Seems 3.2.0 and 3.2.1 have a greater number of rejections of "valid" blocks.
On faster systems, this is multiplied exponentially.
EG, (No other target-block in range) Reject rate about 25-50% submitted 456K, 125K, 214K, 100K, on a 100K rate. (Takes 120K-101K, but seems to reject exact targets, and over-targets, randomly.) Reject rate on a 7K hash-rate, is nearly 75%-80%... Thus, the decaying rates of all scrypt-coins at the moment. (EG, estimated earnings 100 coins returns a real-world earning of only 20 coins. Once again, with no target blocks being the "cause of the rejection".)
This is NOT present with 3.1.0 but that program is no longer compatible with new wallets/servers. Something dramatic changed, and it rejects 100% of those blocks, if it even connects at all. EG, using 3.1.0 and a compatible wallet, this does not happen, but the older wallets are no longer accepted with the network.
This is NOT limited to scrypt mining. This is happening to sha256 also, and I assume bitcoin... thus, the value remaining steady and the "rate" not rising proportionally with the work-load, because the work-load is only producing 80-75% valids that the network is accepting. (Either they are not actually valid, but the miner thinks they are... or they are, and they are not being delivered and confirmed correctly on the wallet/server side.)
Run a test with a difficulty of 7K... Solo, and with a mini-network. (Using compiled windows programs, as I can not confirm these results with linux, where they might not exist. Present on Win7-64bit and Vista-32bit. Same results.)
CGminer 3.1.0 and lower works on wallets versions 6.3 and lower. CGminer 3.2.x only works on wallets 6.4 and higher. (It has been a scramble to upgrade all the networks, and now they are all failing due to this issue. Falling difficulties yielding lower returns than higher difficulties, which results in even lower difficulties and even higher losses. Since even less coins are mined with each lower difficulty, not more, as expected.)
Hmm. Not sure I understand all this. But I can say I found using 3.1.1 against Bitminter gives me more rejects directly than if I funnel them through the stratum proxy. I prefer using the proxy because it kicks in the variable difficulty and only one machine is connecting to the pool instead of all 4, reducing my (and pool ops) bandwidth usage.
Getting 4 difficulty thru the proxy, I get hardly any rejects. Whereas with 1 difficulty with each miner going there individually, I do see more. I can't quantify it, and it might just be because there are more diff 1 shares submitted than diff 4 shares, therefore I'll see more rejects.
Just my $.02 worth.
M
