GPU Mining on OS X Lion (10.7)

[twmz]

All of my observations are from a Mac Pro with an ATI HD 5870.  Some people have reported that they have none of the problems I am seeing when using miners on iMacs with ATI mobile video cards, so your milage may vary

I have tried several miners and not had success with any:

• Current version of DiabloMiner won't start.  Note, the current version of DiabloMiner didn't work with Snow Leopard either, but had different symptoms.
• Current version of poclbm uses the phatk kernel which still seems to not work (status shows incredibly high hashrates for a single GPU which likely means it is erroring out instead of finding real hashes).  The current version of poclbm didn't work with SL either.
• I tried an old version of poclbm (before the switch to phatk) that had worked well on SL, but on Lion, it generates mostly invalid/stale shares (> 80%).  Unsure what is causing those.

So, has anyone had any success GPU mining on Lion, with its new version of OpenCL?  If so, what software and setup did you use?

[1714039085]

1714039085

[1714039085]

1714039085

[1714039085]

1714039085

[twmz]

Update:  I tried running the old (non-phatk) poclbm with -f 100 and it has been running for the last 30+ minutes without any rejected shares.  So, possibly the default poclbm in that version is just too agressive and that was causing some corruption?

On a side, note, my workstation's HD 5870 is now getting about 285 MH/s (with -f 100) under Lion vs about 192 MH/s (with -f 10) under Snow Leopard.  Still not quite what that card would get under Linux, but much improved.  Given that this workstation has a day job, I am trilled to get almost 50% more hashing out of it while it is idle (this workstation only mines when the screensaver is on).

Edit: and right after I posted this, I got 4 corrupt shares in a row.  Well, I'll keep an eye on it and see how frequent they are and possibly will try an even higher -f setting.

[twmz]

These are the miners I found to work:

* old poclbm (pre-phatk, using -f 100 or higher)
* cgminer v1.3.1 (using -v 1 and either dynamic intensity or intensity <= 6)
* current phoenix (using poclbm kernel with no kernel params)

Miners that don't work:

* current poclbm
* current DiabloMiner
* cgminer v1.4.0

[twmz]

It may be hardware specific then.  I have a Mac Pro with an HD 5870. 

In Snow Leopard the symptom was that DiabloMiner showed me getting something like 1000 GH/s (which is impossible) and never actually found shares.  D3Diablo I think posted that the problem was that the new kernel was crashing in OpenCL but that Apple wasn't reporting errors so it looked like it was just succeeding very fast.  He didn't indicate any desire to workaround or fix the issue it because it is "Apple's Problem".

In Lion, I initially got an error about work size being incorrect (see my post in the DiabloMiner thread), but that was worked around by forcing worksize of 64. I still get the B.S. hashrate though.

I happen to have saved off a very old version of DiabloMiner and it works in Lion (it's before he made the kernel optimizations that are broken with the HD 5870), but it gives a lot of messages about hardware errors.

Note, even in Snow Leopard, no miners that use a phatk-like kernel worked with my HD 5870 and they still don't work in Lion (which is why the most recent poclbm doesn't work).  So I don't think this is a Lion issue specifically.

As far as I can tell, the only Lion specific issue at this point is the DiabloMiner work size error and the fact that all of the "working" miners seem to generate corrupt shares from time to time (about 5%).  I don't know if that is a bug in the new OpenCL 1.1 or what, but given that my hashrate increased by close to 25% going from Snow Leopard to Lion, I am willing to tollerate 5% corrupt shares.

[Druas]

I'm currently running Diablo just fine on my iMac with OSX Lion.
It was working fine with OSX Snow Leopard as well.

What errors are you getting in terminal with Diablo?

Did you notice an increase in hashrate between Snow Leopard and Lion?

[jalen]

I've been getting similar problems.  After going up to lion and reinstalling pyopencl, any time I try to run POCLBM (old version) I get this as an output:


Mac-Computer:bitcoin <user>$ python poclbm.py -d 0 --user <ID> --pass <pass> --host=us.btcguild.com -w 32
Exception in thread Thread-2:
Traceback (most recent call last):
  File "/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/threading.py", line 530, in __bootstrap_inner
    self.run()
  File "/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/threading.py", line 483, in run
    self.__target(*self.__args, **self.__kwargs)
  File "/Users/<user>/bitcoin/BitcoinMiner.py", line 306, in miningThread
    output_buf)
  File "/Library/Python/2.7/site-packages/pyopencl-2011.1.2-py2.7-macosx-10.7-intel.egg/pyopencl/__init__.py", line 240, in kernel_call
    global_offset, wait_for, g_times_l=g_times_l)
LogicError: clEnqueueNDRangeKernel failed: invalid work group size


This never happened in SnowLeopard (10.6).