Topic: python OpenCL bitcoin miner - page 40. (Read 1239072 times)
OK, maybe I will install it as a multiple-monitor setup and just not plug a monitor into the ATI card. I am still a bit confused about how OpenCL works; I thought the fglrx driver supported it directly. Is the SDK necessary?
I'm about to set up m0mchil's miner on openSUSE 11.3, and I have a couple of (hopefully) simple questions. Can I continue to use my motherboard-integrated Nvidia GPU for display purposes while running an ATI 5XXX card strictly for mining? I presume I add the ATI driver to the kernel but don't make any changes to X so it continues to use the Nvidia. Also, can I use the ati-fglrxG02-kmp-desktop driver from the repo at www2.ati.com/suse/11.3/, or do I need to compile from source? Thanks.
I doubt it, the ATI cards will need X displays on them (even if no physical displays), or they won't work with OpenCL. Stupid, I know. You'd be better off just using one of the ATI cards for display and disabling the on-board card, or if you have the spare slots, add another ATI card just for display.
I'm about to set up m0mchil's miner on openSUSE 11.3, and I have a couple of (hopefully) simple questions. Can I continue to use my motherboard-integrated Nvidia GPU for display purposes while running an ATI 5XXX card strictly for mining? I presume I add the ATI driver to the kernel but don't make any changes to X so it continues to use the Nvidia. Also, can I use the ati-fglrxG02-kmp-desktop driver from the repo at www2.ati.com/suse/11.3/, or do I need to compile from source? Thanks.
Thanks for this. Working fine on:
2010 iMac running OSX 10.6.5 with a 256mb HD4850
Python 2.6.1
PyOpenCL 0.92
SVN revision #19 of python-jsonrpc
Commit 7ed72feab9baac99727c4bafaadb8575787ab627 of the miner
Doing 14,000-16,000 khash/sec. The CPU miner on the Core i7 Quad was only doing 1,500 khash/sec per core.
2010 iMac running OSX 10.6.5 with a 256mb HD4850
Python 2.6.1
PyOpenCL 0.92
SVN revision #19 of python-jsonrpc
Commit 7ed72feab9baac99727c4bafaadb8575787ab627 of the miner
Doing 14,000-16,000 khash/sec. The CPU miner on the Core i7 Quad was only doing 1,500 khash/sec per core.
Everything is running great here except when I play a video my computer crashes. I'm running a Radeon 6970 and as soon as I click on a video (youtube, etc) etc I get a hard lock. If I turn off poclbm.exe no problems with the videos. Has anyone had a similar experience?
I have exactly the same issue as you described: a complete freeze of all Windows (running Vista 64bit), not even ctrl-alt-del working anymore. And I suspect everything from Adobe: not only flash player but even opening a PDF in Acrobat Reader causes my system to freeze.
I had an ATI 4870 before which didn't show this problem.
Fixed issue with intermittent solving for duplicate pool job. Anyone using the miner against a pool should upgrade to latest version to avoid losing shares.
Also, kernel now compiles with vectors on Stream 2.1
Also, kernel now compiles with vectors on Stream 2.1
Everything is running great here except when I play a video my computer crashes. I'm running a Radeon 6970 and as soon as I click on a video (youtube, etc) etc I get a hard lock. If I turn off poclbm.exe no problems with the videos. Has anyone had a similar experience?
Try backing the clock down a little if you're overclocking.
Everything is running great here except when I play a video my computer crashes. I'm running a Radeon 6970 and as soon as I click on a video (youtube, etc) etc I get a hard lock. If I turn off poclbm.exe no problems with the videos. Has anyone had a similar experience?
Yes. On my old 5770 dual crossfire setup when I would have a flash video open (youtube, securitytube, I'm pretty sure anything with flash) I would just get a dip in M/hash/s by about 20M. Never had any lock ups.Same deal with my 5970. It is a dual gpu card and only 1 gpu takes the performance hit by about the same amount in M/hash/s.
Everything is running great here except when I play a video my computer crashes. I'm running a Radeon 6970 and as soon as I click on a video (youtube, etc) etc I get a hard lock. If I turn off poclbm.exe no problems with the videos. Has anyone had a similar experience?
