Open Techn1cks opened 6 years ago
Tried an experiment, ran 2x sgminer with --device "0-4" --remove-disabled and --device "5-9" --remove-disabled respectively and the results speak for themselves:
Running all 10 gpus on 1 sgminer: 8.13Kh/s (after ~18 hours, but it stabilized to that in about 1 hour and kept constant ever since) sgminer 1: Device 0-4: 4.27Kh/s (after 1hr 30 mins) sgminer 2: Device 5-9: 4.33Kh/s (after 1hr 30 mins) Total: 8.60Kh/s ~5.5% slower when running all 10 GPUs in one instance
Now i'm curious what will happen if i run 10 sgminer instances.... on to the next adventure :)
Supper expiriens! :) Try to switch on debug information and look time to set work to GPU(s). May be problem in bus or irq balanses PCIE. Check IOps.
Tested as follows:
I have 6 hynix cards, 4 samsung cards. Hynix alone would do with 2 threads around 840h/s, samsung does about 920h/s. Using 7 GPUs, i was getting the same results as shown above (4 samsung 3 hynix) - for a total of ~6.2Kh/s. Adding the last 3 hynix cards would yield no more than 8.1Kh/s. It should be more around 8.72Kh/s. This is a loss of over 610 h/s, or about 7.6% loss, accounting to almost a whole video card missing!
Any clue how to fix this bug? Are there some settings that i'm not aware of that can help or is this really a bug?
Attached image shows the hash rates, the ones over 800 are the samsungs, below 800 are the hynix.