DiabloMiner GPU Miner

[brunoshady]

and why diablo miner shows xxxx/yyyyy khash ?

what does the yyyy means?

[1713557252]

1713557252

[1713557252]

1713557252

[1713557252]

1713557252

[DiabloD3]

Code:

[6/3/11 6:49:05 PM] Block 10 found on Barts (#2)
[6/3/11 6:49:09 PM] Block 11 found on Barts (#1)
[6/3/11 6:49:29 PM] Block 12 found on Barts (#2)
[6/3/11 6:49:40 PM] Block 13 found on Barts (#1)
[6/3/11 6:49:44 PM] Block 14 found on Barts (#2)
[6/3/11 6:50:25 PM] Block 15 found on Barts (#2)
[6/3/11 6:50:43 PM] Block 16 found on Barts (#2)
[6/3/11 6:50:57 PM] Block 17 found on Barts (#1)

So, if DM is outputting stuff like this, then are the blocks defiantly being found by the cards that are in parentheses?  What I mean is that Barts (#2) doesn't seem to be accessible by certain aticonfig options, and my hash rate seems to be that of a single GPU.  Is it possible that all blocks are being completed by Barts #1 but being labeled otherwise?

Nope. Sure you have the right speed in mind? A 6870 should be doing somewhere around 270 or so at stock speeds.

[DiabloD3]

how can I check my sdk version?

where can I see it?

It spits it out on startup now.

[DiabloD3]

and why diablo miner shows xxxx/yyyyy khash ?

what does the yyyy means?

15 second average/forever average

[DiabloD3]

Okay I'll try Diablo with a smaller pool that has been more reliable for me in the past and let you know if that helps.

Like I've said to others, I believe there is either a widespread bug in how some pools operate (such as giving the wrong diff target out, or incorrectly checking hashes (such as H == 0 only instead of also G <= target)), or Pheonix is incorrectly reporting that shares have been accepted when they haven't.

For me, on Deepbit, I get two or three stale shares for every 1000 give or take.

I connected to a small pool I often use with Diablo. It kept up reporting 304Mhash/s but after several hours the pool reported that of the 1231 requested blocks Diablo had only been able to return 854, for an efficiency of 69%.

With phoenix my efficiency is upwards of 90% even though my Mhash rate is slower by 5.  I don't know enough about the long polling process or the way the rpcs work to identify if this is a bug with my pool or the miner.  But my effective hash rate is certainly better with phoenix.

You're suffering from the bug. It eats about 1/3rd of hash attempts.

Although, interestingly, not using vectors doesn't trigger it.

Sigh. I hate bugs.

[whackedspinach]

Code:

[6/3/11 6:49:05 PM] Block 10 found on Barts (#2)
[6/3/11 6:49:09 PM] Block 11 found on Barts (#1)
[6/3/11 6:49:29 PM] Block 12 found on Barts (#2)
[6/3/11 6:49:40 PM] Block 13 found on Barts (#1)
[6/3/11 6:49:44 PM] Block 14 found on Barts (#2)
[6/3/11 6:50:25 PM] Block 15 found on Barts (#2)
[6/3/11 6:50:43 PM] Block 16 found on Barts (#2)
[6/3/11 6:50:57 PM] Block 17 found on Barts (#1)

So, if DM is outputting stuff like this, then are the blocks defiantly being found by the cards that are in parentheses?  What I mean is that Barts (#2) doesn't seem to be accessible by certain aticonfig options, and my hash rate seems to be that of a single GPU.  Is it possible that all blocks are being completed by Barts #1 but being labeled otherwise?

Nope. Sure you have the right speed in mind? A 6870 should be doing somewhere around 270 or so at stock speeds.

I have two 6850s and I'm getting 230 overclocked (to a core clock of 900).  I can't use aticonfig to set any setting for Barts #2, yet DM says it is finding blocks, so I'm confused if it is contributing to the hash meter or not.

