Cgminer-neoscrypt - Testing / optimisation / Windows 7, R9 270, Catalyst 13.12
-
After reinstalling 13.12 and then upgrading - got cgminer working.
cgminer 3.7.7b, R9 270, AMD 14.9.1 beta - windows 7
I got 45KHash with 3.12 and upgraded to 14.9.1 beta and get 50 KHash
That doesn’t seem too bad as R9 270 = 1/3 of a R9 290 and is giving half the hash.
I’ll try vanilla 14.9.1 tomorrow - (A long and tortuous operation on windows) and test that wasn’t glitchy over night.
It only lasted 6 hours and lost contact with the GPU, I’ve turned it down to I - 13 giving ~ 45kHash/s. and left it on AMD 14.9.1. beta to continue testing why it failed. + raised a Github issue on https://github.com/vehre/neo-gpuminer
I’ve forked that project to have a look at some interface changes I thought might improve it, and if there were any enhancements, like better logs, and longer term analysis of performance, port some of 3.8 ? + extra tests, does it still work with scrypt / other algorithms, integration into a merchant solution / minerX. etc…
I’ll ( I suggest) https://github.com/vehre/neo-gpuminerest) raise a [Dev] cgminer-neoscrypt - thread later for further enhancements.
-
I’ve turned it down to Intensity 13 and left the windows 7 system on AMD 14.9.1. beta driver.
To continue testing why 3.7.7b failed. + updated the issue at Github issue on https://github.com/vehre/neo-gpuminer
It lasted 19.5 hours and terminal shut down.
To me this still indicates a problem, I would like to investigate, if it is the 14.9.1 beta AMD driver first.
That is, it would be good to “debug” the gpuminer get more information when it stops. It could then be a simple problem that is causing the glitch occasionally, so worth persueing
I’m a bit over run / too much other stuff to sort out - to do that (dig more into cgminer debug options) , will set up if anyone has more knowledge of gpuminer / do some research to find if there is a debug switch / compile a debug version…
-
It only lasted 6 hours and lost contact with the GPU, I’ve turned it down to I - 13 giving ~ 45kHash/s. and left it on AMD 14.9.1. beta to continue testing why it failed. + raised a Github issue on https://github.com/vehre/neo-gpuminer
I’ve forked that project to have a look at some interface changes I thought might improve it, and if there were any enhancements, like better logs, and longer term analysis of performance, port some of 3.8 ? + extra tests, does it still work with scrypt / other algorithms, integration into a merchant solution / minerX. etc…
I’ll ( I suggest) https://github.com/vehre/neo-gpuminerest) raise a [Dev] cgminer-neoscrypt - thread later for further enhancements.
It’s not the driver - I compiled my bin and ran on 14.9.1 beta and it works fine - the code is seriously half-broken.
-
It’s not the driver - I compiled my bin and ran on 14.9.1 beta and it works fine - the code is seriously half-broken.
Its a working GPU miner tho which is a million miles better than where we were.
-
Its a working GPU miner tho which is a million miles better than where we were.
Okay, fair.
-
Okay, fair.
That’s not to say that we would not appreciate any help with fixes,optimisation or enhancements. :)
-
That’s not to say that we would not appreciate any help with fixes,optimisation or enhancements. :)
I cleaned out a lot of the stupid in the CL file, but I need to find the change that triggers the 14.9 bug so I can release that.
-
I cleaned out a lot of the stupid in the CL file, but I need to find the change that triggers the 14.9 bug so I can release that.
Good to hear, that the application is improved. :)
Are you using github, so you can create pull requests to Avehre’s version of cgminer?
-
Good to hear, that the application is improved. :)
Are you using github, so you can create pull requests to Avehre’s version of cgminer?
No, but as I said, I’ll publish the code when I find that bit.
-
Ugh, shiiiiit. It’s a combination of things. Basically the block mix operation as well as how it’s called. Now I have to water down my salsa. (pun intended)
-
Ugh, shiiiiit. It’s a combination of things. Basically the block mix operation as well as how it’s called. Now I have to water down my salsa. (pun intended)
Shit works on 14.9: https://github.com/wolf9466/neo-gpuminer
I fixed just enough of that disaster to get it to work.
-
nice work, Wolf, linked to watch the github. “optimisation mode” enabled.
