[ANN] sgminer v5 - optimized X11/X13/NeoScrypt/Lyra2RE/etc. kernel-switch miner - page 224.

nem2k

newbie

Activity: 27

Merit: 0

is it possible to integrate the lasybear x11 and x13 builds of sgminer:

https://bitcointalksearch.org/topic/annx11x13-x11-darkcoinx13-marucoin-miner-based-on-sph-sgminer-623409

id love to jump on the multi algo mining train, but the traditional x11/13 miner builds have been much slower than the lasybear option

platinum4

sr. member

Activity: 547

Merit: 250

Quote from: nicehashdev on June 19, 2014, 08:15:36 AM

NiceHash has all 4xxx ports alive - always, but all except one will drop connection as bad auth. As long as the switching on sgminer isn't performed too often it should work fine.

That's why I posted the logs. It's obvious the switching in sgminer occurs on the scales of 8, 15, and possibly 30 seconds. You've just been saying it should be longer...

Ok ok I see, you're proposing a change... post a github issue?

Code:

[09:02:55] Stratum connection to nh x11 multi interrupted
[09:02:56] nh x11 multi not responding!
[09:02:56] Switching to nh keccak multi
[09:02:56] nh keccak multi difficulty changed to 512
[09:03:05] Initialising kernel maxcoin.cl with bitalign, unpatched BFI, nfactor 10, n 1024
[09:03:05] Initialising kernel maxcoin.cl with bitalign, unpatched BFI, nfactor 10, n 1024
[09:03:18] Accepted Coin 16114 Diff 2.914/512 GPU 0 at nh keccak multi 
[09:03:19] Accepted Coin 16114 Diff 3.370/512 GPU 0 at nh keccak multi 
[09:03:34] Accepted Coin 16114 Diff 12/512 GPU 1 at nh keccak multi 
[09:03:34] Accepted Coin 16114 Diff 4.309/512 GPU 0 at nh keccak multi 
[09:03:50] Accepted Coin 16114 Diff 3.061/512 GPU 0 at nh keccak multi 
[09:03:54] Accepted Coin 15797 Diff 6.801/512 GPU 0 at nh keccak multi 
[09:03:54] nh keccak multi difficulty changed to 1024
[09:03:56] Accepted Coin 15797 Diff 6.801/1.02K GPU 0 at nh keccak multi 
[09:03:58] nh x11 multi alive, testing stability
[09:04:11] Accepted Coin 15797 Diff 4.549/1.02K GPU 1 at nh keccak multi 
[09:04:18] nh x11 multi stable for 30 seconds
[09:04:18] Switching to nh x11 multi
[09:04:19] Initialising kernel darkcoin-mod.cl with bitalign, unpatched BFI, nfactor 10, n 1024
[09:04:19] Initialising kernel darkcoin-mod.cl with bitalign, unpatched BFI, nfactor 10, n 1024

Is this what you're referring to, with the 30 seconds?

EDIT: Here is an example with failover-switch-delay : 90, notice how in 40 seconds, the nh x11 stratum drops, we wouldn't see this with the 30 second setting.

Code:

[09:12:45] Accepted Coin 15818 Diff 4.802/1.02K GPU 1 at nh keccak multi 
[09:13:04] Accepted Coin 15818 Diff 4.104/1.02K GPU 1 at nh keccak multi 
[09:13:07] nh x11 multi alive, testing stability
[09:13:10] Accepted Coin 15818 Diff 159/1.02K GPU 0 at nh keccak multi 
[09:13:19] Accepted Coin 15840 Diff 23/1.02K GPU 1 at nh keccak multi 
[09:13:27] Accepted Coin 15840 Diff 7.130/1.02K GPU 0 at nh keccak multi 
[09:13:50] Stratum connection to nh x11 multi interrupted
[09:13:57] Accepted Coin 15840 Diff 457/1.02K GPU 1 at nh keccak multi 
[09:14:02] Accepted Coin 15809 Diff 13/1.02K GPU 0 at nh keccak multi 
[09:14:05] Accepted Coin 15809 Diff 13/1.02K GPU 0 at nh keccak multi 
[09:14:06] Accepted Coin 15809 Diff 416/1.02K GPU 1 at nh keccak multi 
[09:14:34] Accepted Coin 15804 Diff 8.340/1.02K GPU 1 at nh keccak multi 
[09:14:43] Stratum connection to nh keccak multi interrupted
[09:14:44] nh keccak multi not responding!
[09:14:44] Switching to tmb x13 multiport
[09:14:44] tmb x13 multiport difficulty changed to 0.005
[09:14:53] Initialising kernel marucoin-mod.cl with bitalign, unpatched BFI, nfactor 10, n 1024
[09:14:53] Initialising kernel marucoin-mod.cl with bitalign, unpatched BFI, nfactor 10, n 1024
[09:14:54] nh x11 multi alive, testing stability
[09:15:09] Accepted Coin 80 Diff 0.018/0.005 GPU 1 at tmb x13 multiport 
[09:15:14] Accepted Coin 80 Diff 0.006/0.005 GPU 1 at tmb x13 multiport 
[09:15:16] Accepted Coin 80 Diff 0.006/0.005 GPU 0 at tmb x13 multiport 
[09:15:17] Accepted Coin 80 Diff 0.010/0.005 GPU 1 at tmb x13 multiport 
[09:15:20] Accepted Coin 80 Diff 0.015/0.005 GPU 1 at tmb x13 multiport 
[09:15:20] Accepted Coin 80 Diff 0.007/0.005 GPU 0 at tmb x13 multiport 
[09:15:22] Accepted Coin 80 Diff 0.019/0.005 GPU 0 at tmb x13 multiport 
[09:15:26] Accepted Coin 80 Diff 0.009/0.005 GPU 0 at tmb x13 multiport 
[09:15:33] Accepted Coin 80 Diff 0.006/0.005 GPU 0 at tmb x13 multiport 
[09:15:36] Accepted Coin 80 Diff 0.008/0.005 GPU 1 at tmb x13 multiport 
[09:15:37] Accepted Coin 80 Diff 0.015/0.005 GPU 1 at tmb x13 multiport 
[09:15:38] Accepted Coin 80 Diff 0.019/0.005 GPU 0 at tmb x13 multiport 
[09:15:38] Accepted Coin 80 Diff 0.006/0.005 GPU 1 at tmb x13 multiport 
[09:15:42] Accepted Coin 80 Diff 0.021/0.005 GPU 0 at tmb x13 multiport 
[09:15:43] Accepted Coin 80 Diff 0.022/0.005 GPU 0 at tmb x13 multiport 
[09:15:44] Accepted Coin 80 Diff 0.009/0.005 GPU 1 at tmb x13 multiport 
[09:15:44] Accepted Coin 80 Diff 0.006/0.005 GPU 1 at tmb x13 multiport 
[09:15:46] Accepted Coin 80 Diff 0.619/0.005 GPU 1 at tmb x13 multiport 
[09:15:49] Accepted Coin 80 Diff 0.007/0.005 GPU 0 at tmb x13 multiport 
[09:15:50] Accepted Coin 80 Diff 0.007/0.005 GPU 0 at tmb x13 multiport 
[09:15:51] tmb x13 multiport stale share detected, submitting (user)
[09:15:51] Accepted Coin 80 Diff 0.035/0.005 GPU 0 at tmb x13 multiport 
[09:15:55] Accepted Coin 81 Diff 0.011/0.005 GPU 1 at tmb x13 multiport 
[09:15:57] tmb x13 multiport difficulty changed to 0.018
[09:16:28] nh x11 multi stable for 90 seconds
[09:16:28] Switching to nh x11 multi
[09:16:28] Initialising kernel darkcoin-mod.cl with bitalign, unpatched BFI, nfactor 10, n 1024
[09:16:28] Initialising kernel darkcoin-mod.cl with bitalign, unpatched BFI, nfactor 10, n 1024
[09:16:33] Stratum connection to tmb x13 multiport interrupted
[09:17:09] Accepted Coin 606 Diff 0.058/0.050 GPU 1 at nh x11 multi

See, I think it's actually something inside the miner that needs to be adjusted, because look at the harsh algo switch in around 20 seconds total elapsed time.

Code:

[09:14:43] Stratum connection to nh keccak multi interrupted
[09:14:44] nh keccak multi not responding!
[09:14:44] Switching to tmb x13 multiport
[09:14:44] tmb x13 multiport difficulty changed to 0.005
[09:14:53] Initialising kernel marucoin-mod.cl with bitalign, unpatched BFI, nfactor 10, n 1024
[09:14:53] Initialising kernel marucoin-mod.cl with bitalign, unpatched BFI, nfactor 10, n 1024
[09:14:54] nh x11 multi alive, testing stability

I can't tell if this commit is useful, or not, but I propose we use the idle state prior to initialzing a new algo kernel.

https://github.com/sgminer-dev/sgminer/commit/59bbe1b636cdb4a492e1ba15bc43ea9f20338ab1

nicehashdev

sr. member

Activity: 280

Merit: 250

NiceHash has all 4xxx ports alive - always, but all except one will drop connection as bad auth. As long as the switching on sgminer isn't performed too often it should work fine.

platinum4

sr. member

Activity: 547

Merit: 250

Quote from: nicehashdev on June 19, 2014, 07:01:26 AM

Also it should be considered that detection of dead pool should work flawlessly. Because now, rig is disconnected upon auth (we don't get f and c parameters sooner than with password, so we can't know if rig should be disconnected or not). And yes, there were issues (idlebug) and these issues were solved, but solution might not be perfect.

Code:

[07:23:22] Accepted Coin 836 Diff 0.103/0.040 GPU 0 at nh x11 multi 
[07:23:40] Accepted Coin 4368 Diff 0.044/0.040 GPU 1 at nh x11 multi 
[07:23:55] Accepted Coin 4368 Diff 0.207/0.040 GPU 0 at nh x11 multi 
[07:24:00] Stratum connection to nh x11 multi interrupted
[07:24:01] nh x11 multi not responding!
[07:24:01] Switching to nh keccak multi
[07:24:02] nh keccak multi not responding!
[07:24:02] Switching to tmb x13 multiport
[07:24:03] tmb x13 multiport not responding!
[07:24:03] Switching to tmb x13 multi east2
[07:24:03] tmb x13 multi east2 not responding!
[07:24:03] Switching to tmb x13 multi west
[07:24:05] tmb x13 multi west not responding!
[07:24:05] Switching to tmb x13 multi eu2
[07:24:06] tmb x13 multi eu2 not responding!
[07:24:06] Switching to tmb x11 multiport
[07:24:10] Waiting for work to be available from pools.
[07:24:10] Setting GPUs to idle performance.                <--- successful implementation of if have_ADL set to idle patch!!
[07:24:15] Work available from pools, resuming.
[07:24:16] Accepted Coin 836 Diff 0.008/0.005 GPU 0 at tmb x11 multiport 
[07:24:18] Accepted Coin 836 Diff 0.009/0.005 GPU 0 at tmb x11 multiport 
[07:24:20] tmb x11 multiport stale share detected, submitting (user)

Changing the failover-switch-delay parameter in the config file obviously won't fix the stratum drop. Is there a softer way to do this?

Look at the timestamps... you're actually at half of 30 seconds, and closer to 15 seconds. Is there another way to kill your stratum softer?

Here's how harshly it flips back when it's alive...

Code:

[07:33:35] Accepted Coin 4188 Diff 0.021/0.017 GPU 1 at tmb x11 multi 
[07:33:41] Stratum connection to tmb x11 multi interrupted
[07:33:42] tmb x11 multi not responding!
[07:33:42] Switching to nh x11 multi
[07:33:48] Stratum connection to nh x11 multi interrupted
[07:33:49] nh x11 multi difficulty changed to 0.040
[07:33:58] Accepted Coin 730 Diff 0.097/0.040 GPU 1 at nh x11 multi

8 seconds between a disconnect from tmb and switch back to nh. 1/4 as long as 30s.

nicehashdev

sr. member

Activity: 280

Merit: 250

Also it should be considered that detection of dead pool should work flawlessly. Because now, rig is disconnected upon auth (we don't get f and c parameters sooner than with password, so we can't know if rig should be disconnected or not). And yes, there were issues (idlebug) and these issues were solved, but solution might not be perfect.

platinum4

sr. member

Activity: 547

Merit: 250

Quote from: nicehashdev on June 19, 2014, 04:20:33 AM

I suggest one more extra feature to be added, because current model can cause many rapid switches and as it appears sgminer needs some time to reach full speed when doing switch. We only drop connections every 15 minutes on NiceHash, but due to priority list in sgminer, sgminer will swap to higher priority pool as soon as that one is "stable" for 30 seconds? I suggest to make this value configurable in pool section. Use high values for multialgo pools and default (30) for backup pools and it should work fine (with no rapid switches). This extra feature is only 5 minute of work, so shouldn't be a big deal.

I've extended failover-switch-delay to 300s right now. I will report back if this mitigates SICK->DEAD. Somewhere in the work being served, a thread goes lazy and always shits on the last GPU of a rig.

As far as extending SICK timer, adding in a GPU idle patch for when all pools are down and no work is available, and pthread_testcancel() all still result in rig crashing occasionally. Two others are rock solid and would have 3 days up time if the city hadn't cut the power for a sec to work on the traffic lights I guess.

nicehashdev

sr. member

Activity: 280

Merit: 250

I suggest one more extra feature to be added, because current model can cause many rapid switches and as it appears sgminer needs some time to reach full speed when doing switch. We only drop connections every 15 minutes on NiceHash, but due to priority list in sgminer, sgminer will swap to higher priority pool as soon as that one is "stable" for 30 seconds? I suggest to make this value configurable in pool section. Use high values for multialgo pools and default (30) for backup pools and it should work fine (with no rapid switches). This extra feature is only 5 minute of work, so shouldn't be a big deal.

Ghost__

newbie

Activity: 8

Merit: 0

grivad

newbie

Activity: 3

Merit: 0

Some help here would be greatly appreciated!

Finally got the multi-algo switching working right after much hair pulling, once I realized that a new CGWatcher version was out (auto update didn't pick it up).

However, when sgminer runs with the multi-algo config (both 6/11 and 6/16 builds), after a few minutes, the miner dies saying that GPU0's status is UNKNOWN. Using a config for a specific algo pool doesn't have this problem. Starting sgminer via CGWatcher or command line causes the same issue.

Config:

Quote

{
"pools" : [
{ "name" : "NiceHash_Scrypt",
"url" : "stratum+tcp://stratum.nicehash.com:4333",
"user" : "1Jp4cZcpCjQrqkAaPisktfVgU1uHV7uVQc",
"pass" : "f0=2.4;f2=1.1;f3=10.5;f4=7.5;f5=1200;c0=2.4;c2=2.4;c3=1.4;c4=1.4;c5=1.7",
"pool-algorithm" : "zuikkis",
"pool-nfactor" : "10",
"pool-xintensity" : "400",
"pool-gpu-engine" : "1000-1000",
"pool-gpu-memclock" : "1300-1300",
"pool-gpu-threads" : "1",
"pool-thread-concurrency" : "25601"
},
{ "name" : "NiceHash_Scrypt-N",
"url" : "stratum+tcp://stratum.nicehash.com:4335",
"user" : "1Jp4cZcpCjQrqkAaPisktfVgU1uHV7uVQc",
"pass" : "f0=2.4;f2=1.1;f3=10.5;f4=7.5;f5=1200;c0=2.4;c2=2.4;c3=1.4;c4=1.4;c5=1.7",
"pool-algorithm" : "zuikkis",
"pool-nfactor" : "11",
"pool-xintensity" : "400",
"pool-gpu-engine" : "1000-1000",
"pool-gpu-memclock" : "1300-1300",
"pool-gpu-threads" : "1",
"pool-thread-concurrency" : "20481"
},
{ "name" : "NiceHash_x11",
"url" : "stratum+tcp://stratum.nicehash.com:4336",
"user" : "1Jp4cZcpCjQrqkAaPisktfVgU1uHV7uVQc",
"pass" : "f0=2.4;f2=1.1;f3=10.5;f4=7.5;f5=1200;c0=2.4;c2=2.4;c3=1.4;c4=1.4;c5=1.7",
"pool-algorithm" : "darkcoin-mod",
"pool-nfactor" : "10",
"pool-xintensity" : "400",
"pool-gpu-engine" : "1000-1000",
"pool-gpu-memclock" : "1300-1300",
"pool-gpu-threads" : "4",
"pool-thread-concurrency" : "25601"
},
{ "name" : "NiceHash_x13",
"url" : "stratum+tcp://stratum.nicehash.com:4337",
"user" : "1Jp4cZcpCjQrqkAaPisktfVgU1uHV7uVQc",
"pass" : "f0=2.4;f2=1.1;f3=10.5;f4=7.5;f5=1200;c0=2.4;c2=2.4;c3=1.4;c4=1.4;c5=1.7",
"pool-algorithm" : "marucoin-mod",
"pool-nfactor" : "10",
"pool-xintensity" : "400",
"pool-gpu-engine" : "1000-1000",
"pool-gpu-memclock" : "1300-1300",
"pool-gpu-threads" : "4",
"pool-thread-concurrency" : "25601"
},
{ "name" : "NiceHash_Keccak",
"url" : "stratum+tcp://stratum.nicehash.com:4338",
"user" : "1Jp4cZcpCjQrqkAaPisktfVgU1uHV7uVQc",
"pass" : "f0=2.4;f2=1.1;f3=10.5;f4=7.5;f5=1200;c0=2.4;c2=2.4;c3=1.4;c4=1.4;c5=1.7",
"pool-algorithm" : "maxcoin",
"pool-nfactor" : "10",
"pool-xintensity" : "300",
"pool-gpu-engine" : "1000-1000",
"pool-gpu-memclock" : "1300-1300",
"pool-gpu-threads" : "1",
"pool-thread-concurrency" : "25601"
}
],

"no-submit-stale" : true,
"failover-only" : true,
"failover-switch-delay" : "30",
"no-pool-disable" : true,

"api-allow" : "W:127.0.0.1",
"api-description" : "m1-multi",
"api-listen" : true,
"api-port" : "4096",

"remove-disabled" : true,

"temp-cutoff" : "95",
"temp-overheat" : "87",
"temp-target" : "80",
"gpu-fan" : "50-100",
"auto-fan" : true,

"lookup-gap" : "2",
"worksize" : "128",
"shaders" : "2560",
"queue" : "1",
"expiry" : "1",
"scan-time" : "1",
"timeout" : "60",
"log" : "5",
"temp-hysteresis" : "3",

"kernel-path" : "/usr/local/bin"
}

Any help greatly appreciated!
Posted from Bitcointa.lk - #A7J23pvPIVJKFWfH

winkler

newbie

Activity: 8

Merit: 0

I just followed the directions on a new Ubuntu Trusty headless install, AMD catalyst 14.6 beta drivers and keep getting a Bus error (core dumped) message. Any ideas? Seems to be happening with all the different builds...

compile

newbie

Activity: 51

Merit: 0

Anyone have a 7950 or 270x config? The link on the site isn't working.

shacky

full member

Activity: 158

Merit: 100

kenshirothefist

sr. member

Activity: 457

Merit: 273

Quote from: yotadrivers on June 18, 2014, 12:38:08 PM

hello! im using sgminer_x11x13mod_03_06_2014 mod, how to switch X13 and X11 on fly?

Use newer sgminer https://nicehash.com/software/#sgminer and take a look at this: https://nicehash.com/multialgo/#finetuning

yotadrivers

member

Activity: 114

Merit: 10

hello! im using sgminer_x11x13mod_03_06_2014 mod, how to switch X13 and X11 on fly?

landy68888

newbie

Activity: 9

Merit: 0

Do not support the 5xxx 6xxx card

mannyg

full member

Activity: 142

Merit: 101

2+ days of mining X11 only. Rig is 4x MSI 280x Gamings open aluminium frame, 4GB ram on ASRock Pro BTC board.

Core config:
"vectors" : "1",
"intensity" : "17",
"worksize" : "256",
"lookup-gap" : "2",
"shaders" : "2048",
"nfactor" : "10",
"thread-concurrency" : "8193",
"gpu-engine" : "1125",
"gpu-memclock" : "1500",
"gpu-powertune" : "-20",
"temp-target" : "68",
"gpu-threads" : "2",

mannyg

full member

Activity: 142

Merit: 101

Quote from: LeetPoolsOP on June 17, 2014, 03:42:44 PM

Just thought I should contribute to this thread here.

Here is my config for Trademybit multi algo x11/x13

I was getting 2-3 min delays between algo switches but the problem was resolved by making the port that closes the FIRST pool in the list, and leaving the port that is ALWAYS OPEN as the second one.

The logic here (this will probably be addressed in later releases but for the time being its smooth): First pool opens and closes. If x11 is more profitable, x13 shuts, insta fail-over to x11, when x13 is more profitable, pool port opens up, since its first pool, it will go back to it from x11.

For what ever reason the miner didn't like the fact that both could be closed or overlap... no idea why... but it works now.

This is using Catalyst 14.4

sgminer.conf =

This is for my 290 @ stock clocks using this build:
https://mega.co.nz/#!Ld5QASiI!cS1ETIXI_-jdvx6ZW93rD-IYY91zcWG4B24ZbYs2Ik8
x11 = 4Mhs
x13 = 3Mhs
http://pastebin.com/eNbaahbY

Heres one for my 5x 280x Rig using the same build in the megalink:
x11 = 3.1-3.2
x13 = 2.2-2.4 (Cards are beat to shit from scrypt mining their hashes swing up and down)
http://pastebin.com/FpvMKVBc

Credit goes to Pelk for the pool list configuration at #pool from Trademybit IRC.

Hope this will help.

Why do you have your 290 shaders set to 2048... they have 2560 shaders.

Also, your 280x's might be going up and down because of the 18 intensity you have them set at. I run mine at 17 and they stay solid. These 280x's are 3-5 months old and did most of their time mining scrypt and scrypt-n. Try the 8193 thread concurrency with them as well you may see improvement.

platinum4

sr. member

Activity: 547

Merit: 250

No lie, shit is now extremely stable on well-configured rigs with no mixmatches of card types.

Still getting a SICK -> DEAD on a config with 2x Sapphire R9 290X + 1x Sapphire Tri-OC R9 290X.

Been flipping between x11/x13 flawless, haven't witnessed keccak or scrypt yet as I don't think they've been served as most profitable.

badman74

hero member

Activity: 658

Merit: 500