DiabloMiner GPU Miner

[DiabloD3]

racerx: I've put together a fairly complete guide for getting Diablo up and running on Ubuntu.

Any instructions that require using AMD's packaged version of the drivers are wrong and should never be followed. Please only use distro packaged drivers for fglrx.

Why is that? Working fine for me.
[/quote]

Because AMD is retarded and keeps shipping these things. Use your distro packaging, they know what they are doing.

[DiabloD3]

Investigating a little further: it's only the second core on the 5970 which gives me problems (Invalid block found on Cypress (#2), possible driver or hardware issue) it does however produce correct blocks too, so I'm getting nowhere with that one. The 5850 isn't detected by Diablo, although it does show up in aticonfig.

Any ideas?

Because you didn't turn Crossfire off. Removing the cable doesn't turn it off (and you don't need it for Crossfire anyways, it works fine without it) You have to run aticonfig --initial --adapter=all to write a working xorg.conf. Humble gave you bad advice. Also, for the record, 5970s contain an internal Crossfire bridge.

Also, if your DISPLAY is not set to :0, set it to that otherwise my miner won't see all your GPUs.

[DiabloD3]

Diablo, may you change cli parameters so that they correctly handle the standard URL pattern?
Like this:

Code:

foo://username:password@example.com:8042/over/there/index.dtb?type=animal;name=narwhal#nose

As described http://en.wikipedia.org/wiki/Uniform_Resource_Locator (links to RFC in bottom of page).

I want use your miner with smth like

Code:

....DiabloMiner --host home.agalakov.spb.ru:80/btc/btc-jrproxy.cgi --user a --pass a -w 64 -f 6

, but this don't work. :^(

UPD:
Better example:

Code:

....DiabloMiner --host http://a:a@home.agalakov.spb.ru:80/btc/btc-jrproxy.cgi -w 64 -f 6

Except that URL of yours is bad form with using RPC impls like json rpc and xml rpc. Why not just forward the port instead? There is a --port argument after all.

[adv]

Diablo, may you change cli parameters so that they correctly handle the standard URL pattern?
Like this:

Code:

foo://username:password@example.com:8042/over/there/index.dtb?type=animal;name=narwhal#nose

.....
example:

Code:

....DiabloMiner --host http://a:a@home.agalakov.spb.ru:80/btc/btc-jrproxy.cgi -w 64 -f 6

Except that URL of yours is bad form with using RPC impls like json rpc and xml rpc. Why not just forward the port instead? There is a --port argument after all.

Why this form is bad for the RPC? Maybe I'm wrong, just do not know the reasons.
In addition, a parameter --port also does not working correctly when the miner converts CLI-arguments in standard form:

Code:

--host home.agalakov.spb.ru/btc/btc-jrproxy.cgi --user a --pass b --port 80
....
<p>The requested URL /btc/btc-jrproxy.cgi:80/ was not found on this server.</p>
<address>Apache/2.2.16 (Debian) Server at home.agalakov.spb.ru Port 80</address>

Yes, I can forward port or use an address without a path on server-side. But besides that, within the miner parameters is still represented in the kind of a standard URL and double the work, why not make the program more universal and standard? This is not critical, just a possible nice improvement.

[DiabloD3]

Its the /btc/btc-jrproxy.cgi part of the URL thats wrong. Why use a proxy script when you can just forward the port to what you're trying to proxy? I'll eventually add a --url argument, but I'm not adding it for abuses like that.

[adv]

Its the /btc/btc-jrproxy.cgi part of the URL thats wrong. Why use a proxy script when you can just forward the port to what you're trying to proxy?

For example run proxy on a web-hosting, which does not give access to any action with the ports. This is the first, and probably not very rare use case.

I'll eventually add a --url argument, but I'm not adding it for abuses like that.

Ugh...
In the first place thank you.
And I'm not complaining about your program, I'm just suggesting a possible improvement. Sorry if it annoys you.

[DiabloD3]

Update: Automatically strip .[0-9] from $DISPLAY and set $COMPUTE, also make work group size of 64 the default instead of the hardware default

[Cdecker]

What good is the $COMPUTE? I have a problem with stripping the digit from the $DISPLAY variable, as I'm using it to turn off single cards in my setup, would be nice if there where an easy way to enable/disable single cards by specifying lists of cards on the command line. Actually come to think of it I'll add it myself and file a push request ^^

[Cdecker]

Ok, got it. I filed a pull request and also uploaded a branch to my repo with my JMX enabled code for 5 minute averages.

[DiabloD3]

What good is the $COMPUTE? I have a problem with stripping the digit from the $DISPLAY variable, as I'm using it to turn off single cards in my setup, would be nice if there where an easy way to enable/disable single cards by specifying lists of cards on the command line. Actually come to think of it I'll add it myself and file a push request ^^

$COMPUTE is the env variable, like $DISPLAY, that says which cards should be used for computing. SDK 2.3 officially recognizes it, but still uses $DISPLAY when its not set. I assume, eventually, all OpenCL impls on Linux will use $COMPUTE only (or use a much better system that isn't a pile of shit).

I'll look at your branch in a bit.

[DiabloD3]

According to Diablo, my single 5970 at stock settings, without OC is cruising at ~0.85 Ghash/s. According to BitcoinCharts, it's 1/412th of the current network hashing power, right?

Now, i know that there are more than 3-4 hundred 5970's doing Bitcoin, because most vendors are sold out, so many people have several of them in one machine, and some have entire racks of them, not to mention the communes...

So, which hashing rate number is bogus? Or am i totally missing something obvious?

Thanks in advance.
 

Some of your numbers are bogus. A stock 5970 (which is less than two stock 5870 because of factory underclocking) will around 550 mhash/sec; at full 5870 clocks its closer to around 630.

The current network computational power is around 261 ghash/sec.

261 / 0.550 = 475 stock 5970s or 414 fully clocked ones.

There is probably about 50-100 actual 5970s mining, the bulk of the rest is other Radeons, very few Nvidia users, and a bunch of CPU miner (which, although probably thousands of CPUs mining, basically add very little to the overall effort).

[lfm]

According to Diablo, my single 5970 at stock settings, without OC is cruising at ~0.85 Ghash/s. According to BitcoinCharts, it's 1/412th of the current network hashing power, right?

Now, i know that there are more than 3-4 hundred 5970's doing Bitcoin, because most vendors are sold out, so many people have several of them in one machine, and some have entire racks of them, not to mention the communes...

So, which hashing rate number is bogus? Or am i totally missing something obvious?

I see no conflict. GPUs are used for purposes other than bitcoin you know. 5970s are is demand for many things besides bitcoin, not the least of which is games.

[telestrial]

Hey there,

I was wondering if I could get a point in the right direction here.

I'm using an ATI mobility HD 3400 (super beefy, I know), have ATI STREAM SDK installed using the .deb files floating around.  When I try to start DiabloMiner I get:

Code:

ERROR: OpenCL platform ATI Stream contains no devices

I've tried export DISPLAY=:0 and export DISPLAY=:1 (the latter of which was just a stab in the dark)

Any suggestions?

[gusti]

Hey there,

I was wondering if I could get a point in the right direction here.

I'm using an ATI mobility HD 3400 (super beefy, I know), have ATI STREAM SDK installed using the .deb files floating around.  When I try to start DiabloMiner I get:

Code:

ERROR: OpenCL platform ATI Stream contains no devices

I've tried export DISPLAY=:0 and export DISPLAY=:1 (the latter of which was just a stab in the dark)

Any suggestions?

check with lspci if you can see the video card

[telestrial]

Code:

01:00.0 VGA compatible controller: ATI Technologies Inc Mobility Radeon HD 3400 Series

It does.

[DiabloD3]

3xxx Radeons can't run OpenCL. You need 4xxx and up.

[telestrial]

Alright, Thanks!

[master_shredder]

Has anyone determined whether 11.2 works any better than 11.1?

I currently have 11.2 installed on windows 7 32-bit with a ATI Radeon 5870. It does not work. It just hangs and if I inspect the error I get.

Problem signature:
  Problem Event Name:   BEX
  Application Name:   java.exe
  Application Version:   6.0.230.5
  Application Timestamp:   4cddca02
  Fault Module Name:   atiocl.dll
  Fault Module Version:   2.3.451.0
  Fault Module Timestamp:   4cfd9c01
  Exception Offset:   004245d3
  Exception Code:   c0000417
  Exception Data:   00000000
  OS Version:   6.1.7600.2.0.0.256.1
  Locale ID:   1033
  Additional Information 1:   7f7d
  Additional Information 2:   7f7d4072b1dd0e426b05ec779e23282a
  Additional Information 3:   f860
  Additional Information 4:   f860ea43be1910f329290140707d113d

[DiabloD3]

As far as I can tell, 11.2 retains whatever 11.1 caused.

11.1 on Linux included a change in the Linux kernel module API (which hasn't been done for awhile), so I suspect they really updated shit.

[qet]

Hi, I'm running diablo miner but it's just saying "0/0 khash" and never goes up from there
(one 5970 card)

Code:

[2/27/11 10:25:18 AM] Started
[2/27/11 10:25:20 AM] Added Cypress
            (#1) (20 CU, local work size of 64)
[2/27/11 10:25:23 AM] Added Cypress
            (#2) (20 CU, local work size of 64)
0/0 khash/sec

What could be the cause of this problem?