Change Details

**System Information** Operating system: Windows-10-10.0.19041-SP0 64 Bits Graphics card: GeForce RTX 3090/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 456.38 **Blender Version** Broken: - 2.92.0 Beta, branch: master, commit date: 2021-01-14 20:13, hash: `rB30dd31a7b3d8` - 2.92.0 Alpha, branch: master, commit date: 2020-12-15 23:20, hash: `rB977bd7937a07` Worked: - 2.92.0 Alpha, branch: master, commit date: 2020-11-19 22:15, hash: `rBfdd3032f8fad` - Reports "System is out of GPU and shared host memory" **Short description of error** Denoising Data with large tiles causes instant crash and exit. **Exact steps for others to reproduce the error** # Open attached file # Ensure that your system preferences are configured for the use of a GPU # Render {F9580376} Alternatively from the default startup file: 1. Start Blender. 2. Change render engine to Cycles. 3. Device to GPU. 4. Go to "Passes->Data" and enable "Denoising Data". 5. In Performance->Tiles option change both for Tile X and Tiles Y to 2000. 6. Change scene Dimensions both for Resolution X and Resolution Y to 2000. 7. Click "F12" for rendering. I am working with large tiles for performance reasons my GPU 3090x and CPU with 64gb has plenty of memory. Notes: 1. Turning on the render "OpenImageDenoise" no problem. 2. Enabling all other data passes no problem. 3. It seems related to the render tile prefiltered. ```lines std::_Check_C_return(int _Res) Line 131 C++ std::unique_lock<std::mutex>::unique_lock<std::mutex>(std::mutex & _Mtx) Line 138 C++ ccl::CUDADevice::global_free(ccl::device_memory & mem) Line 1111 C++ ccl::CUDADevice::mem_copy_to(ccl::device_memory & mem) Line 1005 C++ ccl::CUDADevice::move_textures_to_host(unsigned __int64 size, bool for_texture) Line 798 C++ ccl::CUDADevice::generic_alloc(ccl::device_memory & mem, unsigned __int64 pitch_padding) Line 840 C++ ccl::CUDADevice::mem_alloc(ccl::device_memory & mem) Line 994 C++ ccl::device_only_memory<float>::alloc_to_device(unsigned __int64 num, bool shrink_to_fit) Line 300 C++ ccl::DenoisingTask::run_denoising(ccl::RenderTile & tile) Line 332 C++ ccl::CUDADevice::thread_run(ccl::DeviceTask & task) Line 2410 C++ ccl::CUDADevice::task_add::__l11::<Lambdafunktion>() Line 2465 C++ ccl::DedicatedTaskPool::thread_run() Line 222 C++ [External Code] ccl::thread::run(void * arg) Line 53 C++ [External Code] ```

**System Information** Operating system: Windows-10-10.0.19041-SP0 64 Bits Graphics card: GeForce RTX 3090/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 456.38 **Blender Version** Broken: - 2.92.0 Beta, branch: master, commit date: 2021-01-14 20:13, hash: `rB30dd31a7b3d8` - 2.92.0 Alpha, branch: master, commit date: 2020-12-15 23:20, hash: `rB977bd7937a07` - Commit rBbfb6fce6594e9cf133bd18aee311c1e5e32dc799 identified in T85089 as the one that introduced the issue Worked: - 2.92.0 Alpha, branch: master, commit date: 2020-11-19 22:15, hash: `rBfdd3032f8fad` - Reports "System is out of GPU and shared host memory" **Short description of error** Denoising Data with large tiles causes instant crash and exit. From merged ticket T85089: Crashes while loading textures to GPU memory using CUDA. After some digging in the code it seems to happen when the device memory is low and it decides to 'move_textures_to_host' then it ends up trying to lock 'cuda_mem_map_mutex' twice. {F9599185} **Exact steps for others to reproduce the error** # Open attached file # Ensure that your system preferences are configured for the use of a GPU # Render {F9580376} Alternatively from the default startup file: 1. Start Blender. 2. Change render engine to Cycles. 3. Device to GPU. 4. Go to "Passes->Data" and enable "Denoising Data". 5. In Performance->Tiles option change both for Tile X and Tiles Y to 2000. 6. Change scene Dimensions both for Resolution X and Resolution Y to 2000. 7. Click "F12" for rendering. I am working with large tiles for performance reasons my GPU 3090x and CPU with 64gb has plenty of memory. Notes: 1. Turning on the render "OpenImageDenoise" no problem. 2. Enabling all other data passes no problem. 3. It seems related to the render tile prefiltered. ```lines std::_Check_C_return(int _Res) Line 131 C++ std::unique_lock<std::mutex>::unique_lock<std::mutex>(std::mutex & _Mtx) Line 138 C++ ccl::CUDADevice::global_free(ccl::device_memory & mem) Line 1111 C++ ccl::CUDADevice::mem_copy_to(ccl::device_memory & mem) Line 1005 C++ ccl::CUDADevice::move_textures_to_host(unsigned __int64 size, bool for_texture) Line 798 C++ ccl::CUDADevice::generic_alloc(ccl::device_memory & mem, unsigned __int64 pitch_padding) Line 840 C++ ccl::CUDADevice::mem_alloc(ccl::device_memory & mem) Line 994 C++ ccl::device_only_memory<float>::alloc_to_device(unsigned __int64 num, bool shrink_to_fit) Line 300 C++ ccl::DenoisingTask::run_denoising(ccl::RenderTile & tile) Line 332 C++ ccl::CUDADevice::thread_run(ccl::DeviceTask & task) Line 2410 C++ ccl::CUDADevice::task_add::__l11::<Lambdafunktion>() Line 2465 C++ ccl::DedicatedTaskPool::thread_run() Line 222 C++ [External Code] ccl::thread::run(void * arg) Line 53 C++ [External Code] ``` ----- From merged ticket T85089: Select CUDA or OptiX as the device Render a scene with large textures {F9599178}

