as the tittle says, this is going to be an issue for mining.
I have 3 RX cards:
Main PCI-E x16 - RX 480 4GB.
2nd PCI-E x16 - RX 580 4GB.
3rd PCI-E x16 - RX 470 4GB.
CPU is AMD FX-8320, so no iGPU is available.
According to GPU-Z, the 2nd and 3rd have 8MB Dedicated and 1MB Dynamical memory used.
The Main GPU uses ~110MB Dedicated and ~50MB Dynamical.
According to my memories, either the Drivers or the OS management was RESERVING about ~500MB's of VRAM for whatever it needs.
That leaves about 3.4 GB usable for the main GPU, and ~3.6GB for the rest.
Used algorithm is Crypronight - XMR,ETN, Graft, etc.
Compute mode is ON for all GPUs.
Used miner and configuration is:
raw intensity: 896
Using the above mentioned configuration, starting the miner for GPU's 2 and 3, uses 3.6GB of VRAM. No issues, it mines perfectly.
Using the same configuration on the Main GPU, causes instant crash of drivers OR system hang, and i need to hard reset.
Roughly calculated, 1 unit of raw intensity is ~2MB big. Thats how the algo works, it uses 2MB of L3 cache on the CPU's. Same principle.
This makes it : RawIntensity * 2MB * Threads -> 896 * 2 * 2 = 3584MB of used VRAM.
I have tried reducing the Raw Intensity, accordingly to the already used VRAM, with a total of 64, but it still causes issues. And it shouldn't do it anymore. But it does.
Seems like Windows allocates a lot of resources to the main GPU.
With 800 RI it works, but its beyond useful, its better in graphics mode.
Any idea, on how to prevent Windows for allocating resources on the GPU - limit to use X amounts of VRAM? I highly doubt its possible, but still....
Also, why the system crashes if too much VRAM is allocated... Instead of a driver crash?