I have some code that’s been running reliably for a long while on an 8800GTX, but intermittently returns “unspecified launch failure” on my new GTX 260. The failure does not always occur at the same place, which indicates to me some sort of race condition or memory management issue.
Although I moved from compute capability 1.0 to 1.3, I thought that it would run everything in order (no asynchronous behavior), since the old code never specifies any streams or any asynchronous memory operations. Are there any calls that could have implicitly become asynchronous? Alternatively, are there other changes that could be causing this error?
Thanks for any hints about where to look!