Slight misunderstanding. I never threatened to quit if he forked cgminer. He forked it long before I was considering quitting because I knocked back some of his code. I was considering quitting because I had lost interest in the code with FPGAs taking off as I thought they did not make any business sense (and still don't in light of ROI equations with changes coming up). When ASICs were announced, the equation changed dramatically as they make a lot more sense than FPGAs do. As for me being a whiny bitch... 

That has nothing to do with what I said. It's not like I'm making the number up, it reads it back from the device. It does NOT mean the amount of memory the device has. You have read the docs correctly and that is the value reported back for that max alloc size by your device.

Much appreciated, thanks. 

I hate sockets. They never  do what you expect. Ok well it was always going to be a rough introduction.

It keeps hashing for up to 2 minutes before it  can tell the pool is down. How long did you leave it for?

Last time this happened it was a packaging error on my part. I'll reupload shortly. I've reuploaded it.

The reported size is just the reported max size allocatable by opencl, it is NOT the gpu ramsize. I already said that it will try to set it to what you set it to. It fails, and we are back to my original response - I have no idea why it fails, but that is the response to the command asking for that much memory.

First official version of cgminer supporting stratum, v2.8.0, has been released.

Thanks! Glad someone appreciates the effort  It was quite a bit of work...

Yes it works generically by reconnecting if possible after a clear disconnection, or 2 minutes of no communication from the pool.

Not offhand because it depends on how often the pool sends out a notify update and whether they support variable diff targets. The difference is there is only one lot of new information sent from the pool every 30-60 seconds regardless of the hashrate of the miner which is starkly different to the standard "getwork" protocol. If you tie that into variable diff shares, then you can theoretically make the network transfer volume STATIC regardless of the miner's hashrate with stratum.

Stratum support is now in an official cgminer release, version 2.8.0:

https://bitcointalk.org/index.php?topic=28402.msg1252632#msg1252632

Performance is almost entirely tied to the SDK version, not the driver, so no.

It's incompatible with 2.1, and 2.4 if it does work will work poorly. The scrypt kernel in cgminer relies upon features in sdk2.6+

New version - 2.8.0, 7th October 2012

This version is a massive upgrade compared to the stable 2.7 releases and involves over 1000 lines of changed code and a lot of new work implementing the stratum protocol for pooled mining, which is currently only implemented on slush's and btcguild pools. With such a massive change, breakage is likely but there have been numerous people who have graciously tested along the way paving the way for a much more bug tested release to the public. Basically you need to do nothing to make use of this new protocol, as cgminer will try to switch to it automatically if it detects it, you can force it by giving it the direct url of stratum+tcp://pool:port if you have the details from the pool. There were intentionally 2 releases on the say day, the 2.7.7 and 2.8.0 as 2.7.7 builds on the stability of the 2.7 series for those who just want a stable release and 2.8.0 has such massive changes that it should be considered relatively experimental.

More information about the stratum protocol can be found here:
https://bitcointalk.org/index.php?topic=108533.0


Human readable changelog:
Stratum support.


From the readme:

Q: What is stratum and how do I use it?
A: Stratum is a protocol designed for pooled mining in such a way as to
minimise the amount of network communications, yet scale to hardware of any
speed. With versions of cgminer 2.8.0+, if a pool has stratum support, cgminer
will automatically detect it and switch to the support as advertised if it can.
Stratum uses direct TCP connections to the pool and thus it will NOT currently
work through a http proxy but will work via a socks proxy if you need to use
one. If you input the stratum port directly into your configuration, or use the
special prefix "stratum+tcp://" instead of "http://", cgminer will ONLY try to
use stratum protocol mining. The advantages of stratum to the miner are no
delays in getting more work for the miner, less rejects across block changes,
and far less network communications for the same amount of mining hashrate.


