OFFICIAL CGMINER mining software thread for linux/win/osx/mips/arm/r-pi 4.11.1

[greatwolf]

I just downloaded cgminer 2.3.3 and I thought I'd give it a try. My initial impressions of it is that it's rather buggy and general usage is complicated by the long list of options displayed from cgminer --help.

Using this system setup:
  Win7
  Intel Q6600 Quad Core
  Nvidia GTS250 w/ Driver 285.62
  Latest cgminer win32-binary available: version 2.3.3 at time of this posting

Here are the list of problems I'm running into:

Code:

cgminer --benchmark

Causes my video driver to crash and restart. I also get the following cgminer output from the console window:

Code:

 cgminer version 2.3.3 - Started: [2012-04-19 06:56:33]
--------------------------------------------------------------------------------
 (5s):0.2 (avg):64.5 Mh/s | Q:0  A:0  R:0  HW:0  E:0%  U:0.00/m
 TQ: 2  ST: 0  SS: 0  DW: 0  NB: 0  LW: 0  GF: 0  RF: 0
 Connected to Benchmark without LP as user Benchmark
 Block: (null)...  Started:
--------------------------------------------------------------------------------
 [P]ool management [G]PU management [S]ettings [D]isplay options [Q]uit
 GPU 0: 169.3/ 64.5Mh/s | A:0 R:0 HW:0 U:0.00/m I:14
--------------------------------------------------------------------------------

[2012-04-19 06:56:33] Started cgminer 2.3.3
[2012-04-19 06:56:35] Disabling extra threads due to dynamic mode.
[2012-04-19 06:56:35] Tune dynamic intensity with --gpu-dyninterval
[2012-04-19 06:56:41] Thread 1 being disabled
[2012-04-19 06:56:56] Error: clEnqueueReadBuffer failed. (clEnqueueReadBuffer)
[2012-04-19 06:56:56] Thread 0 failure, exiting

When this happens, memory usage spikes up to 1.951+ gb and it stays there, like a memory leak has just occurred. Here's a screenshot of it:



I get the same results for all 4 kernels:

Code:

-k diablo
-k diakgcn
-k poclbm
-k phatk

In the case of using cgminer -k poclbm, I can't even get cgminer to quit by pressing 'q' or Ctrl-C. The only way to terminal the process was to kill it from task manager using 'end process'; yuck!

Here's an example output when I try to connect and test mine on a pool. The follow command was used:

Code:

C:\cgminer-2.3.3-win32> cgminer -o http://us.eclipsemc.com:8337 -u username -p password

produces the follow output:

Code:

 cgminer version 2.3.3 - Started: [2012-04-19 07:18:53]
--------------------------------------------------------------------------------
 (5s):5.4 (avg):12.3 Mh/s | Q:103  A:0  R:0  HW:0  E:0%  U:0.00/m
 TQ: 2  ST: 0  SS: 0  DW: 0  NB: 1  LW: 1562  GF: 133  RF: 0
 Connected to http://us.eclipsemc.com:8337 with LP as user username
 Block: 000006f95499ba1ca3844065a373e62a...  Started: [07:18:53]
--------------------------------------------------------------------------------
 [P]ool management [G]PU management [S]ettings [D]isplay options [Q]uit
 GPU 0:   0.0/ 12.3Mh/s | A:0 R:0 HW:0 U:0.00/m I:14
--------------------------------------------------------------------------------

[2012-04-19 07:19:10] Pool 0 not providing work fast enough
[2012-04-19 07:19:10] Pool 0 not providing work fast enough
[2012-04-19 07:19:10] Pool 0 not providing work fast enough
[2012-04-19 07:19:10] Pool 0 not providing work fast enough
[2012-04-19 07:19:10] Pool 0 not providing work fast enough
[2012-04-19 07:19:10] Pool 0 not providing work fast enough
[2012-04-19 07:19:10] Pool 0 not providing work fast enough
[2012-04-19 07:19:10] Pool 0 not providing work fast enough
[2012-04-19 07:19:10] Pool 0 not providing work fast enough
[2012-04-19 07:19:10] Pool 0 not providing work fast enough
[2012-04-19 07:19:11] Pool 0 not providing work fast enough
[2012-04-19 07:19:11] Pool 0 not providing work fast enough

Other miners I've used like phoenix and ufa-soft miner do not exhibit such problems and they work as expected. On the other hand, I can't even get a proper benchmark read out from cgminer. A search on this thread shows someone else with nvidia hardware encountering similar problems and difficulty though they were using an earlier version 2.2.3. Apparently the latest cgminer build hasn't address these problems yet.

[1713971500]

1713971500

[1713971500]

1713971500

[1713971500]

1713971500

[1713971500]

1713971500

[kano]

...

Luke-jr's name is in the code and the AUTHORS file - he added his name after I did, but 'above' mine coz I guess he felt he was more important

I've left it in that order but in the hope that one day I could make this comment ... which I just did

Anyway, on the subject of accepting code - yep even Luke-jr's code gets accepted.
However, like my code, it gets vetted by ckolivas before it goes in.

The problematic things he seems to like to do are related to changing the way things look because he wants it to look different - and they have been rejected in the past.
Otherwise, I think most of his code changes have gone in.

Anyone who uses a BFL (or even an Icarus) is using the code that Luke-jr wrote.
Icarus code was based on the BFL code, a lot was rewritten due to the different way they work and then other changes have happened to the Icarus code since then (hmm I should add my latest LP Icarus change to the BFL code whenever I get around to sending a pull request ...)

[Luke-Jr]

The problematic things he seems to like to do are related to changing the way things look because he wants it to look different - and they have been rejected in the past.
Otherwise, I think most of his code changes have gone in.

No, generally I haven't touched (nor have any particular interest in touching) display/output. The only times I've affected it were accidental/secondary from the real code change; the first, refactoring the code to support different drivers (this initially moved the driver-specific information - temperature and fan speed for GPUs - to the end of the device info line), and the second, fixing a bug I introduced with that first change when I used "BFL" to represent "FPGA" in the internal driver API (this was fixed by correcting it to "PGA" which kanoi suggested).

[Epoch]

I want the best possible code, putting a purity test of code (which is pure logic) is just asinine.  Luke has found bugs (and fixes) in cgminer and bitcoind.  Would the community be better served by running buggier software because you dislike him.  As long as merges are vetted and not done without Due Diligence I honestly don't care if the unibomber wants to make a pull request.

Either the code changes have value or they doesn't.  THAT (and only that) should be the metric.

To try and get this somewhat back on topic my opininon is that changes to the interface on cgminer should be a low priority.  The API paid for by many of us, coded by Kano, and integrated and tested by conman provide the perfect path forward.  Nobody will ever agree on perfect interface.  There is no such thing.  The API allows multiple front ends to be developed independently of cgminer.

It allows separation of responsibilities:
kernel = hashing engine
cgminer = control & management
GUI = user interface, reporting, charting, etc

cgminer just needs enough of a native interface to allow low level troubleshooting.   So many people cling to the obsolete GUIminer that I am surprised nobody has made a Windows GUI interface for cgminer (maybe I should).

+1 for maintaining sanity. Also agree with keeping UI development on cgminer minimal. There is great value and benefit keeping the UI decoupled from the 'work' code. It is a good design/implementation practice which manages complexity while adding flexibility. I once wondered why cgminer didn't have a Windows front-end, but it didn't take long to realize that the existing curses-based text UI is more than adequate.

[DeathAndTaxes]

Does cgminer config file have a "comment" symbol?

If not could it be added to next version?  #?

The simplest/easiest method would simply to only support leading charecter comment symbol. 

Code:

#This line is a comment
# "auto-gpu" : true
#The command above has been commented out.

"auto-gpu" : true  #This is invalid because the format only support leading char comment symbols

More comprehensive would be to allow parse right comments (like the last line in code above) but I would be happy for a leading comment check.

[jjiimm_64]

Does cgminer config file have a "comment" symbol?

If not could it be added to next version?  #?

The simplest/easiest method would simply to only support leading charecter comment symbol. 

Code:

#This line is a comment
# "auto-gpu" : true
#The command above has been commented out.

"auto-gpu" : true  #This is invalid because the format only support leading char comment symbols

More comprehensive would be to allow parse right comments (like the last line in code above) but I would be happy for a leading comment check.

this works:

Code:

"__auto-gpu" : true

precede any attribute with underscores will prevent that attribute from loading.

[jddebug]

I want the best possible code, putting a purity test of code (which is pure logic) is just asinine.  Luke has found bugs (and fixes) in cgminer and bitcoind.  Would the community be better served by running buggier software because you dislike him.  As long as merges are vetted and not done without Due Diligence I honestly don't care if the unibomber wants to make a pull request.

Either the code changes have value or they doesn't.  THAT (and only that) should be the metric.

