01-04-2013, 07:53 PM
I have just finished running some tests. First I did as you said, and used -n 1. Took a very long to complete but no issues. Then I set -n to 160. Few minutes and it froze. I set it to 80. Again, it froze. I set it to 40 and is OK some of the time depending on the dictionaries I am using.
And then it hit me. Every time there has been an issue, I have seen the message: "NOTE: autotuned --gpu-accel from <some value> to <some value> ". I have seen it autotuned to anywhere from 8 to 56. So every time I set -n to a higher value than the autotuned value, it will eventually freeze.
This means if I am running an automated script, I could either set -n to 40 and avoid the issue some of the time or set it to 8 and avoid it completely.
So I am kind of curious, what's the purpose of autotune if you still need to worry about it. I thought it was there to prevent you from setting it too high in the first place.
Although, it's not really a big issue, just a bit frustrating, for example, when you have been running it for several hours with no results and you would like to bypass. Anyway, I will keep doing a few more tests.
And then it hit me. Every time there has been an issue, I have seen the message: "NOTE: autotuned --gpu-accel from <some value> to <some value> ". I have seen it autotuned to anywhere from 8 to 56. So every time I set -n to a higher value than the autotuned value, it will eventually freeze.
This means if I am running an automated script, I could either set -n to 40 and avoid the issue some of the time or set it to 8 and avoid it completely.
So I am kind of curious, what's the purpose of autotune if you still need to worry about it. I thought it was there to prevent you from setting it too high in the first place.
Although, it's not really a big issue, just a bit frustrating, for example, when you have been running it for several hours with no results and you would like to bypass. Anyway, I will keep doing a few more tests.