[GODLIKE]

I have an i5 cpu, ATI HD5770, Win7 Pro... that's why I can't understand it isn't working

[DiabloD3]

Code:

[6/3/11 6:49:05 PM] Block 10 found on Barts (#2)
[6/3/11 6:49:09 PM] Block 11 found on Barts (#1)
[6/3/11 6:49:29 PM] Block 12 found on Barts (#2)
[6/3/11 6:49:40 PM] Block 13 found on Barts (#1)
[6/3/11 6:49:44 PM] Block 14 found on Barts (#2)
[6/3/11 6:50:25 PM] Block 15 found on Barts (#2)
[6/3/11 6:50:43 PM] Block 16 found on Barts (#2)
[6/3/11 6:50:57 PM] Block 17 found on Barts (#1)

So, if DM is outputting stuff like this, then are the blocks defiantly being found by the cards that are in parentheses?  What I mean is that Barts (#2) doesn't seem to be accessible by certain aticonfig options, and my hash rate seems to be that of a single GPU.  Is it possible that all blocks are being completed by Barts #1 but being labeled otherwise?

Nope. Sure you have the right speed in mind? A 6870 should be doing somewhere around 270 or so at stock speeds.

I have two 6850s and I'm getting 230 overclocked (to a core clock of 900).  I can't use aticonfig to set any setting for Barts #2, yet DM says it is finding blocks, so I'm confused if it is contributing to the hash meter or not.

230 sounds closer to a single 6850.

No, it can't accidentally label blocks coming from the wrong device. It is contributing to the hash meter.

You're on Linux, so did you forget to turn off Crossfire? You need to run sudo aticonfig --initial --adapter=all -f and then restart X.

[whackedspinach]

Code:

[6/3/11 6:49:05 PM] Block 10 found on Barts (#2)
[6/3/11 6:49:09 PM] Block 11 found on Barts (#1)
[6/3/11 6:49:29 PM] Block 12 found on Barts (#2)
[6/3/11 6:49:40 PM] Block 13 found on Barts (#1)
[6/3/11 6:49:44 PM] Block 14 found on Barts (#2)
[6/3/11 6:50:25 PM] Block 15 found on Barts (#2)
[6/3/11 6:50:43 PM] Block 16 found on Barts (#2)
[6/3/11 6:50:57 PM] Block 17 found on Barts (#1)

So, if DM is outputting stuff like this, then are the blocks defiantly being found by the cards that are in parentheses?  What I mean is that Barts (#2) doesn't seem to be accessible by certain aticonfig options, and my hash rate seems to be that of a single GPU.  Is it possible that all blocks are being completed by Barts #1 but being labeled otherwise?

Nope. Sure you have the right speed in mind? A 6870 should be doing somewhere around 270 or so at stock speeds.

I have two 6850s and I'm getting 230 overclocked (to a core clock of 900).  I can't use aticonfig to set any setting for Barts #2, yet DM says it is finding blocks, so I'm confused if it is contributing to the hash meter or not.

230 sounds closer to a single 6850.

No, it can't accidentally label blocks coming from the wrong device. It is contributing to the hash meter.

You're on Linux, so did you forget to turn off Crossfire? You need to run sudo aticonfig --initial --adapter=all -f and then restart X.

Do you know how I would run aticonfig to set up both of these cards with one of them having a dual head setup?

Code:

sudo aticonfig --initial=dual-head 

allows me to boot, but

Code:

sudo aticonfig --initial=dual-head --adapter=all -f

seems to generate a dual head setup for both cars that screws everything up.  What is required for the miner to use a card? Does it just need a device section in xorg.conf, or does it need to be connected to a screen?

EDIT: Okay, now (after some careful editing of xorg.conf) aticonfig shows that Barts #1 is at a GPU load of 99% while Barts #2 of 0%.  Other stats, such as temperature are confirming this.  While I know there is something wrong with my setup, are you sure there isn't a bug in the miner?  It still reports the blocks being solved by both cards.
EDIT2: If I run DM with -D 2, aticonfig still says all of the load and temp increase is on Barts #1.  Also, Crossfire is disabled.

[DiabloD3]

Code:

[6/3/11 6:49:05 PM] Block 10 found on Barts (#2)
[6/3/11 6:49:09 PM] Block 11 found on Barts (#1)
[6/3/11 6:49:29 PM] Block 12 found on Barts (#2)
[6/3/11 6:49:40 PM] Block 13 found on Barts (#1)
[6/3/11 6:49:44 PM] Block 14 found on Barts (#2)
[6/3/11 6:50:25 PM] Block 15 found on Barts (#2)
[6/3/11 6:50:43 PM] Block 16 found on Barts (#2)
[6/3/11 6:50:57 PM] Block 17 found on Barts (#1)

So, if DM is outputting stuff like this, then are the blocks defiantly being found by the cards that are in parentheses?  What I mean is that Barts (#2) doesn't seem to be accessible by certain aticonfig options, and my hash rate seems to be that of a single GPU.  Is it possible that all blocks are being completed by Barts #1 but being labeled otherwise?

Nope. Sure you have the right speed in mind? A 6870 should be doing somewhere around 270 or so at stock speeds.

I have two 6850s and I'm getting 230 overclocked (to a core clock of 900).  I can't use aticonfig to set any setting for Barts #2, yet DM says it is finding blocks, so I'm confused if it is contributing to the hash meter or not.

230 sounds closer to a single 6850.

No, it can't accidentally label blocks coming from the wrong device. It is contributing to the hash meter.

You're on Linux, so did you forget to turn off Crossfire? You need to run sudo aticonfig --initial --adapter=all -f and then restart X.

Do you know how I would run aticonfig to set up both of these cards with one of them having a dual head setup?

Code:

sudo aticonfig --initial=dual-head 

allows me to boot, but

Code:

sudo aticonfig --initial=dual-head --adapter=all -f

seems to generate a dual head setup for both cars that screws everything up.  What is required for the miner to use a card? Does it just need a device section in xorg.conf, or does it need to be connected to a screen?

EDIT: Okay, now (after some careful editing of xorg.conf) aticonfig shows that Barts #1 is at a GPU load of 99% while Barts #2 of 0%.  Other stats, such as temperature are confirming this.  While I know there is something wrong with my setup, are you sure there isn't a bug in the miner?  It still reports the blocks being solved by both cards.
EDIT2: If I run DM with -D 2, aticonfig still says all of the load and temp increase is on Barts #1.  Also, Crossfire is disabled.

Each GPU uses 3 queues in three different threads and are setup in a way that can never cause this.

If you do -D 2, only Barts #2 should have a load.

Also for dual-head on one card but not the other, I'd carefully edit the redundant screen definitions out of xorg.conf after running that second command.

[frisco]

After a whole night mining with -dd I have 80 failed attemps out of 1634 not sure if all of them are due "hardware problems" but for sure most are. I am mining on a 5770 at 915GPU 1200 Mem (a bit oced) the good point is that even running with -dd I haven't seen a MHash drop on average (194MHs using -f 60)
Deepbit is more or less accurate (170-190Mhs) but I have 2% of lifetime stales (41K shares) I don't know why but I had some problems with windows suspending the screen and losing mining capacity (going back to 70-100Mhs) and increasing the stale ratio, I think I fixed that and haven't seen the problem for some time but the stats are already there.

I am using "-w 128 -v 2  -f 60 -dd" but for sure I am not losing 1/3 of the shares but what bothers me as is I have a 5% of failed attempts misscalculated then 5% of the total hashes should be also misscalculated including some attempts that arent being recognized as such so effectively I have 185MHs rate.  If this is due to the oc probably it is better to just reset to normal speeds and get 100% efficiency on 180MHs

[gentakin]

I am using "-w 128 -v 2  -f 60 -dd" but for sure I am not losing 1/3 of the shares [..]

With -v 2, I'm not getting any hardware-issue-failed-shares (but only 262mhash/s). The problem with 1/3 hardware issues only happens with -v 19 for me. So your problem is probably with overclocking.

BTW, my temperature usually is ~71°C, so that shouldn't be a problem. It's either some nicely hidden bug in diablo, or it's a driver/hardware bug that shows when interleaving 3 sha256 runs. After digging through the .cl and .java code for some time, I gave up and am now using -v 2. 

[DiabloD3]

After a whole night mining with -dd I have 80 failed attemps out of 1634 not sure if all of them are due "hardware problems" but for sure most are. I am mining on a 5770 at 915GPU 1200 Mem (a bit oced) the good point is that even running with -dd I haven't seen a MHash drop on average (194MHs using -f 60)
Deepbit is more or less accurate (170-190Mhs) but I have 2% of lifetime stales (41K shares) I don't know why but I had some problems with windows suspending the screen and losing mining capacity (going back to 70-100Mhs) and increasing the stale ratio, I think I fixed that and haven't seen the problem for some time but the stats are already there.