To try and get this somewhat back on topic my opininon is that changes to the interface on cgminer should be a low priority.  The API paid for by many of us, coded by Kano, and integrated and tested by conman provide the perfect path forward.  Nobody will ever agree on perfect interface.  There is no such thing.  The API allows multiple front ends to be developed independently of cgminer.

It allows separation of responsibilities:
kernel = hashing engine
cgminer = control & management
GUI = user interface, reporting, charting, etc

cgminer just needs enough of a native interface to allow low level troubleshooting.   So many people cling to the obsolete GUIminer that I am surprised nobody has made a Windows GUI interface for cgminer (maybe I should).

+1 for maintaining sanity. Also agree with keeping UI development on cgminer minimal. There is great value and benefit keeping the UI decoupled from the 'work' code. It is a good design/implementation practice which manages complexity while adding flexibility. I once wondered why cgminer didn't have a Windows front-end, but it didn't take long to realize that the existing curses-based text UI is more than adequate.

While it may be true that a minimal interface is adequate, those of us that happen to have more than 8 devices on one cgminer and can no longer monitor them individually, are not receiving the same output that others see with less than 9 devices.

I can no longer monitor the temps or fan speeds or individual hash rates, rejects and hardware errors. No way of knowing IF one of the devises has a problem and, even if you knew, no way of figuring out which one has a problem either.

I have a mix of GPU and PGA's. That output is important to me. Why limit it to 8 devices?

[Epoch]

While it may be true that a minimal interface is adequate, those of us that happen to have more than 8 devices on one cgminer and can no longer monitor them individually, are not receiving the same output that others see with less than 9 devices.

I can no longer monitor the temps or fan speeds or individual hash rates, rejects and hardware errors. No way of knowing IF one of the devises has a problem and, even if you knew, no way of figuring out which one has a problem either.

I have a mix of GPU and PGA's. That output is important to me. Why limit it to 8 devices?

Valid concern, yes. I haven't tried this, but if you had 16 devices (as an example) could you run 2 instances of cgminer on the same machine, each servicing 8 devices?

[jddebug]

While it may be true that a minimal interface is adequate, those of us that happen to have more than 8 devices on one cgminer and can no longer monitor them individually, are not receiving the same output that others see with less than 9 devices.

I can no longer monitor the temps or fan speeds or individual hash rates, rejects and hardware errors. No way of knowing IF one of the devises has a problem and, even if you knew, no way of figuring out which one has a problem either.

I have a mix of GPU and PGA's. That output is important to me. Why limit it to 8 devices?

Valid concern, yes. I haven't tried this, but if you had 16 devices (as an example) could you run 2 instances of cgminer on the same machine, each servicing 8 devices?

That is kind of what i am thinking. Not nearly as slick but doable. I'll be testing that later today.

[kano]

Does cgminer config file have a "comment" symbol?

If not could it be added to next version?  #?

The simplest/easiest method would simply to only support leading charecter comment symbol. 

Code:

#This line is a comment
# "auto-gpu" : true
#The command above has been commented out.

"auto-gpu" : true  #This is invalid because the format only support leading char comment symbols

More comprehensive would be to allow parse right comments (like the last line in code above) but I would be happy for a leading comment check.

this works:

Code:

"__auto-gpu" : true

precede any attribute with underscores will prevent that attribute from loading.

It's the issue of using a 'standard' that doesn't have what you want ...
Thus the answer to the first question is no.
Of course jjiimm_64's solution works to comment out an option.
Just be sure to never save the config ...

[kano]

While it may be true that a minimal interface is adequate, those of us that happen to have more than 8 devices on one cgminer and can no longer monitor them individually, are not receiving the same output that others see with less than 9 devices.

I can no longer monitor the temps or fan speeds or individual hash rates, rejects and hardware errors. No way of knowing IF one of the devises has a problem and, even if you knew, no way of figuring out which one has a problem either.

I have a mix of GPU and PGA's. That output is important to me. Why limit it to 8 devices?

Valid concern, yes. I haven't tried this, but if you had 16 devices (as an example) could you run 2 instances of cgminer on the same machine, each servicing 8 devices?

That is kind of what i am thinking. Not nearly as slick but doable. I'll be testing that later today.

Um ... the sample miner.php now allows you to show multiple rigs with a single script
(that can run anywhere on your network you like - however you'd need the latest miner.php version from my git until it's committed)
Just add an auto refresh to the page and it can even act just like the cgminer screen ...

And of course there are much more expansive monitoring tools for cgminer.

