Pages:
Author

Topic: Phoenix - Efficient, fast, modular miner - page 37. (Read 760839 times)

newbie
Activity: 41
Merit: 0
Small request... but could you please add a display for an "efficiency" ratio/percentage? It's easy to just glance at accepted/rejected, but I'd like to see a comparison of "get works" vs accepted.
Currently I have to manually count how many times the log prints that it received new work, then compare that to the accepts and rejects (rejects have been very rare as of the new version.)

Also, probably a dumb question, but when I load the miner, I get a double request... is this something that can be addressed, or is it a server-side issue?

Example:
Code:
[10/05/2011 20:57:51] Phoenix 1.46 starting...
[10/05/2011 20:57:51] Connected to server
[10/05/2011 20:57:51] Server gave new work; passing to WorkQueue
[10/05/2011 20:57:51] New block (WorkQueue)
[10/05/2011 20:57:51] Server gave new work; passing to WorkQueue
I always get new work after a new block, but there's always a 'new' block after I start the app.

Also... could there be a way to dump a temp file on close, so if you're just restarting, it can resume the work unit it had previously instead of requesting a new one? Most likely I'd expect it would be timestamped, and ignored if older than a few seconds. One such instance where this is useful is when you release a new version and I want to close out, extract the new files, then relaunch it.

If I was on a linux box, I'd just manually edit the source and merge with your svn changes, but since I'm on windows I am relying on your binary. I'll probably end up just learning the basics of python and compiling this myself, but for now I'm just being lazy and hoping you'd be kind enough to consider these modifications. Especially since it may benefit people other than myself.

Thanks for your time, sorry if I'm incoherent, drowsy from allergy meds about to pass out :-|
newbie
Activity: 8
Merit: 0
Can someone post instructions from a barebox to making this work ?

I'm having trouble getting it to run. I have bitcoind running as a daemon and that's about it.



 File "./phoenix.py", line 123, in
    miner.start(options)
  File "/home/fahayek/Downloads/phoenix-1.46/Miner.py", line 74, in start
    self.kernel = self.options.makeKernel(KernelInterface(self))
  File "./phoenix.py", line 111, in makeKernel
    kernelModule = imp.load_module(module, file, filename, smt)
  File "kernels/poclbm/__init__.py", line 22, in
    import pyopencl as cl
  File "/usr/lib/pymodules/python2.7/pyopencl/__init__.py", line 3, in
    import pyopencl._cl as _cl
ImportError: libOpenCL.so.1: cannot open shared object file: No such file or directory

member
Activity: 78
Merit: 10
same here in 1.45.  it reports 0% stales but deepbit reports much more.

also, the % stales I had before (~.27%) is now way up to .31% so it is definitely worse...
member
Activity: 118
Merit: 10
exactly the same with deepbit.
although i see 0 rejected in all workers, some workers have 5% stales!
others have 0%
very strange!

I can confirm this behaviour with Bitcoinpool. Phoenix miner is reporting more accepted shares than the site is.
sr. member
Activity: 288
Merit: 250
As I asked before .. how does Phoenix calculate local hashrate?

I observed that at aggression 7 I can get 430MH/sec, while at aggression 10 I can get only 405, but at this lower local hashrate I am able to finish succesfully more work shares than at aggression 7 (tested for half a day, card compared to card, etc.).

So, better aggression yields in hotter GPU and better yield, while the local hashrate is actually reported lower.


The displayed hashrate is the average over several kernel executions. (configure with -a, default is 10) Each sample in the average is calculated using: (nonces per execution/time taken)

Higher AGGRESSION increases both the number of nonces per execution and the time taken.

I have tried AGGRESSION values up to 14 and the displayed hashrate continues to increase with each step.
At 909Mhz It shows me 351Mhash/s, when i bump it to 910Mhz i get 362Mhash/s. But then at 911Mhz I get 351Mhash/s again. Using agression 12, fastloop=false and worksize=128 on my unlocked 6950.
Seems somewhat a bug in the miner.



DUDUDUDUDUDUDUDUDE, If you go to AMD CCC and make powertune +20 you can get MUCH MORE HASH, due to the card not downclocking when it gets over 200w.

ALSO: you must OC with CCC
First of all, you don't have to OC with CCC. I'm using afterburner and that's what i have always used. Powertune is already set to +20%.
It's not downclocking, heaven benchmark shows me increased fps at 950 then at 910.
newbie
Activity: 35
Merit: 0
As I asked before .. how does Phoenix calculate local hashrate?

I observed that at aggression 7 I can get 430MH/sec, while at aggression 10 I can get only 405, but at this lower local hashrate I am able to finish succesfully more work shares than at aggression 7 (tested for half a day, card compared to card, etc.).

So, better aggression yields in hotter GPU and better yield, while the local hashrate is actually reported lower.


The displayed hashrate is the average over several kernel executions. (configure with -a, default is 10) Each sample in the average is calculated using: (nonces per execution/time taken)

Higher AGGRESSION increases both the number of nonces per execution and the time taken.

