New release: Version 3.11.0, 25th January 2014
Finally a proper release for hashfast devices and a substantially updated bab driver.Human readable changelog:
- A proper working hashfast driver tested working on a real device, including the windows version.
- Substantially updated BAB driver courtesy of Kano. Hopefully he can give us a summary of the changes there.
- Generic fixes for the reconnect bug on btcguild (unsure if other issues still exist).
- Work is discarded on a stratum reconnect message from the pool now (as btcguild uses) to avoid working on invalid work on switching URLs.
- Fixed the stuck line at the top of the log window.
- Message about block change detected no longer mentions longpoll.
- API now has a field "Last getwork" in summary which can be used to see that we are still getting work from pools. This is useful if you are trying to determine if a device is dead for device reasons or simply isn't getting work from any pools. It uses the same numbering in seconds with the "Last Valid Work" returned in the device API. If "Last getwork" is not incrementing, there is no work for any devices.
- Added a --nfu-bits command to allow you to set the clock speed on nanofury/icefury usb sticks. Note that the default was 54 so is now 50 to be in keeping with USB2 power limit standards. This means IT WILL BE SLOWER compared to 3.10.0 unless you change it with this option back to 54. The driver is otherwise unchanged so any other differences you see are pure variance.
- Threads names have been changed so they will show up with different, consistent names in your process manager of choice.
- Building will now not include libbitfury on every configuration unnecessarily.
- The crash on device removal has been fixed.
- Fixes for lean configurations that failed to build.
- Hashfast driver changes:
-- Should be more robust with respect to initialisation and dropouts
-- If the device stops returning shares, cgminer will reset the device, and if it cannot, it will drop it, which usually allows it to be re-hotplugged again.
-- If a device needs to be reset for not returning shares, and the clockspeed is still overclocked, cgminer will lower the clocks further on every reset.
-- Devices will be throttled if they reach 85 degrees. With water cooling they usually drop in temperature very rapidly. This can be configured with --hfa-temp-overheat
-- Hashrate displayed will be based on valid shares returned so should more accurately represent what the pool will see you hashing at. It will look very unstable initially. Earlier boards do have some loss so the previously displayed hashrate was always over. You can compare the total hashes estimated by this as "Calc hashcount" to compare with "Raw hashcount" the device has worked on in the API stats to see how they differ.
-- Max temp on screen will not show if it's an obvious error showing something like 512
-- The API will show board temperatures
-- Windows builds work
-- Babyjets show up as HFB while Sierras show up as HFS. Custom devices will come up with the generic HFA name. Note that if you are trying to select them with the --usb command, they are all seen as HFA due to the fact that cgminer can only tell them apart after they have been initialised and running.Full changelog:
- Add hashfast documentation to ASIC README
- Support the variable HFA naming throughout the driver notices.
- Set the global hfa hash clock rate to equal the lowest if we are lowering it
for a device reset since it may be re-hotplugged after failing reset.
- Decrease the hfa clock rate if it is overclocked and we have had to try
- Put a sanity check on the measured temperature in the hfa driver for obviously
- Avoid calling applog from within hfa statline before to avoid a deadlock.
- Add throttling control to hfa driver, configurable at command line, nominally
set to 85 degrees.
- Reset hfa device if no valid hashes are seen for 1 minute from the last work.
- Store when the last getwork was retrieved and display it in the API summary.
- bab - also report dead chip count screen
- Count share based hashes in the hfa driver with the device diff to get more
- Only count 2/3 of the accumulated hashes on each pass through the hfa scan
work loop to smooth out displayed hashrate.
- bab add total history HW% to API stats
- Test valid nonces in the hashfast driver allowing us to check against the
target when trying to submit them.
- No point casting a double to a uint64
- Convert the hfa hashmeter to one based on successful share return and display
the raw and calculated hash totals in the API.
- bab - remove libbitfury dependency since it requires USB
- Add description to hfa hash clock command.
- Add hfa board temperatures to API output.
- Wait for up to 0.5 seconds in the hashfast scanwork loop if no jobs are
- Label HFA devices as B or S when their configuration matches babyjet or
- Fix libbitfury being compiled in always by mistake.
- bab - spelling
- Add bab-options
- bab - tune the chip speed based on error rates
- bab record/report spie and miso errors
- Win32 falsely comes up as big endian pulling in the wrong hf protocol header.
- Remove unused components in hashfast driver.
- Check in all usb communication places for hashfast driver that the device
- Do not send a usb reset on a usb read pipe error.
- Don't replace usb pipe errors with the pipe reset return code.
- Updated hf protocol header.
- The search for extra nonce is not worth performing in the hashfast driver.
- Add core address to hfa parse nonce debugging.
- Retry sending a frame once if it has failed in hfa_send_frame
- Add extra hfa usb init errors.
- Quiet now unused variable warning in hfa detect.
- Remove unused variable.
- Add board temperature to hfa debug
- Make submit_tested_work return a bool about whether it meets the work target
- Provide a helper function for determining dev runtime and use it in the
- Look for hfa usb init header for 2 seconds, then resend the init twice more
- Really only set up the hfa crc table once.
- Generically increase the queue if we are mining on a pool without local work
generation each time we run out of work.
- Change new block detection message since longpoll is rarely relevant today.
- Change the default clockspeed bits on nanofury devices to 50 and add a command
line option to allow it to be changed.
- Use unused line at the top of the log window which often gets stuck
- Clear pool work on a stratum reconnect message.
- bab record/report spie and miso errors
- bab - cleanup old work for dead chips also
- bab add avg fail tests to API stats
- bab report bank/board/chip for dead and v.slow chips
- bab process all nonce replies per chip together
- bab reduce work delays
- bab record the number of E0s discarded
- bab - modified result parsing
- bab restore removed unused flag
- configure - correct minion name
- bab only scan valid nonce offsets
- bab record continuous (and max) bad nonces
- bab display Banks/Boards/Chips in the device window
- Modify thread naming to make them easier to identify
- bab reduce the work send delay
- bab remove results polling
- bab report SPI wait in seconds
- bab report missing chips at start and API
- bab ensure there's enough space for the nonce reply
- bab correct stats 'Send Max'
- bab allow long enough wait on ioctl() per board
- bab more I/O stats
- api.c 2014
- api allow any size stats data
- bab add processed links which excludes expired links skipped
- bab report chips per bank, hw% and ghs per chip
- bab lock access to new_nonces to ensure correct reporting
- bab report V2 banks/boards during initialisation
- bab expire chip work
- bab use only k_lists and make work handling more refined
- klist - allow adding to tail
- bab remove old unused #define
- bab correct for master git
- correct klist reallocs
- klist lists for bab
- api.c correct DEVICECODE and ordering
- Maxchips should be 384 (16 chips/board 24 boards/controller)
- bab more detailed stats and delay less when waiting for a buffer
- api add data type AVG float 3 decimal
- bab - add V2 detect with bug fix in detect
- api.c set the actual version number to 3.0
- API V3.0 unlimited socket reply size
- README update --usb
- Check for loss of device in usb read before any other code on the usbdev
- Change stratum strings under stratum_lock in reconnect and free old strings.
- Add mcp2210 compilation to want_libbitfury configs.
- Fix HF driver typo.