Try again ... it works ... I wrote it
PWC solo & 5750 (about 80 kHash), Last Valid Work is 1751 seconds old just now and still growing. I already noticed that this value was higher than 2000 seconds. On the other side, faster card (7950) and different scrypt coin has lower values but I see 244 seconds timeout just now... These values are related to coin and GPU hashrate.
The time will of course relate to the speed of the device.
Until you find a nonce, the device has done nothing successfully.
How long on average ... well that of course will directly relate to the hash rate and how many hashes are required on average per nonce.
For BTC sha256 you need to do on average 2^32 hashes per nonce - so for 100MH/s that's 43seconds for 2^32 hashes.
21 times that is 15 minutes - so yeah if a 100MH/s device finds no valid shares in 15 minutes then you can be pretty sure something is wrong.
Until a device actually returns a valid nonce, you can't be sure it really is working properly.
I guess on scrypt you have a problem coz it may take
months minutes to do any work, so I guess you can't really be sure there is something wrong very quickly. However, in your case you went way, way past the point where "Last Valid Work" would definitely mean something is wrong.
I will also add that with scrypt, if you try to push the performance up too much, you could easily get exactly what you said, where the device might keep returning invalid nonces but still appear to be OK from cgminer's point of view - send work, get reply.
The problem there of course is that you'd have to keep a history of HW errors since if it works OK for 3 days and then starts spitting out constant HW errors due to whatever is the cause, the HW error rate won't increase very fast.
I've never bother to scrypt GPU mine ... I prefer to have my GPUs idle or playing Minecraft for my kids - they use so much energy mining ...