I am using "-w 128 -v 2  -f 60 -dd" but for sure I am not losing 1/3 of the shares but what bothers me as is I have a 5% of failed attempts misscalculated then 5% of the total hashes should be also misscalculated including some attempts that arent being recognized as such so effectively I have 185MHs rate.  If this is due to the oc probably it is better to just reset to normal speeds and get 100% efficiency on 180MHs

80 out of 1634 seems to be fine for an OC'ed card.

Do not try to suspend the monitor during mining, Windows tries to get clever and fails.

HW check errors do not effect stale shares on the pool, as they are never sent in.

[DiabloD3]

I am using "-w 128 -v 2  -f 60 -dd" but for sure I am not losing 1/3 of the shares [..]

With -v 2, I'm not getting any hardware-issue-failed-shares (but only 262mhash/s). The problem with 1/3 hardware issues only happens with -v 19 for me. So your problem is probably with overclocking.

BTW, my temperature usually is ~71°C, so that shouldn't be a problem. It's either some nicely hidden bug in diablo, or it's a driver/hardware bug that shows when interleaving 3 sha256 runs. After digging through the .cl and .java code for some time, I gave up and am now using -v 2. 

Certain combinations drive the hardware harder. If you're borderline overclocked too far, some settings could push you over.

Also, -f 60 could also drive the hardware harder over -f 30, so ymmv. (And no, you don't win here, 60 is slower than 30 on Windows, and causes a more variable hashrate otherwise).

[DiabloD3]

Update: I think I've refixed mtrlt's mining bug. I hope.

The math to calculate this, you need three things, mhash, number of shares (at least 100), and the number of minutes between the first and last shares.

mhash * 60 * minutes
4296 * shares

So, for example 367 mhash, 100 shares,  18 minutes

367 * 60 * 18 = 396360
4296 * 100 = 429600

So in this case, I was luckier than normal.

If you have the bug, the share hashrate will be approximately 2/3rds of the actual hashrate.

[jasonKidd]

I'm on a macbook air and i can't seem to get more than 700 khash/sec

any suggestions?
 ./DiabloMiner-OSX.sh ... -w 64
[6/4/11 12:24:51 PM] Started
[6/4/11 12:24:51 PM] Connecting to: http://..../
[6/4/11 12:24:51 PM] Using Apple OpenCL 1.0 (Dec 23 2010 17:30:26)
[6/4/11 12:24:51 PM] Added GeForce 320M (#1) (6 CU, local work size of 64)
442/619 khash/sec

[whackedspinach]

Each GPU uses 3 queues in three different threads and are setup in a way that can never cause this.

If you do -D 2, only Barts #2 should have a load.

Also for dual-head on one card but not the other, I'd carefully edit the redundant screen definitions out of xorg.conf after running that second command.

Okay, sorry to keep bothering you like this, but I'd like to figure this out.  After a night of mining, here are some stats:
Command:

Code:

./DiabloMiner-Linux.sh --url http://user.worker:pass@mining.bitcoin.cz:8332/ -v 19 -w 192 -D 1,2

Output:

Code:

6/4/11 1:55:59 PM] Started
[6/4/11 1:55:59 PM] Connecting to: http://mining.bitcoin.cz:8332/
[6/4/11 1:56:00 PM] Using AMD Accelerated Parallel Processing OpenCL 1.1 AMD-APP-SDK-v2.4 (595.10)
[6/4/11 1:56:03 PM] BFI_INT patching enabled, disabling hardware checking
[6/4/11 1:56:03 PM] Added Barts (#1) (12 CU, local work size of 192)
[6/4/11 1:56:09 PM] BFI_INT patching enabled, disabling hardware checking
[6/4/11 1:56:09 PM] Added Barts (#2) (12 CU, local work size of 192)
[6/4/11 1:56:19 PM] Block 1 found on Barts (#1)
[6/4/11 1:56:52 PM] Block 2 found on Barts (#2)
[6/4/11 1:57:05 PM] Block 3 found on Barts (#2)
[6/4/11 1:58:04 PM] Block 4 found on Barts (#2)
[6/4/11 1:58:16 PM] Block 5 found on Barts (#2)
[6/4/11 1:58:59 PM] Block 6 found on Barts (#1)
[6/4/11 2:00:21 PM] Block 7 found on Barts (#1)
226765/231103 khash/sec

aticonfig --odgc --adapter=all

Code:

