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

[outsidefactor]

is it possible to underclock the memory down to 300 using cgminer?

I clock my 7970s down to 150 with it quite successfully.

how?

GPU Settings -> Change Settings -> Select GPU -> Memory

From there you can set the memory speed.

You can also use the "gpu-memclock" config stanza in your .conf file.

I usually get a miner running how I like and then just write out the config file (Settings -> Write Config File).

[1714957866]

1714957866

[1714957866]

1714957866

[1714957866]

1714957866

[1714957866]

1714957866

[cypherdoc]

is it possible to underclock the memory down to 300 using cgminer?

I clock my 7970s down to 150 with it quite successfully.

how?

GPU Settings -> Change Settings -> Select GPU -> Memory

From there you can set the memory speed.

You can also use the "gpu-memclock" config stanza in your .conf file.

have you verified with a sensor like gpu-z that your card is actually running at 150?

when i do that with my 6970, it defaults back to stock settings.

[kano]

Hello to the cgminer community.

I have been trying to implement a background mining setup here at my work (with the consent and support of my boss). I have 20 CAD stations I want to run miners on, but only when the machines are idle (the engineers never shut them off and we source all our power from solar panels on the roof).

I have tried various options, from the old RCP screen savers (sooooooo out of date) to commandsaver. Commandsaver was close, but cgminer doesn't seem to like console redirection, plus when the screensaver ends and goes to shut down cgminer it rarely seems to exit cleanly.

People directed me to this thread: https://bitcointalk.org/?topic=8403.0, but I can't see how it's any better than command saver and still probably has the clean exiting issue.

Any ideas? cgminer is my preference (it's awesomely efficient and only needs one worker account per machine). Maybe if enough of us annoy con he'll fix the console redirection and clean shutdown issues?

You could just run it intensity 'D' ...

Or if that is an issue, set it to intensity '-10' and then have a web page/script/program that talks to each one at certain times of the day to switch the intensity up and back down as required using the RPC API ...
Basically play with the lower intensity values and work out what doesn't cause any noticeable effect on each machine and have it run at that all the time and switch it up/back at the appropriate out of hours times.

[kano]

is it possible to underclock the memory down to 300 using cgminer?

I clock my 7970s down to 150 with it quite successfully.

how?

GPU Settings -> Change Settings -> Select GPU -> Memory

From there you can set the memory speed.

You can also use the "gpu-memclock" config stanza in your .conf file.

have you verified with a sensor like gpu-z that your card is actually running at 150?

when i do that with my 6970, it defaults back to stock settings.

When you check cgminer not long after trying to change it - it will tell you what the card is really running at
... assuming the ATI ADL isn't bug ridden for you

[outsidefactor]

have you verified with a sensor like gpu-z that your card is actually running at 150?

when i do that with my 6970, it defaults back to stock settings.

I just downloaded GPU-z and confirmed the underclock.

Try the latest catalyst driver? I am running 7970s, FYI.

[TheHarbinger]

when i do that with my 6970, it defaults back to stock settings.

I think this is an issue with the 6xxx series, not the miner software.

All 6000 series cards are restricted to GPU clock -125 for memclock.  For example, if you are running at 900 on GPU, the lowest you can clock the memory is 775.  Anything lower than 125 below the GPU clock will reset the memclock to stock speed.

[-ck]

when i do that with my 6970, it defaults back to stock settings.

I think this is an issue with the 6xxx series, not the miner software.

All 6000 series cards are restricted to GPU clock -125 for memclock.  For example, if you are running at 900 on GPU, the lowest you can clock the memory is 775.  Anything lower than 125 below the GPU clock will reset the memclock to stock speed.

Windows applications have extra backdoors to the hardware to get around this. We have no cross platform->linux code that can do this, nor do I have access to how it's done on windows. So afterburner and others can work around this limitation.

[-ck]

Updated git tree. (Thanks go to sveetsnelda for most of these fixes)

Update poclbm kernel for better performance on GCN and new SDKs with bitalign support when not BFI INT patching. Update phatk kernel to work properly for non BFI INT patched kernels, providing support for phatk to run on GCN and non-ATI cards.

If you're desperate to try these on windows, downloading the new files and using the old filenames will work

https://raw.github.com/ckolivas/cgminer/master/phatk120203.cl
https://raw.github.com/ckolivas/cgminer/master/poclbm120203.cl

use them to overwrite the existing
phatk110817.cl
poclbm110817.cl

This is not a magic performance enhancement. It might improve on nvidia if you can use the phatk kernel, and it will be significantly better on GCN (79xx cards) but will not match other kernels out there yet.

[P4man]

Okay so on the temp-cutoff feature.

... . Thus auto-gpu must be enabled before temp-target, temp-overheat, or temp-cutoff are active. ....
.... The default for temp-cutoff, as described in the readme  , is 95 degrees.

Just to be clear, I had my cutoff temp configured at 80C. Without auto-gpu, cgminer will not do anything with that (my card was over 90C). Or are you saying it would have cut off at 95C?

[jake262144]

the default value of 95°C for temp-cutoff is only good for determining cutoff temperature in case the user failed to specify one.
You still need auto-gpu to enforce the temp-cutoff check.

Let me reiterate: if you want any thermal throttling in cgminer, auto-gpu must be defined.

