[HOWTO] Building cgminer for the Xeon Phi

[mscman]

Unfortunately can't post this in any of the boards except this one... but for those interested in trying to get a miner running on the Xeon Phi:

1) Download the opencl SDK from http://software.intel.com/en-us/vcsource/tools/opencl-sdk-xe

2) Download the cgminer source from GitHub: https://github.com/ckolivas/cgminer

3) Edit ocl.c and change all references of CL_DEVICE_TYPE_GPU to CL_DEVICE_TYPE_ACCELERATOR

There should be 5 of them. Without this change, it will build and try to run (will even show the devices found in debug output) but won't be able to get the device ids and will bail.

Performance with 4 vectors and a work of around 32 is pretty poor, around 20GH/s per card. Perhaps someone with more time and better knowledge of optimizing for the Xeon Phi can optimize this better and give back to the cgminer project. I'll try to dig into it more eventually as time permits.

[fruor]

very well done. i hope others will find this as well.

although it seems pretty uninteresting for btc mining, there will be more algorithms coming up this year where lots of memory will be of good use (I saw 16gb versions of this card). so don't throw it away ;-)

did you try to run cgminer in scrypt mode? how many khash do you get out of this without overclocking?