Updated git tree:
The longpoll message was almost always being dropped in 1.2.8 and it thought it always detected the new block itself - should be fixed in the git tree.
It was possible to queue an ever increasing amount of staged work which would eventually lead to a rise in rejects as the shares submitted would eventually be quite old. Fixed.

Not sure, but it's likely due to the high CPU requirements of the windows build (which I can't do much about since I'm led to believe it's in the pthread library in use for the windows build). I can't reproduce the problem on linux. What happens if you run two instances of cgminer, one CPU mining and one GPU mining?

New release, version 1.2.8:
Source tarball:
http://ck.kolivas.org/apps/cgminer/cgminer-1.2.8.tar.bz2

Windows build:
http://ck.kolivas.org/apps/cgminer/cgminer-1.2.8-win32.zip

Changes:
- More OSX build fixes.
- Add an sse4 algorithm to CPU mining.
- Fix CPU mining with other algorithms not working.
- Rename the poclbm file to ensure a new binary is built.
- We now are guaranteed to have one fresh work item after a block change and we
should only discard staged requests.
- Don't waste the work we retrieve from a longpoll.
- Provide a control lock around global bools to avoid racing on them.
- Iterating over 1026 nonces when confirming data from the GPU is old code
and unnecessary and can lead to repeats/stales.
- The poclbm kernel needs to be updated to work with the change to 4k sized
output buffers.
- longpoll seems to work either way with post or get but some servers prefer
get so change to httpget.


Executive summary:
Less rejects, less idle time at longpoll, less false messages, cpu mining fixed, nvidia gpu mining fixed, should build on osx.

No hashes should be lost. It doesn't work that way. It queues everything asynchronously for upload in a separate thread that has nothing to do with the hashing so everything eventually gets pushed to the server if it was found.

Efficiency 14 can take up to 15 seconds to return data from the GPU. That's a long time to not have reported back any shares. So it usually is less efficient to give such a big chunk of work to the GPU at any one time without reporting in more frequently, even though in theory it would be nice to keep the GPU as busy as possible. It's the law of diminishing returns where there doesn't seem to be any benefit beyond 8 or 9, and it starts introducing more problems.

The hashrate reported by cgminer is just how many hashes it's doing. Nothing more, nothing less. The pool is reporting an estimated hash rate based on returns. If your returns are relatively low (for whatever reason) then the pool will think you're hashing less. Now as for why you're having a bad return, I'm not really sure, but the high intensity is only doing you harm. Are you putting in a big queue? Are you running lots of threads?

Yes sorry, noted and fixed in the git tree with nvidia mining. The poclbm kernel was not updated to match the other changes and has been since fixed. Need to do a git clean -f and get the latest tree to fix it. The cpu mining issue... I dunno yet but the c algorithm seems dodgy though the assembly 64 bit one works fine here. Too much else to look at.

Because your efficiency is low. The pool reports back what it thinks your hash rate is based on the number of accepted shares you return.

Your efficiency is particularly low. That may be the result of setting the intensity too high. Drop it from 14 to 8.

The windows build has gpu mining built into it. You cannot do plain cpu mining on the generic windows build unless you build it without GPU mining support.

Ego? That's easy to say. I don't see where they're coming from and why they're lower with other miners so I don't know what to fix.

Fine, it's broken. Go use guiminer thanks.

There are instructions for the most convenient way of installing libs and includes in the README.

You build from a release tarball. Then you can go straight to ./configure
Building from git you need to apt-get install autoconf automake
Then you can ./autogen.sh fine
Don't try to start building it any other way.

Prolonged outage is meant to be longer than that. There's some race there which is occasionally hit and does it very soon after. I'll work on that.

Lots of stales? Stale shares are harmless as they never get counted, it's just the app telling you it won't submit them.
Lots of rejects? You get paid according to accepted, so judge the effectiveness on the accepted. cgminer just happens to be brutally honest.

Still using POST? Yes we've been through this, check the discussion earlier on.

Windows build:
http://ck.kolivas.org/apps/cgminer/cgminer-1.2.7-win32.zip

No it does not. It compiles one kernel once for all similar hardware and then loads the binary from that point on. Since you have to run cgminer from the source directory currently, that's where it also stores the binary file.

New version (windows build soon):
http://ck.kolivas.org/apps/cgminer/cgminer-1.2.7.tar.bz2

Changelog:
- Show last 8 characters of share submitted in log.
- Display URL connected to and user logged in as in status.
- Display current block and when it was started in the status line.
- Only pthread_join the mining threads if they exist as determined by
pthread_cancel and don't fail on pthread_cancel.
- Create a unique work queue for all getworks instead of binding it to thread 0
to avoid any conflict over thread 0's queue.
- Clean up the code to make it clear it's watchdog thread being messaged to
restart the threads.
- Check the current block description hasn't been blanked pending the real
new current block data.
- Re-enable signal handlers once the signal has been received to make it
possible to kill cgminer if it fails to shut down.
- Disable restarting of CPU mining threads pending further investigation.
- Update longpoll messages.
- Add new block data to status line.
- Fix opencl tests for osx.
- Only do local generation of work if the work item is not stale itself.
- Check for stale work within the mining threads and grab new work if
positive.
- Test for idle network conditions and prevent threads from being restarted
by the watchdog thread under those circumstances.
- Make sure that local work generation does not continue indefinitely by
stopping it after 10 minutes.
- Tweak the kernel to have a shorter path using a 4k buffer and a mask on the
nonce value instead of a compare and loop for a shorter code path.
- Allow queue of zero and make that default again now that we can track how
work is being queued versus staged. This can decrease reject rates.
- Queue precisely the number of mining threads as longpoll_staged after a
new block to not generate local work.