Running this miner on 32 bit Windows XP it works very well, about 290Mhash/sec on a HD5870 and very small CPU use.
Running on 32 bit Windows 7, the same 290Mhash/sec performance is observed, but the miner is keeping 1 CPU core fully busy.
In both cases the latest (10.12) Catalyst version is installed.
I would prefer to run on Windows 7, so... is there anything that can be done to reduce the CPU use in the Windows 7 case?
Running on 32 bit Windows 7, the same 290Mhash/sec performance is observed, but the miner is keeping 1 CPU core fully busy.
In both cases the latest (10.12) Catalyst version is installed.
I would prefer to run on Windows 7, so... is there anything that can be done to reduce the CPU use in the Windows 7 case?
Just thought I should add I am using two 5770s in crossfire. Both are overclocked to core @ 1020mhz / memory @ stock, volt @ 1.35 and are producing a total of 364 million hashes per second with vectors enabled. Everything is running fine with my setup. Blocks are accepted with no problems.
Thanks for the info. By the way, does Crossfire mean that the client sees a single GPU with doubled performance, or do you need to use two clients?
It seems I may have some thermal issues, as my card is passively cooled and needs extra care with the case airflow. When using slush's pool, a few % of my results were invalid, but the error rate goes down as I improve ventilation.
The card does feature thermal throttling, which became evident as the case fan was accidentally disconnected, and the hash rate went down. Unfortunately, it is not sensitive enough. I assume the error tolerance in GPUs is higher than in CPUs, considering the majority of uses. In any case it is hard to say if the GPU is faulty, since it works fine most of the time.
As for Sawzall's RPC problems, I wonder if this is relevant: I used to have a http_proxy variable in my shell, as some browsers (notably Chrome) could not be configured to use a proxy otherwise. Since json-rpc uses http, it was confused by the proxy, and it took me a while to see where the problem was.
Update: I swapped the PSU, and have been running error-free for 20 hours since then, about 2700 shares in the pool. Also tested with AMD's tools (aticonfig, atiode) that the GPU is fine. So it was most likely a PSU problem.
I've not had any luck connecting to slush's pool using this miner with the -o flag. I set it to mining.bitcoin.cz and I get an error reading "Problems communicating with bitcoin rpc". I got the actual IP address for the server (109.74.195.190) and put that in for -o and got the same error.
I tried setting the -p flag to 8332, and get an error saying non-numeric port.
I tried setting the -p flag to 8332, and get an error saying non-numeric port.
try "--port 8332"
I've not had any luck connecting to slush's pool using this miner with the -o flag. I set it to mining.bitcoin.cz and I get an error reading "Problems communicating with bitcoin rpc". I got the actual IP address for the server (109.74.195.190) and put that in for -o and got the same error.
I tried setting the -p flag to 8332, and get an error saying non-numeric port.
I tried setting the -p flag to 8332, and get an error saying non-numeric port.
try "--port 8332"
I have experienced some further oddities with the vectors option, since upgrading to a HD5770. The option seems to work, as in increasing the hash rate (as opposed to the slowdown on HD5570). However, the blocks found are invariably "invalid or stale". Without vectors the blocks are fine, at least on testnet.
As with the HD5570, going back to commit 99c57637ca0d1db70187 helps. The vectors option increases the rate, and blocks are OK. Unfortunately, this is somewhat slower than using the new version with vectors (about 156 vs. 164 Mhash/s) but I think I will choose the way that actually generates blocks
The new version without vectors gives about 153 Mhash/s.
The system is a Gentoo Linux, kernel 2.6.36.2, with catalyst 10.11 drivers and SDK 2.2. The PCIe slots are only 1x wide, but this probably only affects performance.
As with the HD5570, going back to commit 99c57637ca0d1db70187 helps. The vectors option increases the rate, and blocks are OK. Unfortunately, this is somewhat slower than using the new version with vectors (about 156 vs. 164 Mhash/s) but I think I will choose the way that actually generates blocks
The new version without vectors gives about 153 Mhash/s.The system is a Gentoo Linux, kernel 2.6.36.2, with catalyst 10.11 drivers and SDK 2.2. The PCIe slots are only 1x wide, but this probably only affects performance.
