUVendorFromPCIVendor(uint32_t vendorID)
+{
+  // temporarily disable clang-format to make this more readable.
+  // Ideally we'd use a simple switch() but VS2015 doesn't support that :(.
+  // clang-format off
+  return vendorID == 0x13B5 ? GPUVendor::ARM
+       : vendorID == 0x1002 ? GPUVendor::AMD
+       : vendorID == 0x1010 ? GPUVendor::Imagination
+       : vendorID == 0x8086 ? GPUVendor::Intel
+       : vendorID == 0x10DE ? GPUVendor::nVidia
+       : vendorID == 0x5143 ? GPUVendor::Qualcomm
+       : GPUVendor::Unknown;
+  // clang-format on
+}
+
 DOCUMENT(R"(Identifies a Graphics API.
 
 .. data:: D3D11
diff --git a/renderdoc/driver/d3d11/d3d11_replay.cpp b/renderdoc/driver/d3d11/d3d11_replay.cpp
index b18556f9d..9254663de 100644
--- a/renderdoc/driver/d3d11/d3d11_replay.cpp
+++ b/renderdoc/driver/d3d11/d3d11_replay.cpp
@@ -93,6 +93,14 @@ void D3D11Replay::CreateResources()
     IDXGIAdapter *pDXGIAdapter;
     hr = pDXGIDevice->GetParent(__uuidof(IDXGIAdapter), (void **)&pDXGIAdapter);
 
+    DXGI_ADAPTER_DESC desc = {};
+    pDXGIAdapter->GetDesc(&desc);
+
+    m_Vendor = GPUVendorFromPCIVendor(desc.VendorId);
+
+    if(m_WARP)
+      m_Vendor = GPUVendor::Software;
+
     if(FAILED(hr))
     {
       RDCERR("Couldn't get DXGI adapter from DXGI device");
@@ -450,6 +458,7 @@ APIProperties D3D11Replay::GetAPIProperties()
 
   ret.pipelineType = GraphicsAPI::D3D11;
   ret.localRenderer = GraphicsAPI::D3D11;
+  ret.vendor = m_Vendor;
   ret.degraded = m_WARP;
   ret.shadersMutable = false;
 
diff --git a/renderdoc/driver/d3d11/d3d11_replay.h b/renderdoc/driver/d3d11/d3d11_replay.h
index 9aff49a83..ee9c336c0 100644
--- a/renderdoc/driver/d3d11/d3d11_replay.h
+++ b/renderdoc/driver/d3d11/d3d11_replay.h
@@ -231,6 +231,8 @@ private:
   bool m_WARP;
   bool m_Proxy;
 
+  GPUVendor m_Vendor = GPUVendor::Unknown;
+
   D3D11DebugManager *GetDebugManager();
   // shared by BuildCustomShader and BuildTargetShader
   void BuildShader(std::string source, std::string entry, const ShaderCompileFlags &compileFlags,
diff --git a/renderdoc/driver/d3d12/d3d12_replay.cpp b/renderdoc/driver/d3d12/d3d12_replay.cpp
index b45015b02..60a5a5d75 100644
--- a/renderdoc/driver/d3d12/d3d12_replay.cpp
+++ b/renderdoc/driver/d3d12/d3d12_replay.cpp
@@ -83,6 +83,26 @@ void D3D12Replay::CreateResources()
       RDCERR("Couldn't create DXGI factory! HRESULT: %s", ToStr(hr).c_str());
     }
 
+    if(m_pFactory)
+    {
+      LUID luid = m_pDevice->GetAdapterLuid();
+
+      IDXGIAdapter *pDXGIAdapter;
+      hr = m_pFactory->EnumAdapterByLuid(luid, __uuidof(IDXGIAdapter), (void **)&pDXGIAdapter);
+
+      if(FAILED(hr))
+      {
+        RDCERR("Couldn't get DXGI adapter by LUID from D3D device");
+      }
+      else
+      {
+        DXGI_ADAPTER_DESC desc = {};
+        pDXGIAdapter->GetDesc(&desc);
+
+        m_Vendor = GPUVendorFromPCIVendor(desc.VendorId);
+      }
+    }
+
     m_DebugManager = new D3D12DebugManager(m_pDevice);
 
     CreateSOBuffers();
@@ -128,6 +148,7 @@ APIProperties D3D12Replay::GetAPIProperties()
 
   ret.pipelineType = GraphicsAPI::D3D12;
   ret.localRenderer = GraphicsAPI::D3D12;
+  ret.vendor = m_Vendor;
   ret.degraded = false;
   ret.shadersMutable = false;
 
diff --git a/renderdoc/driver/d3d12/d3d12_replay.h b/renderdoc/driver/d3d12/d3d12_replay.h
index df978843a..443cb79c3 100644
--- a/renderdoc/driver/d3d12/d3d12_replay.h
+++ b/renderdoc/driver/d3d12/d3d12_replay.h
@@ -262,6 +262,8 @@ private:
 
   bool m_Proxy;
 
