SkipDaShu
Newbie
Offline
Activity: 27
Merit: 0
|
|
July 01, 2011, 04:43:31 AM |
|
Update: Added bitless's hack.
My 5850@918 on 2.1 went from 369 to 378, so a 2.4% increase.
Just installed the one with June 27th .jar file on the machines out in the 'server oven'. Great stuff! Pretty much got +10 MegaHootzels per card. Mostly running #! Linux AMD64, drivers v11.2 / 11.3, OpenCL 2.1, BOINC running 100% on all CPU cores/threads: 1x5850@775/875 went from 304 to 314 2x5830x@875/900 went from 533 to 553 1x5830@875/900 went from 266 to 276 6970@885/800 + 6870@995/910 went from 670 to 694. driver v11.6, OpenCL 2.4.
|
|
|
|
kwukduck
Legendary
Offline
Activity: 1937
Merit: 1001
|
|
July 01, 2011, 04:01:42 PM |
|
Could you add a version number or date in the file?
|
14b8PdeWLqK3yi3PrNHMmCvSmvDEKEBh3E
|
|
|
padrino
Legendary
Offline
Activity: 1428
Merit: 1000
https://www.bitworks.io
|
|
July 01, 2011, 11:57:12 PM |
|
Still haven't been able to get anything from bitcoins.lc with regard to the hgih rate of stales but other pools are looking really good. Added 2 more GPUs, nothing like one instance of Diablo churning at 3.2Gh/s
|
|
|
|
DiabloD3 (OP)
Legendary
Offline
Activity: 1162
Merit: 1000
DiabloMiner author
|
|
July 03, 2011, 02:03:28 AM |
|
Update: Finished adding all the old optimizations, increase speed like 1-2%
Although, I seem to be at the limit here. I went from 378 to 379 on SDK 2.1. I think I'll work on undoing more frankenkernel insanity later.
|
|
|
|
DiabloD3 (OP)
Legendary
Offline
Activity: 1162
Merit: 1000
DiabloMiner author
|
|
July 04, 2011, 02:36:39 PM |
|
Update: Removed a lot of dead code that the compiler should remove, and I think it might have been missing some.
|
|
|
|
Diapolo
|
|
July 04, 2011, 03:22:54 PM |
|
Update: Removed a lot of dead code that the compiler should remove, and I think it might have been missing some.
Did you take a look at my kernel mod ? Dia
|
|
|
|
Diapolo
|
|
July 04, 2011, 04:35:28 PM |
|
Update: Removed a lot of dead code that the compiler should remove, and I think it might have been missing some.
Did you take a look at my kernel mod ? Dia Hehe, was just going to suggest him to take a look at it Here's the link Diablo, if you need it: http://forum.bitcoin.org/index.php?topic=25860.0Hope you can use it Btw. my nick is a long time one and no offense to you Diablo . Well some changes seem like we've got similar ideas or he already took a look ^^. Dia
|
|
|
|
SkipDaShu
Newbie
Offline
Activity: 27
Merit: 0
|
|
July 04, 2011, 06:56:25 PM |
|
About 3:30am CDT all my miners when belly up except 1. The triggering event was something between me and btcguild's central server. The reason one remained up was it was connected to the east server. Anyway, here's best I could pen and paper copy from the ones that went down.... all the same major message. Exception in thread "DiabloMiner LongPollAsync" java.lang.OutOfMemoryError: Java heap space
what came after that varied a bit from nothing to either at java.util.Formatter.parse(Formatter.java:2480) at java.util.Formatter.parse(Formatter.java:2414) at java.util.Formatter.parse(Formatter.java:2367) or it referred to Getwork Parser. I suspect this was triggered by the long not able to connect to btcguild central but I suppose it could be coincidental. Installed the version with the .jar file dated 7/2/11 yesterday around 8pm CDT PS: btw, they were disconnected from central for about 8 hours because I decided to sleep in today :-(
|
|
|
|
zerokwel
|
|
July 04, 2011, 08:25:34 PM Last edit: July 04, 2011, 08:58:03 PM by zerokwel |
|
Hey all.
I decided to give Diablo miner a try as it has a option I like
Anyway every time I try and Mine using it I get a popup window saying Java has shit its self. I bet it is a Simple fix as well. Any ideas ?
Problem Event Name: BEX Application Name: java.exe Application Version: 6.0.260.3 Application Timestamp: 4dc11607 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: 6.1.7601.2.1.0.256.1 Locale ID: 2057
EDIT again its not working. Reinstalled it worked once everything fine. Closed it down to change a setting and its crashing again.
|
|
|
|
DiabloD3 (OP)
Legendary
Offline
Activity: 1162
Merit: 1000
DiabloMiner author
|
|
July 05, 2011, 01:55:55 AM |
|
Update: Removed a lot of dead code that the compiler should remove, and I think it might have been missing some.
Did you take a look at my kernel mod ? Dia Your mod looks almost identical to the changes I made when I unmacro'ed frankenkernel.
|
|
|
|
Diapolo
|
|
July 05, 2011, 05:09:08 AM |
|
Update: Removed a lot of dead code that the compiler should remove, and I think it might have been missing some.
Did you take a look at my kernel mod ? Dia Your mod looks almost identical to the changes I made when I unmacro'ed frankenkernel. I guess it is some kind of logical evolution, I didn't look into any other kernel than the original phatk. But I like this open-idea with a little feeling of competition . Dia
|
|
|
|
ancow
|
|
July 05, 2011, 08:58:31 AM |
|
Machine stats: Mac Mini with OSX 10.6.7 uname -a: Darwin zentrale1.local 10.7.0 Darwin Kernel Version 10.7.0: Sat Jan 29 15:17:16 PST 2011; root:xnu-1504.9.37~1/RELEASE_I386 i386 Apple OpenCL 1.0 (Dec 23 2010 17:30:26) Nvidia GeForce 9400 Update: Finish async networking, now everything is async. getwork, sendwork, and LP now use 1 thread per miner instance instead of 3 per GPU. Ever since this update, whenever long polling fails (because of network errors), there is a high chance that it won't pick up again (I've had >1h go by without a long polling message in debug mode). Pretty soon after the nonce saturation message, I get an OOM Exception (will paste it in here as soon as I get it again). Uodate: Behold, the frankenkernel. A mix of DiabloKernel and phatk.
Ever since this update, the hashing speed went from "mhash 2,2/2,2" to "mhash 1,4/1,4". The lates git updates don't change anything here.
|
BTC: 1GAHTMdBN4Yw3PU66sAmUBKSXy2qaq2SF4
|
|
|
DiabloD3 (OP)
Legendary
Offline
Activity: 1162
Merit: 1000
DiabloMiner author
|
|
July 05, 2011, 10:51:27 AM |
|
Machine stats: Mac Mini with OSX 10.6.7 uname -a: Darwin zentrale1.local 10.7.0 Darwin Kernel Version 10.7.0: Sat Jan 29 15:17:16 PST 2011; root:xnu-1504.9.37~1/RELEASE_I386 i386 Apple OpenCL 1.0 (Dec 23 2010 17:30:26) Nvidia GeForce 9400 Update: Finish async networking, now everything is async. getwork, sendwork, and LP now use 1 thread per miner instance instead of 3 per GPU. Ever since this update, whenever long polling fails (because of network errors), there is a high chance that it won't pick up again (I've had >1h go by without a long polling message in debug mode). Pretty soon after the nonce saturation message, I get an OOM Exception (will paste it in here as soon as I get it again). Uodate: Behold, the frankenkernel. A mix of DiabloKernel and phatk.
Ever since this update, the hashing speed went from "mhash 2,2/2,2" to "mhash 1,4/1,4". The lates git updates don't change anything here. OSX's OpenCL implementation is a disaster. At any rate, the LP bug you're having doesn't make sense. It almost sounds like the pool ls intentionally not returning on LP. Look at the code yourself: if the LP either returns or errors out, it tells all the mining threads to go refresh their getworks, it then goes back to starting another LP. It almost sounds like you're mining on a pool that is down, and my miner (obviously) isn't shutting down. The only way you'd realistically run out of memory (since I tell Java to limit it to a small size) is if the sendwork queue fills up.
|
|
|
|
padrino
Legendary
Offline
Activity: 1428
Merit: 1000
https://www.bitworks.io
|
|
July 05, 2011, 10:57:45 AM |
|
8 GPUs in one box, I can't say exactly what causes it be it pool downtime, etc. but I was seeing OOM errors after a few hours, changed maximum heap from 16 to 32mb and haven't seen an OOM since.
|
|
|
|
zerokwel
|
|
July 05, 2011, 12:16:55 PM |
|
Hey all.
I decided to give Diablo miner a try as it has a option I like
Anyway every time I try and Mine using it I get a popup window saying Java has shit its self. I bet it is a Simple fix as well. Any ideas ?
Problem Event Name: BEX Application Name: java.exe Application Version: 6.0.260.3 Application Timestamp: 4dc11607 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: 6.1.7601.2.1.0.256.1 Locale ID: 2057
EDIT again its not working. Reinstalled it worked once everything fine. Closed it down to change a setting and its crashing again.
Found out what my Issue was which was causing the crashes I had app 2.3 . Installed 2.4 and running fine. getting 502mhash/s out of a 5770 + 6870. Dont know how this was causing the issue but nevermind Love the fact you can set all your miners away with 1 window.
|
|
|
|
DiabloD3 (OP)
Legendary
Offline
Activity: 1162
Merit: 1000
DiabloMiner author
|
|
July 05, 2011, 12:49:02 PM |
|
Hey all.
I decided to give Diablo miner a try as it has a option I like
Anyway every time I try and Mine using it I get a popup window saying Java has shit its self. I bet it is a Simple fix as well. Any ideas ?
Problem Event Name: BEX Application Name: java.exe Application Version: 6.0.260.3 Application Timestamp: 4dc11607 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: 6.1.7601.2.1.0.256.1 Locale ID: 2057
EDIT again its not working. Reinstalled it worked once everything fine. Closed it down to change a setting and its crashing again.
Found out what my Issue was which was causing the crashes I had app 2.3 . Installed 2.4 and running fine. getting 502mhash/s out of a 5770 + 6870. Dont know how this was causing the issue but nevermind Love the fact you can set all your miners away with 1 window. Ahh, weird. 2.3 should have just refused to work altogether. 2.3 does not support 68xx afaict, only 2.4.
|
|
|
|
DiabloD3 (OP)
Legendary
Offline
Activity: 1162
Merit: 1000
DiabloMiner author
|
|
July 05, 2011, 12:50:27 PM |
|
8 GPUs in one box, I can't say exactly what causes it be it pool downtime, etc. but I was seeing OOM errors after a few hours, changed maximum heap from 16 to 32mb and haven't seen an OOM since.
Yeah, I might make that change myself. 16 is just cutting it too thin.
|
|
|
|
ancow
|
|
July 05, 2011, 04:50:29 PM |
|
OSX's OpenCL implementation is a disaster.
Tell me about it. And it's an Nvidia card... Anyhow, at least with this configuration, there is a serious performance regression and you should know about it... At any rate, the LP bug you're having doesn't make sense. It almost sounds like the pool ls intentionally not returning on LP. Look at the code yourself: if the LP either returns or errors out, it tells all the mining threads to go refresh their getworks, it then goes back to starting another LP.
It almost sounds like you're mining on a pool that is down, and my miner (obviously) isn't shutting down. The only way you'd realistically run out of memory (since I tell Java to limit it to a small size) is if the sendwork queue fills up.
It's not intentional (and for reference, I usually mine at deepbit). Our phone somehow causes some interference with the DSL connection, triggering a reconnect on every phone call and therefore usually a new IP address. LP obviously can't work properly after this. Also note that I said there is a high chance it doesn't recover - sometimes it does. Anyway, I'm logging the current run, and will send you the complete log once the OOM occurs. So far there has been no longpolling debug message for >5 hours and there were some before the first reconnect. 1 share and three stales found so far. Here's the log up to this point (using swepool to see whether it would make a difference; so far it hasn't): [05.07.11 10:28:26] Started [05.07.11 10:28:26] Connecting to: http://swepool.net:8337/ [05.07.11 10:28:26] Using Apple OpenCL 1.0 (Dec 23 2010 17:30:26) [05.07.11 10:28:40] Added GeForce 9400 (#1) (2 CU, local work size of 128) [05.07.11 10:28:41] DEBUG: Enabling long poll support [05.07.11 10:35:42] DEBUG: Long poll returned [05.07.11 10:41:46] DEBUG: Long poll returned [05.07.11 13:07:06] DEBUG: Attempt 1 found on GeForce 9400 (#1) [05.07.11 13:07:07] Rejected block 1 found on GeForce 9400 (#1) [05.07.11 14:17:05] DEBUG: Attempt 2 found on GeForce 9400 (#1) [05.07.11 14:17:05] Rejected block 2 found on GeForce 9400 (#1) [05.07.11 15:54:03] DEBUG: Forcing getwork update due to nonce saturation [05.07.11 16:07:54] DEBUG: Attempt 3 found on GeForce 9400 (#1) [05.07.11 16:07:54] Accepted block 1 found on GeForce 9400 (#1) [05.07.11 17:35:49] DEBUG: Forcing getwork update due to nonce saturation [05.07.11 17:45:00] DEBUG: Attempt 4 found on GeForce 9400 (#1) [05.07.11 17:45:01] Rejected block 3 found on GeForce 9400 (#1) mhash 1,3/1,3 | a/r/hwe: 1/3/0 | ghash: 38,9 | fps: 21,0 (I just had a quick look at the code and long polling doesn't time out, which is obviously problematic. This would also fail if the network connection broke long enough for the long polling answer to fail to reach the client. jgarzik's cpuminer manages to revive the long polling connection after it is broken, so it can't be only the pool.)
|
BTC: 1GAHTMdBN4Yw3PU66sAmUBKSXy2qaq2SF4
|
|
|
DiabloD3 (OP)
Legendary
Offline
Activity: 1162
Merit: 1000
DiabloMiner author
|
|
July 06, 2011, 02:04:54 PM |
|
Update: Anyone having problems on pools with extremely broken LP (such as connections never returning), it should be fixed now (it times out after 10 minutes instead of never). I've also upped the stack size so you shouldn't see OOM errors when the queues get overloaded (at least, it will take much longer). Also, the bug with slushpool where multiple execution threads would not be brought up properly has also been fixed.
|
|
|
|
ancow
|
|
July 06, 2011, 02:59:52 PM |
|
Thanks a bunch, this should significantly reduce the amount of stales I get!
|
BTC: 1GAHTMdBN4Yw3PU66sAmUBKSXy2qaq2SF4
|
|
|
|