The pools report based on your U. So if you're unlucky for a while, your U goes down, and your pool reports it slower. Doesn't mean you're actually hashing slower for that period.

Nothing particularly nice  

I'll use GPUs while they're profitable. If, when the block reward drops, the GPUs stop being profitable, you can ALWAYS sell a GPU (and I was being really conservative saying they'll sell for 50% of the price). Investing in FPGAs on the other hand, despite what anyone says, you are 100% betting on bitcoin being a sure thing long term since there is NO short term profit and NO guaranteed sale of the hardware should bitcoin tank. The return time is absolutely huge. All the so-called other uses for 2nd hand FPGAs are very few and far between, and there is hardly any market for a whole lot of 2nd hand FPGAs unlike GPUs. Difficulty slowly lags behind and parallels BTC value, not the other way around, no matter how much people want it to be, so difficulty will drop somewhat, but perhaps stay above the GPU mining profitability point. If the only way to mine profitably after the drop is with FPGAs, but only with a minimum turnaround of something like 1 year (factoring in reward drop), you can sell all the GPUs then and decide whether you want to bet 100% on bitcoin then and move to FPGAs or jump ship entirely.

In investment, risk should parallel likely gain. The risks of FPGAs are much higher than GPUs, but the potential gains, are they there? Unless bitcoin explodes in popularity and price jumps up dramatically, and you have been hoarding them instead of selling them along the way, they're not. You have to be honest to yourself about the risk here when you're talking potentially tens of thousands of dollars, especially if it's not money you have to burn. Don't get wrapped up in the momentum and not see that it's just another risky venture.

You ARE out of work after a longpoll. The cached work is absolutely useless belonging to a now dead previous block height so you do NOT work on it no matter how much work is "cached". The only thing that is "cached" with submit-stale is SHARES, if they have been found at approximately the same point in time as the longpoll, not extra work. You cannot work again at maximum rate until you have received enough work from either the longpoll itself (since it provides one work unit), rolls work from that longpoll (if the pool supports rolltime), and/or you get enough new work units. The fact that the BFL device twiddles its thumbs for up to 5 seconds on the wrong block work after longpoll means up to 5 seconds of work is wasted. You can submit those shares, but most pools will just say "too late" and they'll be rejected.

The official binary most definitely is compiled with bitforce support.

Interesting.  Is this because each client is sending more LP requests, or is it just because anyone with EMC as a backup pool is now requesting a longpoll regardless of whether they're currently connected or not?

Slightly different. The first number on per-device is updated differently and is more a current average rather than over any particular time period. Even the global 5s value is not truly a 5s value, it just decays over 5 seconds.

Glad you appreciate it. I think it's one of the most underrated features myself too since most miners tend to just use dedicated rigs. Just remember you can also tune the intensity of the dynamic mode as well with the --gpu-dyninterval option.

Nonsense. 7970s will be still hot property in December. The next generation of GPUs is miles away and that's what usually causes the price drop. Your up front price for FPGAs is still a rip off at the moment. Come the block reward halving, the goal posts all change, but so will difficulty as the inefficient miners drop off. There is no foregone conclusion here, only a business model as far as I can see. Again if I sell my 7970s for even half their price, the current up front price/return on investment time for FPGAs makes no sense at all - to me. If electricity is ludicrously expensive, things might be different as Eleuthria suffers.

That's not possible in a cross platform way. It's not even possible to cleanly close cgminer once a driver has crashed (dead device). You need to use the API or log to monitor it happening and use an external program to do that.

If the rate is very low, then leaving it is fine. However 2.5% is fairly significant. Think of them like the glitches you see on screen when you overclock too much. I'd recommend just come backing a little on the overclock till they go away.

I'm sorry, clearly I don't know what I'm talking about.

6870 Northern Islands has 1120 stream processors of the same architecture as the 5870 which has 1600. Performance is linearly related to architecture * stream processors * Mhz. Therefore at exact same clocks, the 6870 should be (1120/1600) 0.7 * Mhash of a 5870. At 1000Mhz, the 5870 gets at best ~460Mhash. Therefore a number as high as 375 for a 6870 is ... highly unlikely.

Does this mean cgminer disabled the pool for you automatically and then correctly re-enabled it? I need some in-the-field confirmation this code is working.

Sweet, thanks 

That easyminer thing screams of the kind of software lock in proprietary crap that is absolutely nothing like bitcoin, and doesn't even allow you to mine anywhere but one pool even. It's really disappointing and I hope people use something else more open... anything, it doesn't even need to be cgminer. I'd much rather whatever benefits were there in the software could be distributed out to other mining software, or they would work with the existing software developers instead of that ancient corporate proprietary model which has been proven time and again to be based on FUD and long-term ends up being a burden rather than a boon.

Only intermittently as a special offer. Reliably $450.

Most overclocked 7970s will do 700MH, but the power cost consumption rises disproportionately. I've measured about 250W at that speed.

This is why I stopped supporting cpu mining on cgminer above all other reasons.

You still can. Nothing's changed. If you can't get temps usually you haven't export DISPLAY on linux or are running some  remote desktop on windows.

7970: Currently $405USD, ~700Mhash, 250W, warranty 1+ year from many reliable manufacturers available instantly, resale >$200 if sold in 2 years. Also play games and act as GPUs... Return on investment time: Dependent on your local electricity prices. Block reward halving in 7 months...

Fill in the gaps for other devices.

Note: I only have GPUs.

It's as kano said, that extra comma is an error to do with you not running any GPUs. Will need to be fixed in the next version. You'll have to edit it manually for it to work at the moment.