+  GPUVendor m_Vendor = GPUVendor::Unknown;
+
   vector<ID3D12Resource *> m_ProxyResources;
 
   struct OutputWindow
diff --git a/renderdoc/driver/gl/gl_debug.cpp b/renderdoc/driver/gl/gl_debug.cpp
index 6b1f27463..86ad7df54 100644
--- a/renderdoc/driver/gl/gl_debug.cpp
+++ b/renderdoc/driver/gl/gl_debug.cpp
@@ -622,6 +622,74 @@ void GLReplay::InitDebugData()
 
   MakeCurrentReplayContext(&m_ReplayCtx);
 
+  // try to identify the GPU we're running on.
+  {
+    const char *vendor = (const char *)gl.glGetString(eGL_VENDOR);
+    const char *renderer = (const char *)gl.glGetString(eGL_RENDERER);
+
+    // we're just doing substring searches, so combine both for ease.
+    std::string combined = (vendor ? vendor : "");
+    combined += " ";
+    combined += (renderer ? renderer : "");
+
+    // make lowercase, for case-insensitive matching, and add preceding/trailing space for easier
+    // 'word' matching
+    combined = " " + strlower(combined) + " ";
+
+    RDCDEBUG("Identifying vendor from '%s'", combined.c_str());
+
+    struct pattern
+    {
+      const char *search;
+      GPUVendor vendor;
+    } patterns[] = {
+        {" arm ", GPUVendor::ARM},
+        {" mali ", GPUVendor::ARM},
+        {" mali-", GPUVendor::ARM},
+        {" amd ", GPUVendor::AMD},
+        {"advanced micro devices", GPUVendor::AMD},
+        {"ati technologies", GPUVendor::AMD},
+        {"radeon", GPUVendor::AMD},
+        {"broadcom", GPUVendor::Broadcom},
+        {"imagination", GPUVendor::Imagination},
+        {"powervr", GPUVendor::Imagination},
+        {"intel", GPUVendor::Intel},
+        {"geforce", GPUVendor::nVidia},
+        {"quadro", GPUVendor::nVidia},
+        {"nouveau", GPUVendor::nVidia},
+        {"nvidia", GPUVendor::nVidia},
+        {"adreno", GPUVendor::Qualcomm},
+        {"qualcomm", GPUVendor::Qualcomm},
+        {"vivante", GPUVendor::Verisilicon},
+        {"llvmpipe", GPUVendor::Software},
+        {"softpipe", GPUVendor::Software},
+        {"bluestacks", GPUVendor::Software},
+    };
+
+    for(const pattern &p : patterns)
+    {
+      if(combined.find(p.search) != std::string::npos)
+      {
+        if(m_Vendor == GPUVendor::Unknown)
+        {
+          m_Vendor = p.vendor;
+        }
+        else
+        {
+          // either we already found this with another pattern, or we've identified two patterns and
+          // it's ambiguous. Keep the first one we found, arbitrarily, but print a warning.
+          if(m_Vendor != p.vendor)
+          {
+            RDCWARN("Already identified '%s' as %s, but now identified as %s", combined.c_str(),
+                    ToStr(m_Vendor).c_str(), ToStr(p.vendor).c_str());
+          }
+        }
+      }
+    }
+
+    RDCDEBUG("Identified GPU vendor '%s'", ToStr(m_Vendor).c_str());
+  }
+
   // these below need to be made on the replay context, as they are context-specific (not shared)
   // and will be used on the replay context.
 
diff --git a/renderdoc/driver/gl/gl_replay.cpp b/renderdoc/driver/gl/gl_replay.cpp
index ef4334b68..3bf974d29 100644
--- a/renderdoc/driver/gl/gl_replay.cpp
+++ b/renderdoc/driver/gl/gl_replay.cpp
@@ -143,6 +143,7 @@ APIProperties GLReplay::GetAPIProperties()
   ret.pipelineType = GraphicsAPI::OpenGL;
   ret.localRenderer = GraphicsAPI::OpenGL;
   ret.degraded = m_Degraded;
+  ret.vendor = m_Vendor;
   ret.shadersMutable = true;
 
   return ret;
diff --git a/renderdoc/driver/gl/gl_replay.h b/renderdoc/driver/gl/gl_replay.h
index 91f7b63d8..406b82610 100644
--- a/renderdoc/driver/gl/gl_replay.h
+++ b/renderdoc/driver/gl/gl_replay.h
@@ -365,6 +365,8 @@ private:
 
   bool m_Degraded;
 
