It seems, that if the system is not present with enough RAM, even it has a lot of Virtual memory allocated, it will fail to load the binaries. Then the GPU thread would be declared sick.
Example:
The system has 4 AMD GPU's. System RAM is 4GB. System allocated virtual memory is 16GB.
sgminer is set to run with 1 GPU. Using the taskamaner, it can be seen, that the binaries are loaded in the RAM, but once it runs out of RAM, it should drop it to virtual memory - it does so (according to the disk read/write). But the GPU threads wont start. After 2 minutes, they are declared SICK.
Doing the same thing, with 4 cards, causes the system to become unresponsive, since its trying to right away put everything into the RAM.
Tested with the MinGW 5.5.6 build on Windows 10 1709 x64. Algorithm is Cryptonight v7.
It seems, that if the system is not present with enough RAM, even it has a lot of Virtual memory allocated, it will fail to load the binaries. Then the GPU thread would be declared sick.
Example:
The system has 4 AMD GPU's. System RAM is 4GB. System allocated virtual memory is 16GB. sgminer is set to run with 1 GPU. Using the taskamaner, it can be seen, that the binaries are loaded in the RAM, but once it runs out of RAM, it should drop it to virtual memory - it does so (according to the disk read/write). But the GPU threads wont start. After 2 minutes, they are declared SICK. Doing the same thing, with 4 cards, causes the system to become unresponsive, since its trying to right away put everything into the RAM.
Tested with the MinGW 5.5.6 build on Windows 10 1709 x64. Algorithm is Cryptonight v7.