Awesome news!
I hope its gonna be faster than old optiminer
Depends a bit ... it closes the gap between 4G and 8G cards. Atm my code uses 3.6G and does on average 12.7 sol/s on a RX 580 4G (headless) - the 8G model will only be barely faster, maybe 13 sol/s. I think this is not so bad compared to optiminer ^^
For me only equihash96 is working on NVidia cards under Linux.
For equihash144 I get this msg:
lolMiner is configured to mine Bitcoin Gold(BTG), Equihash 144.5
Setup Miner...
Using device with id 0 (GeForce GTX 1060 3GB)
Using device with id 1 (GeForce GTX 1060 3GB)
Using device with id 2 (GeForce GTX 1060 3GB)
Using device with id 3 (GeForce GTX 1060 3GB)
Using device with id 4 (GeForce GTX 1060 3GB)
Using device with id 5 (GeForce GTX 1060 3GB)
Warning: Your device with id 0 is currently not supported by lolMiner or by the specific algorithm you selected.
Warning: Your device with id 1 is currently not supported by lolMiner or by the specific algorithm you selected.
Warning: Your device with id 2 is currently not supported by lolMiner or by the specific algorithm you selected.
Warning: Your device with id 3 is currently not supported by lolMiner or by the specific algorithm you selected.
Warning: Your device with id 4 is currently not supported by lolMiner or by the specific algorithm you selected.
Warning: Your device with id 5 is currently not supported by lolMiner or by the specific algorithm you selected.
as far as i know, you cant use
3gb cards on that specific algo on this specific miner
(4gb+)
so its working as intended technically
Well we have to distinguish here between memory available and the memory that the driver allows us to use in a single buffer. For some reason the Nvidia OpenCL driver limits OpenCL that a single memory region may be at most 1/4th of the total memory in size. That said: the 144.5 miner code uses 2.8G, but requires that two buffers of 1.1G are created (and a 3rd with the missing 600 mb) ... so what happens here:
The check for enough total memory passes, but unfortunately the one for maximal single allocation fails.
Interestingly I just found out that CUDA has no such limitation and also sometimes it seems possible to allocate more then "CL_DEVICE_MAX_MEM_ALLOC_SIZE" tells on Nvidia (but not on AMD ... there it is a hard limit, but its often much higher then on Nvidia) ... so maybe I can do an option to ignore memory check on own risk ... hmm