Sample display:

 cgminer version 1.2.6 - Started: [2011-07-16 23:28:05]
--------------------------------------------------------------------------------
 [(5s):173.9  (avg):182.0 Mh/s] [Q:36  A:37  R:0  HW:0  E:103%  U:3.00/m]
 TQ: 1  ST: 1  LS: 0  SS: 0  DW: 0  NB: 1  LW: 1  LO: 1  RF: 0  I: 3
 Connected to http://ozco.in:8332 as user ckolivas.1
 Block 0001c18ef64e55b54f0d8cbabb09ff12  started: [2011-07-16 23:35:48]
--------------------------------------------------------------------------------
 GPU 0: [182.5 Mh/s] [Q:36  A:37  R:0  HW:0  E:106%  U:3.02/m]
--------------------------------------------------------------------------------

[2011-07-16 23:28:28] Share fbbbfd7a accepted from GPU 0 thread 0
[2011-07-16 23:28:38] Share 366da5b7 accepted from GPU 0 thread 0
[2011-07-16 23:28:46] Share fce072de accepted from GPU 0 thread 1
[2011-07-16 23:29:47] Share e7918c2d accepted from GPU 0 thread 1
[2011-07-16 23:29:47] Share 3ca2631b accepted from GPU 0 thread 1
[2011-07-16 23:30:11] Share 2c8ceab5 accepted from GPU 0 thread 1
[2011-07-16 23:30:58] Share b556d2f9 accepted from GPU 0 thread 0
[2011-07-16 23:31:27] Share cc065b5a accepted from GPU 0 thread 1
[2011-07-16 23:35:46] Server not providing work fast enough, generating work locally
[2011-07-16 23:35:48] New block detected on network before longpoll, waiting on fresh work
[2011-07-16 23:35:48] Resuming with work from server
[2011-07-16 23:35:58] Share e99bc71d accepted from GPU 0 thread 1

About cpu usage:

Currently the windows build always uses a lot more CPU than the linux build. It has to do with the windows library that gives support for the linux features and I'm not sure there's anything to do about it. Beyond that, some drivers use a lot more CPU when more than one card is in use. The other big CPU user is when dynamic mode is enabled, as it constantly monitors the GPU usage and throttles it accordingly, but conversely spends more CPU time doing this. To decrease CPU usage, disable dynamic mode by choosing an intensity with -I.

About dynamic mode:

The intensity level is determined by cgminer to be that which will keep the desktop GUI fluid based on how quickly it refreshes. If you have a low power card, the intensity will almost always be low because of this. If you have a high power GPU, it may well get to quite high intensity levels even when still in use.

About the OSX build:

It should build if you have the right tools for it. Ycros is helping me out there.

About shares rejected:

When doing solo mining, you are constantly trying to find a block for yourself and anything that looks like it might be a block is submitted to the network. Most of the time it will be rejected because it's not the real block you've found.

About stopping in windows:

I'm working on trying to make it more robust, but ctrl-break is the only way to stop it at the moment.

Other stuff:

Low priority while I try to get cgminer as stable as possible with the most frequently requested or sponsored features. While I have spent a LOT of time on this project, I do not have infinite time to implement everything in such a short time space.

x-roll-ntime is supported, but indirectly. The "local generation of work" cgminer does is rolling ntime, and is only done when network connectivity cannot keep up with demand for work. I'm not sure how much I'll be expanding on this feature in the future. I'd be surprised if cgminer couldn't keep massive hashes busy already given how the work is gathered for the worker threads at the moment.

Updated git tree:

- Disable restarting of CPU mining threads pending further investigation.
- Update longpoll messages.
- Add new block data to status line.
- Fix opencl tests for osx.
- Only do local generation of work if the work item is not stale itself.
- Check for stale work within the mining threads and grab new work if
positive.
- Test for idle network conditions and prevent threads from being restarted
by the watchdog thread under those circumstances.
- Make sure that local work generation does not continue indefinitely by
stopping it after 10 minutes.
- Tweak the kernel to have a shorter path using a 4k buffer and a mask on the
nonce value instead of a compare and loop for a shorter code path.
- Allow queue of zero and make that default again now that we can track how
work is being queued versus staged. This can decrease reject rates.
- Queue precisely the number of mining threads as longpoll_staged after a
new block to not generate local work.

Install the ati amd sdk:
http://developer.amd.com/sdks/AMDAPPSDK/downloads/Pages/default.aspx