--- Log opened Tue Sep 03 00:00:03 2013
15:49 < gmaxwell> amusing: people here have been chasing a bug for some users where the workaround has been to just recompile the browser and push an update.. and something non-determinstic in the build makes reported crashes go down.
15:51 < gmaxwell> Turns out there is some revisions of some AMD SOC where when the processor is in some state, and there is a branch misprediction, and then within four intructions from the mispredicted branch there is another branch.. then sometimes it resumes execution 15 bytes later than where it should have. And the build non-determinism seemed to be influencing if that 15 byte offset was a valid instruction or not.
16:00 < jgarzik> gmaxwell, makes one want to distcc each built file to N machines, and verify results match
16:52 < Luke-Jr> [20:52:22] <jorash> So, are we absolutely certain that GPUs are slower than ASIC --- ie. that devs are not missing something in the GPU miner code?
16:52 < Luke-Jr> gmaxwell: ^ lolwut
16:58 < gmaxwell> Luke-Jr: why are you talking to that guy still?
16:58 < Luke-Jr> gmaxwell: I wasn't, he just randomly spit that out :p
17:40 < midnightmagic> lol
17:42 < gmaxwell> For those not in #eligius, he continued on to demand citations for proof that GPUs could not efficiently simulate mining asics, and for proof that mining asics are not turing complete. (and then I gave up and pushed him into a volcano)
17:45 < gmaxwell> ... PM from him:
17:45 < gmaxwell> 14:42 <jorash> you just turned Skywalker into Vader.
17:45 < gmaxwell> just for the record in case my volcano dunking makes him actually go out and prove BQP is in P, so I can collect my credit for this contribution to the effort.
20:15 < midnightmagic> VA-DER va-DER! va-DERRRRR!!!
--- Log closed Wed Sep 04 00:00:06 2013