**System Information** Operating system: Windows-10-10.0.19041-SP0 64 Bits Graphics card: GeForce RTX 3090/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 456.38 **Blender Version** Broken: - 2.92.0 Beta, branch: master, commit date: 2021-01-14 20:13, hash: `rB30dd31a7b3d8` - 2.92.0 Alpha, branch: master, commit date: 2020-12-15 23:20, hash: `rB977bd7937a07` - Commit rBbfb6fce6594e9cf133bd18aee311c1e5e32dc799 identified in T85089 as the one that introduced the issue Worked: - 2.92.0 Alpha, branch: master, commit date: 2020-11-19 22:15, hash: `rBfdd3032f8fad` - Reports "System is out of GPU and shared host memory" **Short description of error** Denoising Data with large tiles causes instant crash and exit. From merged ticket T85089: Crashes while loading textures to GPU memory using CUDA. After some digging in the code it seems to happen when the device memory is low and it decides to 'move_textures_to_host' then it ends up trying to lock 'cuda_mem_map_mutex' twice. {F9599185} **Exact steps for others to reproduce the error** # Open attached file # Ensure that your system preferences are configured for the use of a GPU # Render {F9580376} Alternatively from the default startup file: 1. Start Blender. 2. Change render engine to Cycles. 3. Device to GPU. 4. Go to "Passes->Data" and enable "Denoising Data". 5. In Performance->Tiles option change both for Tile X and Tiles Y to 2000. 6. Change scene Dimensions both for Resolution X and Resolution Y to 2000. 7. Click "F12" for rendering. I am working with large tiles for performance reasons my GPU 3090x and CPU with 64gb has plenty of memory. Notes: 1. Turning on the render "OpenImageDenoise" no problem. 2. Enabling all other data passes no problem. 3. It seems related to the render tile prefiltered. ```lines std::_Check_C_return(int _Res) Line 131 C++ std::unique_lock<std::mutex>::unique_lock<std::mutex>(std::mutex & _Mtx) Line 138 C++ ccl::CUDADevice::global_free(ccl::device_memory & mem) Line 1111 C++ ccl::CUDADevice::mem_copy_to(ccl::device_memory & mem) Line 1005 C++ ccl::CUDADevice::move_textures_to_host(unsigned __int64 size, bool for_texture) Line 798 C++ ccl::CUDADevice::generic_alloc(ccl::device_memory & mem, unsigned __int64 pitch_padding) Line 840 C++ ccl::CUDADevice::mem_alloc(ccl::device_memory & mem) Line 994 C++ ccl::device_only_memory<float>::alloc_to_device(unsigned __int64 num, bool shrink_to_fit) Line 300 C++ ccl::DenoisingTask::run_denoising(ccl::RenderTile & tile) Line 332 C++ ccl::CUDADevice::thread_run(ccl::DeviceTask & task) Line 2410 C++ ccl::CUDADevice::task_add::__l11::<Lambdafunktion>() Line 2465 C++ ccl::DedicatedTaskPool::thread_run() Line 222 C++ [External Code] ccl::thread::run(void * arg) Line 53 C++ [External Code] ``` ----- From merged ticket T85089: Select CUDA or OptiX as the device Render a scene with large textures {F9599178}