DiabloMiner GPU Miner

[DiabloD3]

I just installed Mac OS X 10.7 Lion in my Mac Book Pro and just wanted to let you know what I found:

It seems Java is not installed by default, the first time one runs Java, Software Update kicks in and installs it.

I get lower speed with respect to Snow Leopard, about 25% less.

I get also get a couple of error messages when the program starts and finishes:

Code:

$ ./DiabloMiner-OSX.sh -u yomi.mac -p pepitoria -o mining.bitcoin.cz -w 192 -f 1
[4/03/11 16:12:17] Started
[4/03/11 16:12:17] Connecting to: http://mining.bitcoin.cz:8332/
[4/03/11 16:12:19] ERROR: OpenCL Build Warning : Compiler build log:
<program source>:42:8: warning: unused variable 'it'
  uint it;
       ^


[4/03/11 16:12:19] Added GeForce 9400M (#1) (2 CU, local work size of 192)
Waiting...2011-03-04 16:12:28.010 java[2464:5f03] Failed to write public plist to URL file://localhost/Users/edu/Library/Saved%20Application%20State/com.apple.javajdk16.cmd.savedState/windows.plist

It seems also that the local work size has decreased from 256 to 192.

That's all I found so far, I hope someone can make good use of this information!

10.7 seems to be broken so far. Its assuming I want -Werror (if I wanted it, I would have passed it to the OpenCL compiler); two, this kernel works fine on Nvidia hardware with -w 256... however, the default is -w 64, so this should not effect anyone.

And a loss of 25% sounds horrible when 10.6 already loses 25%.

[planalpha]

Hi experts.
I'm getting this error:
ERROR: OpenCL platform ATI Stream contains no devices

lcpic shows:
05:00.0 VGA compatible controller: ATI Technologies Inc Device 6739

I followed this tutorial:
http://bitcointalk.org/index.php?topic=3356.msg47489#msg47489

Any ideas what I'm missing?

Many thanks!

[DiabloD3]

Hi experts.
I'm getting this error:
ERROR: OpenCL platform ATI Stream contains no devices

lcpic shows:
05:00.0 VGA compatible controller: ATI Technologies Inc Device 6739

I followed this tutorial:
http://bitcointalk.org/index.php?topic=3356.msg47489#msg47489

Any ideas what I'm missing?

Many thanks!

Thats a 6850, and 6xxx requires SDK 2.3 and a new enough version of fglrx.

[planalpha]

Thanks, Diablo3!
That got it working. But any idea why I'm only getting about 5000khps? Seems like something's wrong...

[DiabloD3]

Its a known problem with fglrx. It effects mostly Windows users, but some Linux users too.

It showed up in 11.1, and 11.2 doesn't fix it, and it only effects some users, and no one can figure out the exact conditions to replicate it. Even using SDK 2.1 (for 5xxx users) with 11.1/11.2 doesn't fix it (and for some, it triggers it while switching to 2.3 fixes it).

The bug also effects some polcbm users too, so its not specific to my miner.

[Cerebrum]

Thanks, Diablo3!
That got it working. But any idea why I'm only getting about 5000khps? Seems like something's wrong...

Did you try using -f 1? That improved my performance on a 5970 from about 15M Hash/s to about 330M Hash/s.

I have no idea why DiabloMiner is affected by the -f option so much more than poclbm, but apparently it is.

Also, is there a way that we can have DiabloMiner output all those interesting statistics about mining like the modified version of poclbm that's been going around here? (Invalid/Stale Percentage, Current position in the getwork, Ratio of getworks to submissions, hashrate, ratio of getworks to answers with more work in them) Those were great for getting a very condensed picture of how your mining is going.

[DiabloD3]

Thanks, Diablo3!
That got it working. But any idea why I'm only getting about 5000khps? Seems like something's wrong...

Did you try using -f 1? That improved my performance on a 5970 from about 15M Hash/s to about 330M Hash/s.

I have no idea why DiabloMiner is affected by the -f option so much more than poclbm, but apparently it is.

Also, is there a way that we can have DiabloMiner output all those interesting statistics about mining like the modified version of poclbm that's been going around here? (Invalid/Stale Percentage, Current position in the getwork, Ratio of getworks to submissions, hashrate, ratio of getworks to answers with more work in them) Those were great for getting a very condensed picture of how your mining is going.

-f effects it because my global run size code is more finely tuned than his. If larger -fs work, that indicates either a driver bug or a bug in the OS's scheduler.