+  GPUVendor m_Vendor = GPUVendor::Unknown;
+
   HighlightCache m_HighlightCache;
 
   // eventId -> data
diff --git a/renderdoc/driver/vulkan/vk_common.cpp b/renderdoc/driver/vulkan/vk_common.cpp
index f3199d74a..52e49b466 100644
--- a/renderdoc/driver/vulkan/vk_common.cpp
+++ b/renderdoc/driver/vulkan/vk_common.cpp
@@ -27,10 +27,6 @@
 #include "vk_manager.h"
 #include "vk_resources.h"
 
-const uint32_t AMD_PCI_ID = 0x1002;
-const uint32_t NV_PCI_ID = 0x10DE;
-const uint32_t QUALCOMM_PCI_ID = 0x5143;
-
 // utility struct for firing one-shot command buffers to begin/end markers
 struct ScopedCommandBuffer
 {
@@ -1578,14 +1574,11 @@ INSTANTIATE_SERIALISE_TYPE(VkInitParams);
 
 VkDriverInfo::VkDriverInfo(const VkPhysicalDeviceProperties &physProps)
 {
-  if(physProps.vendorID == AMD_PCI_ID)
-    m_Vendor = AMD;
-  else if(physProps.vendorID == NV_PCI_ID)
-    m_Vendor = NV;
-  else if(physProps.vendorID == QUALCOMM_PCI_ID)
-    m_Vendor = QUALCOMM;
-  else
-    m_Vendor = UNKNOWN;
+  m_Vendor = GPUVendorFromPCIVendor(physProps.vendorID);
+
+  // add non-PCI vendor IDs
+  if(physProps.vendorID == 0x10002)
+    m_Vendor = GPUVendor::Verisilicon;
 
   m_Major = VK_VERSION_MAJOR(physProps.driverVersion);
   m_Minor = VK_VERSION_MINOR(physProps.driverVersion);
@@ -1594,7 +1587,7 @@ VkDriverInfo::VkDriverInfo(const VkPhysicalDeviceProperties &physProps)
   // nvidia uses its own version packing:
   //   10 |  8  |        8       |       6
   // major|minor|secondary_branch|tertiary_branch
-  if(IsNV())
+  if(m_Vendor == GPUVendor::nVidia)
   {
     m_Major = ((uint32_t)(physProps.driverVersion) >> (8 + 8 + 6)) & 0x3ff;
     m_Minor = ((uint32_t)(physProps.driverVersion) >> (8 + 6)) & 0x0ff;
@@ -1605,7 +1598,7 @@ VkDriverInfo::VkDriverInfo(const VkPhysicalDeviceProperties &physProps)
     m_Patch = (secondary << 8) | tertiary;
   }
 
-  if(IsNV())
+  if(m_Vendor == GPUVendor::nVidia)
   {
     // drivers before 372.54 did not handle a glslang bugfix about separated samplers,
     // and disabling texelFetch works as a workaround.
@@ -1618,7 +1611,7 @@ VkDriverInfo::VkDriverInfo(const VkPhysicalDeviceProperties &physProps)
 // using the AMD official driver, but there's not a great other way to distinguish it from
 // the RADV open source driver.
 #if ENABLED(RDOC_WIN32)
-  if(IsAMD())
+  if(m_Vendor == GPUVendor::AMD)
   {
     // for AMD the bugfix version isn't clear as version numbering wasn't strong for a while, but
     // any driver that reports a version of >= 1.0.0 is fine, as previous versions all reported
@@ -1636,7 +1629,7 @@ VkDriverInfo::VkDriverInfo(const VkPhysicalDeviceProperties &physProps)
 
 // same as above, only affects the AMD official driver
 #if ENABLED(RDOC_WIN32)
-  if(IsAMD())
+  if(m_Vendor == GPUVendor::AMD)
   {
     // not fixed yet
     amdStorageMSAABrokenDriver = true;
@@ -1644,5 +1637,5 @@ VkDriverInfo::VkDriverInfo(const VkPhysicalDeviceProperties &physProps)
 #endif
 
   // not fixed yet
-  qualcommLeakingUBOOffsets = IsQualcomm();
+  qualcommLeakingUBOOffsets = m_Vendor == GPUVendor::Qualcomm;
 }
diff --git a/renderdoc/driver/vulkan/vk_common.h b/renderdoc/driver/vulkan/vk_common.h
index 5e4f73a0d..3c607b0c7 100644
--- a/renderdoc/driver/vulkan/vk_common.h
+++ b/renderdoc/driver/vulkan/vk_common.h
@@ -178,16 +178,10 @@ struct GPUBuffer
 // in vk_<platform>.cpp
 extern const char *VulkanLibraryName;
 