As for the limit - well, crashing cgminer or messing up the display completely sounds like a good idea to have a limit ...

... and yes you can run 10 copies of cgminer if you really want to and don't have a very small memory limit - one for each GPU/PGA if you really want to.

[jjiimm_64]

Just be sure to never save the config ...

speaking of saving the config..  i finally got around to trying this thru the api.  when I did the voltages were all 0.00

[-ck]

I just downloaded cgminer 2.3.3 and I thought I'd give it a try. My initial impressions of it is that it's rather buggy and general usage is complicated by the long list of options displayed from cgminer --help.

Other miners I've used like phoenix and ufa-soft miner do not exhibit such problems and they work as expected. On the other hand, I can't even get a proper benchmark read out from cgminer. A search on this thread shows someone else with nvidia hardware encountering similar problems and difficulty though they were using an earlier version 2.2.3. Apparently the latest cgminer build hasn't address these problems yet.

You can't have your cake and eat it too. You either have lots of power and lots of features and with that, ways to set up and use all those features. It doesn't look like you ever actually managed to get it working at all and blaming the software is a great way to sort out the problem. At the very least start the program with -D --verbose -T and then give us the output, the way the documentation in the readme says when you have a problem. Didn't read the readme? Of course not, it's too long. Why is it too long? Because cgminer has 100 times as many features as other miners so the documentation needs to be extensive. It's a catch 22. cgminer is extensive machinery, it is NOT a fancy gui app. For that, use guiminer.

[-ck]

By the way, another example to demonstrate multiple points that have come up tonight...

Gigavps recently came in the #CGMiner channel to report a bug about the "semi-graphical" command line display malfunctioning with more than 16 devices - he had just turned on a total of 17+ FPGAs.

How did I confirm this bug? Not with 17 FPGAs - can't expect every developer to have that kind of equipment handy for testing - but by using CPUmining to generate 17 CPU threads. So yet another thing CPUmining helps test is CGMiner's basic frameworks themselves.

Unfortunately, ckolivas expressed that he would refuse to merge a fix for this issue even if I wrote it. Pretty much defeats the point. (Though I did still offer to debug and write the fix for Gigavps, at a reasonable per-hour cost; I can't blame him for declining, considering it wouldn't get merged)

P.S. Kanoi, thanks for digging out the logs which show you alone agreed to abide by your poll, but #CGMiner is a private channel and posting logs publicly is technically forbidden.

Yes, I'm a real bad guy, I don't accept code, and I keep the cgminer channel private. Playing on words makes this whole discussion look ridiculous. There is no rule about posting IRC logs publicly. It is not even remotely a private channel, and I happily accept code when it's good, useful, and not controversial. The code in question, Kano had a reasonable objection to in an area that affected him and his code, and I don't have an opinion regarding it, so I wasn't going to make the decision (unlike other times). I don't care how you spin this, but unless you and Kano can agree on it, I will continue to refuse to accept it.

[Luke-Jr]

By the way, another example to demonstrate multiple points that have come up tonight...

Gigavps recently came in the #CGMiner channel to report a bug about the "semi-graphical" command line display malfunctioning with more than 16 devices - he had just turned on a total of 17+ FPGAs.

How did I confirm this bug? Not with 17 FPGAs - can't expect every developer to have that kind of equipment handy for testing - but by using CPUmining to generate 17 CPU threads. So yet another thing CPUmining helps test is CGMiner's basic frameworks themselves.

Unfortunately, ckolivas expressed that he would refuse to merge a fix for this issue even if I wrote it. Pretty much defeats the point. (Though I did still offer to debug and write the fix for Gigavps, at a reasonable per-hour cost; I can't blame him for declining, considering it wouldn't get merged)

P.S. Kanoi, thanks for digging out the logs which show you alone agreed to abide by your poll, but #CGMiner is a private channel and posting logs publicly is technically forbidden.

Yes, I'm a real bad guy, I don't accept code, and I keep the cgminer channel private. Playing on words makes this whole discussion look ridiculous. There is no rule about posting IRC logs publicly. It is not even remotely a private channel, and I happily accept code when it's good, useful, and not controversial. The code in question, Kano had a reasonable objection to in an area that affected him and his code, and I don't have an opinion regarding it, so I wasn't going to make the decision (unlike other times). I don't care how you spin this, but unless you and Kano can agree on it, I will continue to refuse to accept it.

No, there was no objection to this fix. You're thinking of something else, where Kano admitted his objection was unreasonable.

