I have an RTX Titan, don’t get @ me, and I was just wondering if anyone thought it would be possible to run two instances of grin miner on it to fully utilize the memory capacity? Naively, I tried simply running another instance from a different directory (just copied the grin-miner directory I already had into it), and it worked, but all it seemed to do was divide the GPS between the two instances, that is to say, it slowed down the GPS for one instance. In any case, merely curious if anyone had any insight into whether or not this could be possible, why or why not?
I tried something similar with two AR29 on one 2080ti, with the GPS split between the two.
My understanding is that while the algorithm requires some minimum vRAM (11GB in the case of AT31), its speed is actually determined by the number of SM/cores in the GPU. So with double the memory you are able to run two miners in parallel, but each one gets half the processors and half the gps.
You can also just copy the same config (keeping the same device ID):
[[mining.miner_plugin_config]] plugin_name = "cuckaroo_cuda_29" [mining.miner_plugin_config.parameters] device = 0 ... [[mining.miner_plugin_config]] plugin_name = "cuckaroo_cuda_29" [mining.miner_plugin_config.parameters] device = 0 ...
It’s true that the two miners fight for GPU cores and memory througput, but this should slightly increase utilization. With a single instance there are times when some cores wait for stragglers to finish. These gaps should be filled when running two instances. The sum of two GPS values should thus be slightly above 1 instance. But you won’t see 2x, that’s for sure.
On the other hand latency increases so this may be a net loss.
Yeah. Using a little more power, and getting about 1.83 GPS vs 1.79 on the higher end average, while using 21.7 GB of memory. A useful overclock I guess.
since day 3 or so after mainnet launch, i’ve been mining C31 on 4GB AMD Cards (ocl-cuckatoo31),
i’ve been mining with two instances, is does increase by 15% total GraphRate… BUT, i see a lot of flutuation on the poolside… ironically, the miner shows a more stable total GraphRate… with a delta of 2-3%… whiles with one instance I have 5-10% delta on the total…
i did try with C29, bu ran out of memory…