-extern const uint32_t AMD_PCI_ID;
-extern const uint32_t NV_PCI_ID;
-extern const uint32_t QUALCOMM_PCI_ID;
-
 class VkDriverInfo
 {
 public:
-  bool IsAMD() { return m_Vendor == AMD; }
-  bool IsNV() { return m_Vendor == NV; }
-  bool IsQualcomm() { return m_Vendor == QUALCOMM; }
+  GPUVendor Vendor() { return m_Vendor; }
   uint32_t Major() { return m_Major; }
   uint32_t Minor() { return m_Minor; }
   uint32_t Patch() { return m_Patch; }
@@ -206,13 +200,7 @@ public:
   // rendering on other descriptor sets that don't use offsets at all.
   bool QualcommLeakingUBOOffsets() { return qualcommLeakingUBOOffsets; }
 private:
-  enum
-  {
-    AMD,
-    NV,
-    QUALCOMM,
-    UNKNOWN,
-  } m_Vendor;
+  GPUVendor m_Vendor;
 
   uint32_t m_Major, m_Minor, m_Patch;
 
diff --git a/renderdoc/driver/vulkan/wrappers/vk_device_funcs.cpp b/renderdoc/driver/vulkan/wrappers/vk_device_funcs.cpp
index a0735f081..58dfcd799 100644
--- a/renderdoc/driver/vulkan/wrappers/vk_device_funcs.cpp
+++ b/renderdoc/driver/vulkan/wrappers/vk_device_funcs.cpp
@@ -1237,6 +1237,8 @@ bool WrappedVulkan::Serialise_vkCreateDevice(SerialiserType &ser, VkPhysicalDevi
       }
     }
 
+    APIProps.vendor = GetDriverVersion().Vendor();
+
     m_ShaderCache = new VulkanShaderCache(this);
 
     m_DebugManager = new VulkanDebugManager(this);
diff --git a/renderdoc/driver/vulkan/wrappers/vk_get_funcs.cpp b/renderdoc/driver/vulkan/wrappers/vk_get_funcs.cpp
index 0edb97782..e988bada3 100644
--- a/renderdoc/driver/vulkan/wrappers/vk_get_funcs.cpp
+++ b/renderdoc/driver/vulkan/wrappers/vk_get_funcs.cpp
@@ -165,7 +165,7 @@ void WrappedVulkan::vkGetImageMemoryRequirements(VkDevice device, VkImage image,
   // allow for this. The variability isn't quite clear, but for now we assume aligning size to
   // alignment * 4 should be sufficient (adding on a fixed padding won't help the problem as it
   // won't remove the variability, nor will adding then aligning for the same reason).
-  if(GetDriverVersion().IsAMD() && pMemoryRequirements->size > 0)
+  if(GetDriverVersion().Vendor() == GPUVendor::AMD && pMemoryRequirements->size > 0)
   {
     VkMemoryRequirements &memreq = *pMemoryRequirements;
 
@@ -247,7 +247,7 @@ void WrappedVulkan::vkGetImageMemoryRequirements2KHR(VkDevice device,
   // allow for this. The variability isn't quite clear, but for now we assume aligning size to
   // alignment * 4 should be sufficient (adding on a fixed padding won't help the problem as it
   // won't remove the variability, nor will adding then aligning for the same reason).
-  if(GetDriverVersion().IsAMD() && pMemoryRequirements->memoryRequirements.size > 0)
+  if(GetDriverVersion().Vendor() == GPUVendor::AMD && pMemoryRequirements->memoryRequirements.size > 0)
   {
     VkMemoryRequirements &memreq = pMemoryRequirements->memoryRequirements;
 
diff --git a/renderdoc/replay/renderdoc_serialise.inl b/renderdoc/replay/renderdoc_serialise.inl
index 7bb6a95bc..d1e62282f 100644
--- a/renderdoc/replay/renderdoc_serialise.inl
+++ b/renderdoc/replay/renderdoc_serialise.inl
@@ -425,6 +425,7 @@ void DoSerialise(SerialiserType &ser, APIProperties &el)
 {
   SERIALISE_MEMBER(pipelineType);
   SERIALISE_MEMBER(localRenderer);
+  SERIALISE_MEMBER(vendor);
   SERIALISE_MEMBER(degraded);
   SERIALISE_MEMBER(shadersMutable);
 
@@ -434,7 +435,7 @@ void DoSerialise(SerialiserType &ser, APIProperties &el)
   SERIALISE_MEMBER(MultiGPU);
   SERIALISE_MEMBER(D3D12Bundle);
 
-  SIZE_CHECK(16);
+  SIZE_CHECK(20);
 }
 
 template <typename SerialiserType>