I don't intend on adding all those numbers to -d because they have no use. You have zero control over those, and they are not informative.

[Cerebrum]

Thanks, Diablo3!
That got it working. But any idea why I'm only getting about 5000khps? Seems like something's wrong...

Did you try using -f 1? That improved my performance on a 5970 from about 15M Hash/s to about 330M Hash/s.

I have no idea why DiabloMiner is affected by the -f option so much more than poclbm, but apparently it is.

Also, is there a way that we can have DiabloMiner output all those interesting statistics about mining like the modified version of poclbm that's been going around here? (Invalid/Stale Percentage, Current position in the getwork, Ratio of getworks to submissions, hashrate, ratio of getworks to answers with more work in them) Those were great for getting a very condensed picture of how your mining is going.

-f effects it because my global run size code is more finely tuned than his. If larger -fs work, that indicates either a driver bug or a bug in the OS's scheduler.

I don't intend on adding all those numbers to -d because they have no use. You have zero control over those, and they are not informative.

I'd debate whether those statistics are informative or not. I don't give a damn about that efficiency statistic, nor the current getwork number, or the spot in the current getwork. However, a statistic on how many of the hash solutions found were invalid/stale would be great. That's what's telling me that I should stick with DiabloMiner instead of switching to poclbm, since poclbm reports that 3-5% of hash solutions are invalid, but gets about 2% better hashrate than DiabloMiner.

However, since DiabloMiner remains mute on that subject, I assume that no hashes computed by DiabloMiner are invalid or stale. This leads me to the conclusion that I should use DiabloMiner over poclbm, so it would be good to include that statistic, if only to allow the users to make a more informed decision about which miner they should use to get the most out of their hardware investments.

And about that not being controllable: As I understand it, that metric is within the user's control as soon as overclocking is involved. If overclocked too aggressively, cards start producing invalid hashes more frequently, no? This may be a useful tool for properly tuning overclocking settings.

[DiabloD3]

Cerebrum, I think you've overlooked -d. It tells you lots of information, including when it submits a hash, and if the hash is accepted or not.

Also, my miner double checks all solutions before submitting... if there is a hw error, it bitches very loudly without -d.

And I don't believe poclbm's hash meter is correct. Mine lists two numbers, 15 second average, forever average. poclbm does not have a meter that is comparable to either. My miner is also faster because poclbm does not try to keep work queued for the hardware.

In addition, since I queue multiple kernels in parallel, network activity for remote pools costs much less than it does on poclbm.

[Cerebrum]

Cerebrum, I think you've overlooked -d. It tells you lots of information, including when it submits a hash, and if the hash is accepted or not.

Also, my miner double checks all solutions before submitting... if there is a hw error, it bitches very loudly without -d.

And I don't believe poclbm's hash meter is correct. Mine lists two numbers, 15 second average, forever average. poclbm does not have a meter that is comparable to either. My miner is also faster because poclbm does not try to keep work queued for the hardware.

In addition, since I queue multiple kernels in parallel, network activity for remote pools costs much less than it does on poclbm.

Wonderful. Thanks for these clarifications. It is never complaining about invalid or stale hashes so I guess I'm never making any.

I was noticing that when I run a single instance of DiabloMiner with all 4 graphics cores, I get worse performance than running 4 seperate instances with one for each core. Do you have any idea why that is? The performance difference isn't huge (about 4-5%) but it's definitely present. Any idea as to why it works better with 4 independent instances than with all 4 in the same instance?

EDIT: Forgot to ask. Is there any way to start up a fake X session from the command line so that I can start the miner via SSH? Or do I have to look into getting VNC installed on my desktop so I can log into the graphical environment remotely?

[DiabloD3]

Cerebrum, I think you've overlooked -d. It tells you lots of information, including when it submits a hash, and if the hash is accepted or not.

Also, my miner double checks all solutions before submitting... if there is a hw error, it bitches very loudly without -d.

And I don't believe poclbm's hash meter is correct. Mine lists two numbers, 15 second average, forever average. poclbm does not have a meter that is comparable to either. My miner is also faster because poclbm does not try to keep work queued for the hardware.

In addition, since I queue multiple kernels in parallel, network activity for remote pools costs much less than it does on poclbm.

Wonderful. Thanks for these clarifications. It is never complaining about invalid or stale hashes so I guess I'm never making any.

I was noticing that when I run a single instance of DiabloMiner with all 4 graphics cores, I get worse performance than running 4 seperate instances with one for each core. Do you have any idea why that is? The performance difference isn't huge (about 4-5%) but it's definitely present. Any idea as to why it works better with 4 independent instances than with all 4 in the same instance?