Adapter 0 - AMD Radeon HD 6800 Series 
                            Core (MHz)    Memory (MHz)
           Current Clocks :    900           1000
             Current Peak :    900           1000
  Configurable Peak Range : [600-1000]     [1000-1250]
                 GPU load :    99%

Adapter 1 - AMD Radeon HD 6800 Series 
                            Core (MHz)    Memory (MHz)
           Current Clocks :    300           300
             Current Peak :    900           1000
  Configurable Peak Range : [600-1000]     [1000-1250]
                 GPU load :    0%

aticonfig --odgt --adapter=all

Code:

Adapter 0 - AMD Radeon HD 6800 Series 
            Sensor 0: Temperature - 82.00 C

Adapter 1 - AMD Radeon HD 6800 Series 
            Sensor 0: Temperature - 52.00 C

With one configuration last night I actually saw the meter go up to 460, but I can't remember what I did.  My xorg.conf is at https://gist.github.com/1008220 to show how these cards are set up. I would really appreciate it if anyone could come up with an explanation for what is going on.  Thanks in advance. 

[DiabloD3]

Each GPU uses 3 queues in three different threads and are setup in a way that can never cause this.

If you do -D 2, only Barts #2 should have a load.

Also for dual-head on one card but not the other, I'd carefully edit the redundant screen definitions out of xorg.conf after running that second command.

Okay, sorry to keep bothering you like this, but I'd like to figure this out.  After a night of mining, here are some stats:
Command:

Code:

./DiabloMiner-Linux.sh --url http://user.worker:pass@mining.bitcoin.cz:8332/ -v 19 -w 192 -D 1,2

Output:

Code:

6/4/11 1:55:59 PM] Started
[6/4/11 1:55:59 PM] Connecting to: http://mining.bitcoin.cz:8332/
[6/4/11 1:56:00 PM] Using AMD Accelerated Parallel Processing OpenCL 1.1 AMD-APP-SDK-v2.4 (595.10)
[6/4/11 1:56:03 PM] BFI_INT patching enabled, disabling hardware checking
[6/4/11 1:56:03 PM] Added Barts (#1) (12 CU, local work size of 192)
[6/4/11 1:56:09 PM] BFI_INT patching enabled, disabling hardware checking
[6/4/11 1:56:09 PM] Added Barts (#2) (12 CU, local work size of 192)
[6/4/11 1:56:19 PM] Block 1 found on Barts (#1)
[6/4/11 1:56:52 PM] Block 2 found on Barts (#2)
[6/4/11 1:57:05 PM] Block 3 found on Barts (#2)
[6/4/11 1:58:04 PM] Block 4 found on Barts (#2)
[6/4/11 1:58:16 PM] Block 5 found on Barts (#2)
[6/4/11 1:58:59 PM] Block 6 found on Barts (#1)
[6/4/11 2:00:21 PM] Block 7 found on Barts (#1)
226765/231103 khash/sec

aticonfig --odgc --adapter=all

Code:

Adapter 0 - AMD Radeon HD 6800 Series 
                            Core (MHz)    Memory (MHz)
           Current Clocks :    900           1000
             Current Peak :    900           1000
  Configurable Peak Range : [600-1000]     [1000-1250]
                 GPU load :    99%

Adapter 1 - AMD Radeon HD 6800 Series 
                            Core (MHz)    Memory (MHz)
           Current Clocks :    300           300
             Current Peak :    900           1000
  Configurable Peak Range : [600-1000]     [1000-1250]
                 GPU load :    0%

aticonfig --odgt --adapter=all

Code:

Adapter 0 - AMD Radeon HD 6800 Series 
            Sensor 0: Temperature - 82.00 C

Adapter 1 - AMD Radeon HD 6800 Series 
            Sensor 0: Temperature - 52.00 C

With one configuration last night I actually saw the meter go up to 460, but I can't remember what I did.  My xorg.conf is at https://gist.github.com/1008220 to show how these cards are set up. I would really appreciate it if anyone could come up with an explanation for what is going on.  Thanks in advance. 

