I am running:
- Windows 10
- Nvidia GTX1070-8GB
- 24GB System RAM
- i7
I am running into an issue in my video editor, complaining about depleting the GPU memory.
When inspecting my resources, I noted that Windows allocated "shared GPU memory" to my card.
In my case, 12GB is added to "total" my reported GPU memory to 20GB. (8GB Nvidia + 12gb Shared)
For the life of me, I can't find any setting to change this. Not in the Nvidia driver app, not in windows and not in the BIOS.
Does anyone know how to turn off this feature? I suspect it may be influencing my app complaining about video memory, as when it crashes and reports the GPU out of memory error, task manager shows physical GPU memory at under 6GB (of 8gb total).
How the heck do I get rid of this "shared GPU memory" so I can test my theory?
This is what my resources look like while my video editor is doing its thing… Windows is definitely allocating and using "shared video memory"
My actual error details:
App: Davinci Resolve Studio 16.2
Debug Log:
[0x000019c0] | GPUManager | INFO | 2020-04-28 20:34:00,244
| Flushing GPU memory…[0x000019c0] | DVIP | ERROR | 2020-04-28 20:34:00,815
| cudaMallocArray failed: Requested size 126.6 MiB (7680×4320-1
datatypesize 4)[0x000019c0] | DVIP | ERROR | 2020-04-28 20:34:00,815
| DeviceResourceTypeCUDA.cpp:177: CUDA error cudaErrorMemoryAllocation
(2)[0x000019c0] | DVIP | ERROR | 2020-04-28 20:34:00,816
| Cuda Memory Status : free 62.0 MiB total 8192.0 MiB[0x000019c0] | DVIP | ERROR | 2020-04-28 20:34:00,822
| cudaMallocArray failed: Requested size 126.6 MiB (7680×4320-1
datatypesize 4)[0x000019c0] | DVIP | ERROR | 2020-04-28 20:34:00,822
| DeviceResourceTypeCUDA.cpp:177: CUDA error cudaErrorMemoryAllocation
(2)[0x000019c0] | DVIP | ERROR | 2020-04-28 20:34:00,822
| Cuda Memory Status : free 62.0 MiB total 8192.0 MiB[0x000019c0] | GPUManager | ERROR | 2020-04-28 20:34:00,828
| Exception caught while running GPU algorithms:[0x000019c0] | DVIP | ERROR | 2020-04-28 20:34:00,828
| DVIP Exception: Position DeviceResourceTypeCUDA.cpp:177
– API: CUDA
– API Error Code: cudaErrorMemoryAllocation (2)
Related issues from other people.
Google Search
My Resources when the App runs.
Sometimes my Shared video memory is utilized more than my physical video memory. 2nd pic
Best Answer
There is no way to do that normally. There may be some hacks specific to your video card, like editing the registry, however its not recommended. The RAM memory is only used if the graphics cards GDDR becomes too full.
Its not likely that your problem comes from the shared memory.