EDIT: Forgot to ask. Is there any way to start up a fake X session from the command line so that I can start the miner via SSH? Or do I have to look into getting VNC installed on my desktop so I can log into the graphical environment remotely?

You shouldn't be getting less. Are you sure you're doing your math right? Run with -f 1 connecting to a local, and let it run for 5-10 minutes and check the second number.

If you're on Radeon hardware, you require a real X session running. You don't have to use VNC or use X for anything at all, just start your miner with DISPLAY=:0 ./DiabloMiner over ssh as the user that started X.

[Cerebrum]

Cerebrum, I think you've overlooked -d. It tells you lots of information, including when it submits a hash, and if the hash is accepted or not.

Also, my miner double checks all solutions before submitting... if there is a hw error, it bitches very loudly without -d.

And I don't believe poclbm's hash meter is correct. Mine lists two numbers, 15 second average, forever average. poclbm does not have a meter that is comparable to either. My miner is also faster because poclbm does not try to keep work queued for the hardware.

In addition, since I queue multiple kernels in parallel, network activity for remote pools costs much less than it does on poclbm.

Wonderful. Thanks for these clarifications. It is never complaining about invalid or stale hashes so I guess I'm never making any.

I was noticing that when I run a single instance of DiabloMiner with all 4 graphics cores, I get worse performance than running 4 seperate instances with one for each core. Do you have any idea why that is? The performance difference isn't huge (about 4-5%) but it's definitely present. Any idea as to why it works better with 4 independent instances than with all 4 in the same instance?

EDIT: Forgot to ask. Is there any way to start up a fake X session from the command line so that I can start the miner via SSH? Or do I have to look into getting VNC installed on my desktop so I can log into the graphical environment remotely?

You shouldn't be getting less. Are you sure you're doing your math right? Run with -f 1 connecting to a local, and let it run for 5-10 minutes and check the second number.

If you're on Radeon hardware, you require a real X session running. You don't have to use VNC or use X for anything at all, just start your miner with DISPLAY=:0 ./DiabloMiner over ssh as the user that started X.

Thanks for pointing out my obvious error. It turns out that I wasn't waiting long enough when doing that test on single vs. multiple instances, and I'm actually getting about 500KH/s more per core from the combined instance.

[FnuGk]

hi i was wondering if there is a list of all the diablominer flags. i havnt found any in the op or at any of the links provided in the op.

also if it is not asking too much could you give an explanation of what these flags do in layman terms.

im asking as i just read some of the last posts in this thread talking about the -f 1 flag resulting in more kh/s
true enough the relatively weak 320m in my macbook went from 5100kh/s to about 7000kh/s (it got a 2sec peak at more than 10mh/s) so the increase is very significant. The only draw back was the machine became unusable so i would like some info so i can find a sweet spot.

[DiabloD3]

hi i was wondering if there is a list of all the diablominer flags. i havnt found any in the op or at any of the links provided in the op.

also if it is not asking too much could you give an explanation of what these flags do in layman terms.

im asking as i just read some of the last posts in this thread talking about the -f 1 flag resulting in more kh/s
true enough the relatively weak 320m in my macbook went from 5100kh/s to about 7000kh/s (it got a 2sec peak at more than 10mh/s) so the increase is very significant. The only draw back was the machine became unusable so i would like some info so i can find a sweet spot.

Did you try starting the miner with -h?

And the sweet spot is user specific. The default for -f is 60, try moving it up and down by multiple/divisors of 60 (1, 15, 30, 60, 120, etc) until you get what you want.

[FnuGk]

i use

./DiabloMiner-OSX.sh -u MYUSERNAME -p PASSWORD -o deepbit.net -r 8332 -g 5

right now as i simply copy pasted the command from deepbit

what exactly does the -f do?


-f,--fps <arg>        target execution timing

does not make alot of sense to me

[DiabloD3]

fps means "frames per second", its how many times the OpenCL kernel runs per second. Lower values means the kernel runs for longer, reduces overhead, but decreases desktop interactivity. Higher values means the kernel runs for less time, increases overhead, but increases desktop interactivity.

Tune it until it feels right for you. The default is 60, a lot of people just say fuck it and run at 1000

[echoecho]

Just a head's up:

With the a Raedon 6850 and Catalyst 11.4 preview drivers (which have huge performance improvements for games), I'm only getting 5670/5650 khash/sec.
Unfortunately with me downgrading the Catalyst drivers is not an option.