Full changelog

- Major upgrade - support for the stratum mining protocol.
- Fix various modminer warnings on mingw.
- Fix sign warning on windows build for bitforce.
- Cast socketfail to integer since SOCKET is an unsigned int on windows.
- Use strtod not strtol for bitforce temp backup.
- Cope with broken drivers returning nonsense values for bitforce temperatures.
- Minor warning fixes.
- Use the stratum thread to detect when a stratum pool has died based on no
message for 2 minutes.
- Only set the stratum auth flag once and once the stratum thread is started,
use that to set/unset the stratum active flag.
- Only hand off to stratum from getwork if we succeed in initiating the
protocol.
- Target should only be 32 bytes copied.
- Use a static array for work submission data instead of stack memory.
- Clear the buffer data before sprinting to it.
- Clear work stratum strings before setting them and add them to debug output.
- Drop stratum connect failed message to verbose level only since it's a regular
probing message.
- TCP Keepalive in curl is only in very recent versions and not required with
regular messages on stratum anyway.
- Move stratum sockets to curl infrastructure with locking around send+recv to
begin support for proxies and ssl.
- Make detect stratum fail if a proxy has been set up.
- Stratum does not currently have any proxy support so do not try to switch to
stratum if a proxy has been specified.
- Windows doesn't work with MSG_PEEK on recv so move to a continuously updating
buffer for incoming messages.
- Alloca is unreliable on windows so use static arrays in util.c stratum code.
- Begin support for mingw stratum build.
- Add space to reject reason.
- Parse the reject reason where possible from stratum share submission.
- Pass json error value to share result function to be able to parse reject
reason in stratum.
- Don't try to parse unneeded parameters in response to mining.subscribe.
- Remove the sshare hash entry if we failed to send it.
- Change notify message to info level to avoid spamming repeatedly when a pool
is down.
- Check the stratum pool difference has not changed compared to the work diff
when testing whether a share meets the target or not and retarget if necessary.
- Bit error in target calculation for stratum.
- Set work_block in gen_stratum_work for when work is reused to avoid thinking
it's all stale.
- Offset the current block detection to the prev block hash.
- We should be testing for id_val, not id in parse stratum response.
- Make target on stratum scale to any size by clearing sequential bits according
to diff.
- Correct target calculation in gen_stratum_work.
- If a share result has an error code but still has an id, it is likely a
reject, not an error.
- Initiate stratum the first time in pool_active only, allowing us to switch to
it on getting a failed getwork and detecting the presence of stratum on the url
at that time.
- Use 5 second timeout on sock full for now as a temporary workaround.
- If no stratum url is set by the end of the detect stratum routine, copy the
sockaddr url.
- Make all buffers slightly larger to prevent overflow.
- Make the stratum recv buffer larger than the recvsize.
- Userpass needs to be copied to user and pass earlier to allow stratum
authorisation to work with it.
- Store a sockaddr url of the stripped url used in determining sockaddr to not
confuse it with the stratum url and fix build warnings.
- Decrease the queued count with stratum work once it's staged as well.
- Allow the stratum retry to initiate and auth stratum in pool_alive to make
sure the stratum thread is started.
- Avoid duplicating pool->rpc_url and setting pool->stratum_url twice to itself.
- Detect if a getwork based pool has the X-Stratum header on startup, and if so,
switch to the stratum based pool.
- Comment update.
- Minor message change.
- Create a work item from a "clean" request from stratum allowing the new block
to be detected and the appropriate block change message to be given.
- Use statically allocated stratum strings in struct work to cope with the
inability to safely deallocate dynamically allocated ram.
- Use the current pool when deciding whether to reuse work from a stratum source
rather than the work's previous pool.
- Copy the stratum url to the rpc url to avoid none being set.
- Provide locking around stratum send operations to avoid races.
- Submit shares from stratum through the abstracted submit share function
detecting what message they belong to and showing the data from the associated
work, and then deleting it from the hash.
- Use a more robust mechanism to obtain a \n terminated string over a socket.
- Abstract out share submit as a function to be useable by stratum.
- Rename parse_stratum to parse_method as it is only for stratum messages that
contain methods.
- Display stratum as mechanism in status line when current pool is running it.
- Count each stratum notify as a getwork equivalent.
- Correct nonce submitted with share.
- Extranonce2 should be added before coinbase2.
- We should be hashing the binary coinbase, not the hex one.
- Fix endianness of nonce submitted for stratum.
- Check that stratum is already active in initiate_stratum to avoid
de-authorising ourselves by subscribing again.
- Begin implementing a hash database of submissions and attempt sending results.
- Copy parameters from stratum work required for share submission.
- Set lagging flag on first adding a pool to prevent pool slow warning at
startup.
- Fix work->target being a 32 byte binary in gen_stratum_work.
- Store and display stripped url in its own variable.
- Create machinery to divert work requests to stratum.
- Generate the work target in gen_stratum_work, setting default diff to 1 in
case it is not yet set.
- Generate work data, midstate and hash1 in gen_stratum_work.
- Generate header created from stratum structures in gen_stratum_work.
- Generate merkle root hash in gen_stratum_work.
- Generate the coinbase for generation of stratum based work.
- The number of transactions is variable so make merkle a variable length
dynamically allocated array and track how many there are for stratum.
- Rename nonce2 to n2size reflecting that it's a size variable and not the
actual nonce.
- Provide rudimentary support for stratum clean work command in the stratum
thread.
- Cope with pools being removed in the stratum thread.
- Use the pool sock value directly in the stratum thread in case it changes
after reconnecting.
- Create a stratum thread per pool that has stratum that monitors the socket and
serves received data.
- Check return value of stratum_parse.
- Complete authorisation in stratum.
- Implement stratum parsing of notify parameters and storing them in the pool
stratum work structure.
- Create helper functions for duplicating json strings to avoid keeping json
references in use.
- Append \n in the sock_send function instead of adding it when constructing
json in stratum.
- Don't keep any json references around with stratum structures.
- Create parse_stratum function that hands off stratum parameters to other
functions to manage pool stratum work struct variables. Implement mining
difficulty setting.
- Create helper functions for checking when a socket is ready to read on and
receive a single line at a time. Begin stratum authorisation process.
- Provide a helper function for reading a single \n terminated string from a
socket.
- Create a stratum work structure to store current work variables.
- Test specifically for stratum being active in pool_active.
- Detect stratum in common place when adding urls, and use a bool to tell us
when it's active.
- Fix warnings.
- Extract and store various parameters on stratum init confirming successful
mining notify.
- Use existing socket macros and close the socket on failure in init stratum.
- Initiate stratum and grab first json result.
- Get detailed addressinfo from the parsed URL for future raw socket usage when
possible. IPV4 only for now.
- Prepare for getaddrinfo call.
- Add data structures to pool struct for socket communications.
- Put all socket definitions in util.h to allow reusing by added socket
functions to be used in util.c.