[bulanula]

Updated git tree. (Thanks go to sveetsnelda for most of these fixes)

Update poclbm kernel for better performance on GCN and new SDKs with bitalign support when not BFI INT patching. Update phatk kernel to work properly for non BFI INT patched kernels, providing support for phatk to run on GCN and non-ATI cards.

If you're desperate to try these on windows, downloading the new files and using the old filenames will work

https://raw.github.com/ckolivas/cgminer/master/phatk120203.cl
https://raw.github.com/ckolivas/cgminer/master/poclbm120203.cl

use them to overwrite the existing
phatk110817.cl
poclbm110817.cl

This is not a magic performance enhancement. It might improve on nvidia if you can use the phatk kernel, and it will be significantly better on GCN (79xx cards) but will not match other kernels out there yet.

OK. So currently what is the best possible combination for 5870s ? SDK 2.1 and poclbm ? SDK 2.1 and other kernel ?

Which kernel has the lowest ALU OPs for 5870s ? I don't care about SDK because I can use 2.4 or 2.1 or 2.6 etc. I only care what gets me most MHash/s right now on my 5870s !

Thanks !

[jake262144]

Phatk will almost surely be faster than poclbm on old hardware AFAIK.
Why don't you do some benchmarking and post the results here?

[bulanula]

Phatk will almost surely be faster than poclbm on old hardware AFAIK.
Why don't you do some benchmarking and post the results here?

Really ? I saw people swear by poclbm kernel and ATI SDK 2.1 for 5870s but that is very moderate performance for me right now. About 430 mhash/s at 960 core and 320 memory.  

I thought phatk is to bring poclbm+2.1 performance but with sdk 2.4 ?

[jake262144]

Like I said, this is your chance to shine
Cgminer has a very useful shares parameter (e.g. "shares" : "10000") which tells it to exit once it solves a specified number of shares.
Write a script to launch cgminer consecutively with different SDKs, kernels, vectors, and worksizes and post your results here.
This will be the real deal, far more accurate than any info you can find online.

[cypherdoc]

when i do that with my 6970, it defaults back to stock settings.

I think this is an issue with the 6xxx series, not the miner software.

All 6000 series cards are restricted to GPU clock -125 for memclock.  For example, if you are running at 900 on GPU, the lowest you can clock the memory is 775.  Anything lower than 125 below the GPU clock will reset the memclock to stock speed.

not quite true.  if you use Afterburner with the modif here, you can get memclock down to 685 while running core clock in the 900's:

https://bitcointalk.org/index.php?topic=37271.0

[kano]

Like I said, this is your chance to shine
Cgminer has a very useful shares parameter (e.g. "shares" : "10000") which tells it to exit once it solves a specified number of shares.
Write a script to launch cgminer consecutively with different SDKs, kernels, vectors, and worksizes and post your results here.
This will be the real deal, far more accurate than any info you can find online.

However ... shares are random so the time to generate 10,000 of them can vary quite a lot on the same card each time you do it.

[The00Dustin]

Like I said, this is your chance to shine
Cgminer has a very useful shares parameter (e.g. "shares" : "10000") which tells it to exit once it solves a specified number of shares.
Write a script to launch cgminer consecutively with different SDKs, kernels, vectors, and worksizes and post your results here.
This will be the real deal, far more accurate than any info you can find online.

However ... shares are random so the time to generate 10,000 of them can vary quite a lot on the same card each time you do it.

So capture average MH/s over the 10,000 shares as well, and do each run more than once.  More information is better.  Have backup pools and log downtime so your numbers aren't unknowingly reduced due to outside factors.

[jake262144]

However ... shares are random so the time to generate 10,000 of them can vary quite a lot on the same card each time you do it.

Please, Kano, I never told him to do "time cgminer --shares 10000 ...", did I? The MHash/s info can be extracted from the logs.
Still, good for you to have brought the potential issue up as I was admittedly sparse with implementation details.

OTOH, with high enough shares even this simplistic approach (that is, "time cgminer...") is acceptable due to the law of large numbers.
Syncing the benchmarking with difficulty change gives you plenty of time.
I'd repeat the benchmark across at least two difficulty changes to rule out the possibility of random packet storms biasing the results.

[-ck]

Likely phatk with 2.1/2.4/2.5 will be faster on everything but GCN.

If you are benchmarking, make sure to delete any .bins before you start, and start cgminer twice after that so it can generate the bins the first time (which is slow) and then load them the second time.

Even 50 shares will do for the MH/s quoted at the end provided you have no network connectivity issues. Don't look at all at how long the shares took to generate, but look at the average MH/s.

[bulanula]

Likely phatk with 2.1/2.4/2.5 will be faster on everything but GCN.

If you are benchmarking, make sure to delete any .bins before you start, and start cgminer twice after that so it can generate the bins the first time (which is slow) and then load them the second time.

Even 50 shares will do for the MH/s quoted at the end provided you have no network connectivity issues. Don't look at all at how long the shares took to generate, but look at the average MH/s.

How can I see the values that cgminer chooses for the kernel and worksize by default if none are specified ?

Just check the logs I believe ?

What phatk kernel does cgminer include by default ? Is it made by Phateus or Diapolo user ?

Thanks !