Using suggested command line for Windows, an IRC user suggested adding "-f 30" but nothing changed.

[DiabloD3]

Try -w 256, or very low values of -f like -f 1. The bug doesn't make any sense, and is clearly in the drivers, and as I've said in other posts in this thread, it also has effected some poclbm users.

[DiabloD3]

Update: Merged jwadamson's patch that gets rid of the OSX dock icon

[Cdecker]

I had two complete lock ups of my system (probably due to me using aticonfig to adjust the fan speeds in a cronjob) at 30 hours between each other (100'000 seconds ~= 30 hours). I was wondering whether this is a known bug or one of my cards is dying on me:

Code:

[109453.772] [mi] EQ overflowing. The server is probably stuck in an infinite loop.
[109453.905] 
Backtrace:
[109454.293] 0: /usr/bin/X (xorg_backtrace+0x3b) [0x80ef31b]
[109454.293] 1: /usr/bin/X (mieqEnqueue+0x1ab) [0x80ecb5b]
[109454.293] 2: /usr/bin/X (xf86PostMotionEventP+0xd2) [0x80bd662]
[109454.330] 3: /usr/lib/xorg/modules/input/evdev_drv.so (0x684000+0x4961) [0x688961]
[109454.330] 4: /usr/lib/xorg/modules/input/evdev_drv.so (0x684000+0x4a56) [0x688a56]
[109454.330] 5: /usr/lib/xorg/modules/input/evdev_drv.so (0x684000+0x5296) [0x689296]
[109454.330] 6: /usr/bin/X (0x8048000+0x62e7f) [0x80aae7f]
[109454.330] 7: /usr/bin/X (0x8048000+0x121b4e) [0x8169b4e]
[109454.330] 8: (vdso) (__kernel_sigreturn+0x0) [0xdaa400]
[109454.365] 9: /usr/lib/xorg/extra-modules/modules/drivers/fglrx_drv.so (ulReadMmRegisterUlong+0xd7) [0x12f90b7]
[109454.368] 10: /usr/lib/xorg/extra-modules/modules/drivers/fglrx_drv.so (0xf89000+0x351733) [0x12da733]
[109454.368] 11: /usr/lib/xorg/extra-modules/modules/drivers/fglrx_drv.so (0xf89000+0x351691) [0x12da691]
[109454.368] 12: /usr/lib/xorg/extra-modules/modules/drivers/fglrx_drv.so (0xf89000+0x35163a) [0x12da63a]
[109454.368] 13: /usr/lib/xorg/extra-modules/modules/drivers/fglrx_drv.so (MCIL_WaitFor+0x99) [0x12b4a29]
[109454.368] 14: /usr/lib/xorg/extra-modules/modules/drivers/fglrx_drv.so (Cail_MCILWaitFor+0xc8) [0x12da4f8]
[109454.368] 15: /usr/lib/xorg/extra-modules/modules/drivers/fglrx_drv.so (Cail_WaitForMCIdle_Internal+0x77) [0x12ccd37]
[109454.368] 16: /usr/lib/xorg/extra-modules/modules/drivers/fglrx_drv.so (CAIL_WaitForMCIdle+0x3e) [0x12c50ee]
[109454.368] 17: /usr/lib/xorg/extra-modules/modules/drivers/fglrx_drv.so (swlCailWaitForMCIdle+0x1f) [0x12ab25f]
[109454.369] 18: /usr/lib/xorg/extra-modules/modules/drivers/fglrx_drv.so (MCIL_WaitForMCIdle+0x22) [0x12b4b52]
[109454.369] 19: /usr/lib/xorg/extra-modules/modules/drivers/fglrx_drv.so (swlPPLibCwddepm+0x67) [0x12bb787]
[109454.369] 20: /usr/lib/xorg/extra-modules/modules/drivers/fglrx_drv.so (0xf89000+0x2f3cbb) [0x127ccbb]
[109454.369] 21: /usr/lib/xorg/extra-modules/modules/drivers/fglrx_drv.so (0xf89000+0x2f151c) [0x127a51c]
[109454.369] 22: /usr/bin/X (0x8048000+0x26ee7) [0x806eee7]
[109454.369] 23: /usr/bin/X (0x8048000+0x1a5da) [0x80625da]
[109454.369] 24: /lib/libc.so.6 (__libc_start_main+0xe7) [0x28ace7]
[109454.369] 25: /usr/bin/X (0x8048000+0x1a1b1) [0x80621b1]