New stable release - Version 2.7.7,  7th October 2012

Another stable release update before bringing out a release with the massive changes of stratum support.

Human readable changelog

Fixed the many hardware errors bug and occasional crash with HW errors.
Other minor stuff

Full changelog
- Fix unused warnings on ming build.
- Fix sign warning in ocl.c
- fds need to be zeroed before set in modminer.
- Put scrypt warning on separate line to avoid 0 being shown on windows as
bufsize.
- Display correct pool number when block is found.
- Prevent corrupt values returned from the opencl code from trying to read
beyond the end of the buffer by masking the value to a max of 15.
- Icarus USB write failure is also a comms error
- api.c DEBUG message has no paramter
- Icarus catch more USB errors and close/reopen the port
- API-README update cgminer verison number
- hashmeter fix stats kh/s on 32bit windows

Okay I've done quite a bit of investigation around this "0" displayed issue. Ironically, that is a display bug in windows. The buffer size is actually being worked out to something like 1.5 billion, and if it's put on a separate line you can see that bufsize is  not zero (I'll do it in the next version). However this does not fix your original complaint that you can't set very high thread concurrency counts like you could on raper [sic]. But you've reminded me of what happened when I investigated this originally.

There are a number of problems with the way raper uses the padbuffer there. Firstly it is reused between threads which means that if you set multiple threads per device they fight over and can trash the data in the buffer. That's not a huge problem with raper because its threading is pretty primitive, unlike cgminer which is heavily multithreaded. However, the main problem is that there is NO error checking on setting values to run the opencl commands. If it  returns invalid values, raper just does it again, and assumes the hashes have been done. So it intermittently works, and intermittently just counts up a number of hashes that never happened. So what happens is you get a displayed hash rate that is really high that does not translate into a proportional rise in number of shares generated.

Summary: I implore you to compare the best share generation rate of raper to cgminer rather than the displayed hashrate.

Now that is just bizarre. I tried it even on a windows machine and it didn't give me zero...

EDIT: Nm, can now reproduce.

I'm unable to reproduce this anywhere. Can you give me your whole command line minus any account details?

Right, that is  the code indeed. The fact it's reading zero is worrying, and looks like it might well be a buffer overflow. It will set it to whatever you choose regardless if it can instead of what it detects as the maximum. However clCreateBuffer takes a size_t variable as the buffer size, and by definition that's a 32 bit unsigned integer. The problem with that is it's impossible to be larger than 2^32. Hrm... let's see

size_t bufsize = 128 * ipt * cgpu->thread_concurrency;

ipt is an unsigned integer, but thread_concurrency is a signed one. Maybe that's all there is to the overflow. I'll dick around with it before the next release.

EDIT: and by the way, I can't replicate any of this without windows...