[-ck]

No, there was no objection to this fix. You're thinking of something else, where Kano admitted his objection was unreasonable.

The land of the pedants, this be, where I must be proven wrong. What a fun game. This is starting to feel like the linux kernel mailing list and I will start ignoring people if it continues. Yes there were 3 different code issues in question. One was the PGA display thing, one was the hidden CPU mining bugs, and one was to make cgminer's display work with more than 14 devices. I refused all of them. Your debate was over the first. My issue is you muddied the water and somehow made this discussion about me being a tyrant and not accepting obvious fixes. The display thing was pretty much unrelated. Stretch this out any further, and I'll ignore you.

[Inaba]

Ok, so I haven't been following the cgminer thread... but I stopped in here because of this horrible commit that removes the display of more than 8 devices.  This basically makes cgminer useless for large farms.  I was just getting to the point where I was digging cgminer and getting comfortable with it, so I cranked it up on one of my larger boxes and low and behold it becomes a useless brick.  

Ugh.  I can't really fathom how the thought process went in removing this.  Why would you want to remove literally the most important information from cgminer?  Every other bit of information cgminer provides is less relevant than the current hashrate and temps of the units that are connected.  You can literally take every single other piece of information and do away with it and cgminer will retain value.  However, removing that information makes cgminer less functional than every other mining program out there.  So... what was the thought process behind that?

Before someone mentions the API for a different front end, the API does not work on all machines.  I have 3 machines in my farm that, if the api is enabled, will crash the machine.   Disabling the API is the only thing that will allow cgminer to run on those machines.  The API is not the solution.

[jddebug]

Ok, so I haven't been following the cgminer thread... but I stopped in here because of this horrible commit that removes the display of more than 8 devices.  This basically makes cgminer useless for large farms.  I was just getting to the point where I was digging cgminer and getting comfortable with it, so I cranked it up on one of my larger boxes and low and behold it becomes a useless brick.  

Ugh.  I can't really fathom how the thought process went in removing this.  Why would you want to remove literally the most important information from cgminer?  Every other bit of information cgminer provides is less relevant than the current hashrate and temps of the units that are connected.  You can literally take every single other piece of information and do away with it and cgminer will retain value.  However, removing that information makes cgminer less functional than every other mining program out there.  So... what was the thought process behind that?

+1

I too wish to know why it is limited to 8, why not 6 or 12 or 30? Why not allow as many as one wants and if it gets to be to many, that individual could decide to run more instances or something.

I've had to revert to 2.3.1-2 so I can at least see my individual hash rates.

[-ck]

Who was the moron who thought up 8? Me. 14 was the tipping point. Beyond 14 the display became corrupt. It is impossible to create "windows" off-screen with a curses display meaning that when I tried to make the window larger, it generated all sorts of annoying problems to do with trying to "resuscitate" the display when it's resized to a reasonable size. Now I'm quite sure you really could not care less about the reasons behind the annoying code problems that affect me. The reason I chose 8 was that I figured anyone with more than 8 devices is likely to at some stage get to 16, 20, whatever, and NOT be using GPUs, since most drivers are limited to 8 devices. Bad choice? Maybe, but there is another limit at 14 which will affect you at some stage even if you're okay at 12. So I can easily change it to 14 on the next release, but then you'll be burnt beyond that. When I created the curses interface for cgminer I never anticipated device numbers would be a limit. I CAN rewrite it entirely to make no upper limit but that would involve doing some annoying code rewrites.

[jddebug]

Who was the moron who thought up 8? Me. 14 was the tipping point. Beyond 14 the display became corrupt. It is impossible to create "windows" off-screen with a curses display meaning that when I tried to make the window larger, it generated all sorts of annoying problems to do with trying to "resuscitate" the display when it's resized to a reasonable size. Now I'm quite sure you really could not care less about the reasons behind the annoying code problems that affect me. The reason I chose 8 was that I figured anyone with more than 8 devices is likely to at some stage get to 16, 20, whatever, and NOT be using GPUs, since most drivers are limited to 8 devices. Bad choice? Maybe, but there is another limit at 14 which will affect you at some stage even if you're okay at 12. So I can easily change it to 14 on the next release, but then you'll be burnt beyond that. When I created the curses interface for cgminer I never anticipated device numbers would be a limit. I CAN rewrite it entirely to make no upper limit but that would involve doing some annoying code rewrites.

Thank you for the explanation. Having the 14 would be a great option if you would add it next time around it would definitely help.