Woah, that xorg.conf doesn't look right. Without -D, does it list, say, 4 Barts?

[DiabloD3]

I'm on a macbook air and i can't seem to get more than 700 khash/sec

any suggestions?
 ./DiabloMiner-OSX.sh ... -w 64
[6/4/11 12:24:51 PM] Started
[6/4/11 12:24:51 PM] Connecting to: http://..../
[6/4/11 12:24:51 PM] Using Apple OpenCL 1.0 (Dec 23 2010 17:30:26)
[6/4/11 12:24:51 PM] Added GeForce 320M (#1) (6 CU, local work size of 64)
442/619 khash/sec

You should get about 6 mhash. Try a higher -w value, Apple's OpenCL implementation is extremely buggy.

[whackedspinach]

Each GPU uses 3 queues in three different threads and are setup in a way that can never cause this.

If you do -D 2, only Barts #2 should have a load.

Also for dual-head on one card but not the other, I'd carefully edit the redundant screen definitions out of xorg.conf after running that second command.

Okay, sorry to keep bothering you like this, but I'd like to figure this out.  After a night of mining, here are some stats:
Command:

Code:

./DiabloMiner-Linux.sh --url http://user.worker:pass@mining.bitcoin.cz:8332/ -v 19 -w 192 -D 1,2

Output:

Code:

6/4/11 1:55:59 PM] Started
[6/4/11 1:55:59 PM] Connecting to: http://mining.bitcoin.cz:8332/
[6/4/11 1:56:00 PM] Using AMD Accelerated Parallel Processing OpenCL 1.1 AMD-APP-SDK-v2.4 (595.10)
[6/4/11 1:56:03 PM] BFI_INT patching enabled, disabling hardware checking
[6/4/11 1:56:03 PM] Added Barts (#1) (12 CU, local work size of 192)
[6/4/11 1:56:09 PM] BFI_INT patching enabled, disabling hardware checking
[6/4/11 1:56:09 PM] Added Barts (#2) (12 CU, local work size of 192)
[6/4/11 1:56:19 PM] Block 1 found on Barts (#1)
[6/4/11 1:56:52 PM] Block 2 found on Barts (#2)
[6/4/11 1:57:05 PM] Block 3 found on Barts (#2)
[6/4/11 1:58:04 PM] Block 4 found on Barts (#2)
[6/4/11 1:58:16 PM] Block 5 found on Barts (#2)
[6/4/11 1:58:59 PM] Block 6 found on Barts (#1)
[6/4/11 2:00:21 PM] Block 7 found on Barts (#1)
226765/231103 khash/sec

aticonfig --odgc --adapter=all

Code:

Adapter 0 - AMD Radeon HD 6800 Series 
                            Core (MHz)    Memory (MHz)
           Current Clocks :    900           1000
             Current Peak :    900           1000
  Configurable Peak Range : [600-1000]     [1000-1250]
                 GPU load :    99%

Adapter 1 - AMD Radeon HD 6800 Series 
                            Core (MHz)    Memory (MHz)
           Current Clocks :    300           300
             Current Peak :    900           1000
  Configurable Peak Range : [600-1000]     [1000-1250]
                 GPU load :    0%

aticonfig --odgt --adapter=all

Code:

Adapter 0 - AMD Radeon HD 6800 Series 
            Sensor 0: Temperature - 82.00 C

Adapter 1 - AMD Radeon HD 6800 Series 
            Sensor 0: Temperature - 52.00 C

With one configuration last night I actually saw the meter go up to 460, but I can't remember what I did.  My xorg.conf is at https://gist.github.com/1008220 to show how these cards are set up. I would really appreciate it if anyone could come up with an explanation for what is going on.  Thanks in advance. 

Woah, that xorg.conf doesn't look right. Without -D, does it list, say, 4 Barts?

It lists 3 Barts.  I reran aticonfig --initial=dual-head --adapter=all -f  again and this time I didn't change anything.  Now it works, and I am getting about 460.  I seems that it must have a screen for each card or else neither aticonfig or the miner can see it.  Here is my current configuration if anyone is interested:  https://gist.github.com/1008416  Thanks for all of your help, DiabloD3.