I have tried AGGRESSION values up to 14 and the displayed hashrate continues to increase with each step.
At 909Mhz It shows me 351Mhash/s, when i bump it to 910Mhz i get 362Mhash/s. But then at 911Mhz I get 351Mhash/s again. Using agression 12, fastloop=false and worksize=128 on my unlocked 6950.
Seems somewhat a bug in the miner.



DUDUDUDUDUDUDUDUDE, If you go to AMD CCC and make powertune +20 you can get MUCH MORE HASH, due to the card not downclocking when it gets over 200w.

ALSO: you must OC with CCC
full member
Activity: 226
Merit: 100
exactly the same with deepbit.
although i see 0 rejected in all workers, some workers have 5% stales!
others have 0%
very strange!
member
Activity: 112
Merit: 10
Either ther is a problem with phoenix 1.46 or bitcoinpool
I get no rejects but the pool consistently is showing less accepted shares than phonix
WTF
newbie
Activity: 56
Merit: 0
ok I guess thats easier than posting a url link with hundreds of post
That's usually why it's done that way. Some development threads can get to be hundreds of pages long or more. Sorry for sounding snarky. Mornings, no one loves em. Also, I tend to forget that not everyone spends large amounts of time on various internet forums.
member
Activity: 112
Merit: 10
They're on the first post of this thread? at least, thats where I found them. That's the way proper forum software development usually works. Update software, edit main post to reflect changes.

ok I guess thats easier than posting a url link with hundreds of post
Guess I should have know that the first post was edited... really
I'm not a mind reader
newbie
Activity: 56
Merit: 0
They're on the first post of this thread? at least, thats where I found them. That's the way proper forum software development usually works. Update software, edit main post to reflect changes.
member
Activity: 112
Merit: 10
1.46 has been released which fixes the idle problem.

ok where do you get it ??
Are you serious ?  Roll Eyes First post ...

are you serious.. seemed clear enough... were are the downloads for 1.46
Is it so hard to post the url when announcing an updated version ?
newbie
Activity: 58
Merit: 0
I was trying to build phoenix from sources under windows, but I can't make pyopencl works.
What version are you using in your setup for windows ?
Where can I find compiled one ? Which will work with python2.7

Thanks.
I couldn't get pyopencl to work from source either.
Try the binary package found here, it's what I ended up using: http://www.lfd.uci.edu/~gohlke/pythonlibs/

I've tried this package. But it doesn't work for me with last Python 2.7. When I'm trying to import pyopencl it fails with dll error.
newbie
Activity: 40
Merit: 0
I was trying to build phoenix from sources under windows, but I can't make pyopencl works.
What version are you using in your setup for windows ?
Where can I find compiled one ? Which will work with python2.7

Thanks.
I couldn't get pyopencl to work from source either.
Try the binary package found here, it's what I ended up using: http://www.lfd.uci.edu/~gohlke/pythonlibs/
newbie
Activity: 58
Merit: 0
I was trying to build phoenix from sources under windows, but I can't make pyopencl works.
What version are you using in your setup for windows ?
Where can I find compiled one ? Which will work with python2.7

Thanks.
sr. member
Activity: 288
Merit: 250
1.46 has been released which fixes the idle problem.

ok where do you get it ??
Are you serious ?  Roll Eyes First post ...
member
Activity: 112
Merit: 10
1.46 has been released which fixes the idle problem.

ok where do you get it ??
full member
Activity: 174
Merit: 100
I'd be nice if the first post of this thread was updated with patch notes instead of having to dig for them.
sr. member
Activity: 288
Merit: 250
As I asked before .. how does Phoenix calculate local hashrate?

I observed that at aggression 7 I can get 430MH/sec, while at aggression 10 I can get only 405, but at this lower local hashrate I am able to finish succesfully more work shares than at aggression 7 (tested for half a day, card compared to card, etc.).

So, better aggression yields in hotter GPU and better yield, while the local hashrate is actually reported lower.


The displayed hashrate is the average over several kernel executions. (configure with -a, default is 10) Each sample in the average is calculated using: (nonces per execution/time taken)

Higher AGGRESSION increases both the number of nonces per execution and the time taken.

I have tried AGGRESSION values up to 14 and the displayed hashrate continues to increase with each step.
At 909Mhz It shows me 351Mhash/s, when i bump it to 910Mhz i get 362Mhash/s. But then at 911Mhz I get 351Mhash/s again. Using agression 12, fastloop=false and worksize=128 on my unlocked 6950.
Seems somewhat a bug in the miner.
full member
Activity: 219
Merit: 120
As I asked before .. how does Phoenix calculate local hashrate?

I observed that at aggression 7 I can get 430MH/sec, while at aggression 10 I can get only 405, but at this lower local hashrate I am able to finish succesfully more work shares than at aggression 7 (tested for half a day, card compared to card, etc.).

So, better aggression yields in hotter GPU and better yield, while the local hashrate is actually reported lower.


The displayed hashrate is the average over several kernel executions. (configure with -a, default is 10) Each sample in the average is calculated using: (nonces per execution/time taken)

Higher AGGRESSION increases both the number of nonces per execution and the time taken.

I have tried AGGRESSION values up to 14 and the displayed hashrate continues to increase with each step.
Pages:
Jump to: