On 69xx you can still use 10.11, but you need SDK 2.3 which does still seem to be stable, but it runs around 15% slower over 2.1 for mining.
11.1 and 11.2 seems to suck for everybody. Its strange.
|
|
|
10.9 through 10.11 + sdk 2.1 remains the most reliable option for most
|
|
|
Read the op post again. 11.1 does not run some or all OpenCL apps correctly.
|
|
|
If it starts with 0/0, restart it until it doesn't do that. Something in the driver hoarked, and I can't fix that.
As for attempts... thats not a bug. Those are supposed to happen. This is why they only show up during debug, that is normal operation. It means it found a block that the first 32 bits are 0 (aka a difficulty of 1)
However, yours are all on GPU #2. This means you didn't turn off Crossfire. Since you're on Windows, owning more than one GPU for mining isn't really worth it since disabling Crossfire also disables all GPUs after the first one. I suggest you mine on Linux instead.
|
|
|
As I've said in the DiabloMiner thread repeatedly, with 5xxx, use 10.10 or 10.11 with 2.1. All other combinations are inferior.
|
|
|
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.
|
|
|
3xxx Radeons can't run OpenCL. You need 4xxx and up.
|
|
|
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).
|
|
|
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.
|
|
|
Update: Automatically strip .[0-9] from $DISPLAY and set $COMPUTE, also make work group size of 64 the default instead of the hardware default
|
|
|
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.
|
|
|
Diablo, may you change cli parameters so that they correctly handle the standard URL pattern? Like this: 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 ....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: ....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.
|
|
|
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.
|
|
|
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.
|
|
|
I am trying to start DiabloMiner. What's the issue?
Exception in thread "main" java.lang.NoClassDefFoundError: com/diablominer/DiabloMiner/DiabloMiner Caused by: java.lang.ClassNotFoundException: com.diablominer.DiabloMiner.DiabloMiner at java.net.URLClassLoader$1.run(URLClassLoader.java:217) at java.security.AccessController.doPrivileged(Native Method) at java.net.URLClassLoader.findClass(URLClassLoader.java:205) at java.lang.ClassLoader.loadClass(ClassLoader.java:321) at sun.misc.Launcher$AppClassLoader.loadClass(Launcher.java:294) at java.lang.ClassLoader.loadClass(ClassLoader.java:266) Could not find the main class: com.diablominer.DiabloMiner.DiabloMiner. Program will exit.
You downloaded it from git and forgot to build it. Run mvn package.
|
|
|
Update: Merged proxy patch
|
|
|
Exactly. No reason to remove everything else if you're only accepting shares for the current block. Not round. Block.
I'm kindly asking you, geebus and FairUser, to stop trolling here again, as you did in m0mchil thread few days ago. Dude, our modifications to momchill's miner would seriously reduced the load on your server......yet we're trolling? WTF Ever. Diablo's miner does this already, but momchills doesn't. That's the point we were making on m0mchill's thread. Wait, since I implemented this in my miner, wouldn't this make me the troll?
|
|
|
Just read thru this thread and am still a little confuse on versions required. 1. What version of stream is required? 2.1 only? 2. Does stream need to be V2.1 for linux too? 3. 10.11 catalyst driver needed for windows? 4. What ati drivers are needed for linux? Im finding it hard to find a clear walkthru or setup options for this miner... can anyone help? Thanks! 2.1 runs fastest and doesnt have the CPU use bug. You only need 2.3 on 6xxx hardware. Yes, you need it on Linux too. 10.11 runs best on Linux and Windows. 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.
|
|
|
Update: I updated to LWJGL 2.7.1 yesterday, but some people were experiencing problems, so I rolled it back to 2.6
|
|
|
If you figure out what, tell me. Nope, no flaw. But my miner drives hardware and drivers harder than m0's, so its probably tripping a bug in the driver somewhere. You're not the first person to see that happen, but it should be rare. Going the whole way back to 10.11 has solved it for the people who saw this.
|
|
|
|