Just thought I should add I am using two 5770s in crossfire. Both are overclocked to core @ 1020mhz / memory @ stock, volt @ 1.35 and are producing a total of 364 million hashes per second with vectors enabled. Everything is running fine with my setup. Blocks are accepted with no problems.
Win 7 64bit. latest catalyst drivers and latest stream sdk.
I've not had any luck connecting to slush's pool using this miner with the -o flag. I set it to mining.bitcoin.cz and I get an error reading "Problems communicating with bitcoin rpc". I got the actual IP address for the server (109.74.195.190) and put that in for -o and got the same error.
I tried setting the -p flag to 8332, and get an error saying non-numeric port.
I tried setting the -p flag to 8332, and get an error saying non-numeric port.
Teknohog, could you please try with Stream 2.1. I can't see why 'vectors' would give invalid results.
With SDK 2.1, the miner does not even compile with vectors
Code:
/tmp/OCLQa31hL.cl(128): error: mixed vector-scalar operation not allowed
unless up-convertable(scalar-type=>vector-element-type)
sharound(E,F,G,H,A,B,C,D,R(36),0x650A7354);
But it gets stranger. After reinstalling 2.2, I am getting accepted results with vectors, using the latest version of poclbm. So this looks like an error in my system, and poclbm is fine
(Possible sources of error: overheated GPU, which is now cooler after the SDK changes; pyopencl, which is the only component I have to manually compile and install with every SDK change.)
Just posted my best settings for Windows 7, Stream 2.2, AMD 5770, 5870 (first post). Teknohog, could you please try with Stream 2.1. I can't see why 'vectors' would give invalid results.
I have experienced some further oddities with the vectors option, since upgrading to a HD5770. The option seems to work, as in increasing the hash rate (as opposed to the slowdown on HD5570). However, the blocks found are invariably "invalid or stale". Without vectors the blocks are fine, at least on testnet.
As with the HD5570, going back to commit 99c57637ca0d1db70187 helps. The vectors option increases the rate, and blocks are OK. Unfortunately, this is somewhat slower than using the new version with vectors (about 156 vs. 164 Mhash/s) but I think I will choose the way that actually generates blocks The new version without vectors gives about 153 Mhash/s.
The system is a Gentoo Linux, kernel 2.6.36.2, with catalyst 10.11 drivers and SDK 2.2. The PCIe slots are only 1x wide, but this probably only affects performance.
As with the HD5570, going back to commit 99c57637ca0d1db70187 helps. The vectors option increases the rate, and blocks are OK. Unfortunately, this is somewhat slower than using the new version with vectors (about 156 vs. 164 Mhash/s) but I think I will choose the way that actually generates blocks
The new version without vectors gives about 153 Mhash/s.The system is a Gentoo Linux, kernel 2.6.36.2, with catalyst 10.11 drivers and SDK 2.2. The PCIe slots are only 1x wide, but this probably only affects performance.
FairUser, please describe your setup with as many details as possible. When did it start showing this?
Does anyone else see something similar?
Does anyone else see something similar?
False alarm. My bad.
The log file is not showing repeats. We running on a console, it's fine too. When displayed through my ssh session, it does.
I think it's the way I have my shell setup. Whatever it is that's causing it to show up on my remote session twice, is NOT because of the miner.
I repeat. The problem is NOT because of poclbm. False alarm, my bad. Sorry for the mistake. Please ignore the previous posting.
Thanks again for the awesome work! What's your bitcoin address, I want to send you a little something for your efforts...
FairUser, please describe your setup with as many details as possible. When did it start showing this?
Does anyone else see something similar?
Does anyone else see something similar?
AMD Athlon(tm) X4 620 Processor
8GB RAM
1TB HD
ATI RADEON 5750
Using the following to run my miner:
python poclbm.py --user=
~149,000 khash/s average speed
I'll start dumping the output to a log to keep better track than using the buffer on my terminal.
Jump to: