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

[phzi]

I'm amazed we're still doing all this tweaking blindly with so much guesswork and trial and error.  There's got to be a way to accurately and precisely measure what's possible/stable, based on the numbers alone.

Everybody's clocks have different tolerances.  It'd be easier just to say stock clocks and unedited .cl files, but fuckit; gotta try and get that extra 200Kh/s.  If you don't somebody else might.

But why would they have different tolerances if they're built exactly the same?  Are we talking differences in heat/humidity or something?

GPU ASIC qualities vary significantly.  Quite simply, no 2 GPUs will perform the same, and this is true even when you have cards with sequential serial numbers.

[bullus]

Is this fast  

Code:

sgminer 4.2.2 - Started: [2014-07-25 09:47:50] - [0 days 00:18:36]
--------------------------------------------------------------------------------
(5s):6.211M (avg):6.078Mh/s | A:2  R:0  HW:0  WU:0.083/m
ST: 0  SS: 0  NB: 30  LW: 1796  GF: 1  RF: 0
Connected to am02 x11 (stratum) diff 0.013 as user xxxx
Block: 204bc369...  Diff:119  Started: [10:06:00]  Best share: 1.946
--------------------------------------------------------------------------------
[P]ool management [G]PU management [S]ettings [D]isplay options [Q]uit
GPU 0:  72.0C  85%    | 6.211M/6.092Mh/s | R:  0.0% HW:0 WU:0.084/m xI: 56
--------------------------------------------------------------------------------
GPU 0: 6.2 / 6.1 Mh/s | A:140  R:0  HW:0  U:8.81/m  I:0  xI:56  rI:0
72.0 C  F: 85%  E: 1060 MHz  M: 1500 Mhz  V: 0.000V  A: 100%  P: 15%
Last initialised: [2014-07-25 09:47:48]
Thread 0: 3.1 Mh/s Enabled ALIVE
Thread 1: 3.1 Mh/s Enabled ALIVE

[E]nable  [D]isable  [R]estart GPU  [C]hange settings
[I]ntensity  E[x]perimental intensity  R[a]w Intensity
Or press any other key to continue

No, it can be faster  

Code:

sgminer 4.2.2 - Started: [2014-07-25 09:47:50] - [0 days 00:40:34]
--------------------------------------------------------------------------------
(5s):6.283M (avg):6.139Mh/s | A:3  R:0  HW:0  WU:0.087/m
ST: 0  SS: 0  NB: 90  LW: 3952  GF: 1  RF: 0
Connected to am02 x11 (stratum) diff 0.012 as user xxxx
Block: 683f926f...  Diff:694  Started: [10:28:02]  Best share: 1.946
--------------------------------------------------------------------------------
[P]ool management [G]PU management [S]ettings [D]isplay options [Q]uit
GPU 0:  73.0C  85%    | 6.283M/6.145Mh/s | R:  0.0% HW:0 WU:0.087/m xI: 65
--------------------------------------------------------------------------------
GPU 0: 6.3 / 6.1 Mh/s | A:341  R:0  HW:0  U:8.81/m  I:0  xI:65  rI:0
72.0 C  F: 85%  E: 1060 MHz  M: 1500 Mhz  V: 0.000V  A: 100%  P: 15%
Last initialised: [2014-07-25 09:47:48]
Thread 0: 3.1 Mh/s Enabled ALIVE
Thread 1: 3.1 Mh/s Enabled ALIVE

[E]nable  [D]isable  [R]estart GPU  [C]hange settings
[I]ntensity  E[x]perimental intensity  R[a]w Intensity
Or press any other key to continue

edit
I notice a hashrate drop to below 6mh/s and rise back again with xI=65.
Drop to xI=64 (6274kh/s) and is not dropping anymore.(10kh/s or so)

Edit2
As u can see the Accepted shares are calculate fine in sgminer, but the above value isn't.
Is it hard to change the above value to the value sgminer is showing with "G".?

edit3
Still stable after 3 hours.

Restarted and see if the other algos also be stable...6.346kh/s

Code:

sgminer 4.2.2 - Started: [2014-07-25 12:33:01] - [0 days 00:19:21]
--------------------------------------------------------------------------------
(5s):6.346M (avg):6.312Mh/s | A:2  R:0  HW:0  WU:0.083/m
ST: 0  SS: 0  NB: 22  LW: 1825  GF: 1  RF: 0
Connected to am02 x11 multi (stratum) diff 0.014 as user xxxx
Block: 2ef3e81b...  Diff:1.11K  Started: [12:51:30]  Best share: 1.397
--------------------------------------------------------------------------------
[P]ool management [G]PU management [S]ettings [D]isplay options [Q]uit
GPU 0:  74.0C  85%    | 6.352M/6.326Mh/s | R:  0.6% HW:0 WU:0.083/m xI: 64
--------------------------------------------------------------------------------
GPU 0: 6.4 / 6.3 Mh/s | A:174  R:1  HW:0  U:10.30/m  I:0  xI:64  rI:0
73.0 C  F: 85%  E: 1060 MHz  M: 1500 Mhz  V: 0.000V  A: 100%  P: 10%
Last initialised: [2014-07-25 12:33:00]
Thread 0: 3.2 Mh/s Enabled ALIVE
Thread 1: 3.2 Mh/s Enabled ALIVE

[E]nable  [D]isable  [R]estart GPU  [C]hange settings
[I]ntensity  E[x]perimental intensity  R[a]w Intensity
Or press any other key to continue

[rldep]

Thanks to all who answered :-)

--quiet|-q          Disable logging output, display status and errors
--log|-l <arg>      Interval in seconds between log output (default: 5)

no, -q is very silent ))

--log-file log.txt

It works!!! :-)

[rldep]

Your x11 hashrate will be dependent mostly on memclock speed, if you can hold at 1350, great, I can't though. 

I didn't know that x[n] algorythms are sensitive to memory speed.

Does it mean we must rise memory clock speed to achieve maximum hash rate at x11, x13, x15 ?

[platinum4]

edit
I notice a hashrate drop to below 6mh/s and rise back again with xI=65.
Drop to xI=64 (6274kh/s) and is not dropping anymore.(10kh/s or so)

Interesting.  I've noticed that xI:64 is a slight increase in performance on x11; but absolutely no increase in performance on x13/x14/x15.
Sticking with xI: 50 still provides 4M solid on 290X on x15.  I wanna see if we can get 4.1 somehow.  I'm still stuck at WU: .056/m maximum on x15.  Would be nice to increase it by .001-.003

[platinum4]

Your x11 hashrate will be dependent mostly on memclock speed, if you can hold at 1350, great, I can't though. 

I didn't know that x[n] algorythms are sensitive to memory speed.

Does it mean we must rise memory clock speed to achieve maximum hash rate at x11, x13, x15 ?

Raise that shit too high and you'll be holding down the power button and hard-resetting your hung rigs every 6 minutes.

[bullus]

Your x11 hashrate will be dependent mostly on memclock speed, if you can hold at 1350, great, I can't though.  

I didn't know that x[n] algorythms are sensitive to memory speed.

Does it mean we must rise memory clock speed to achieve maximum hash rate at x11, x13, x15 ?

Raise that shit too high and you'll be holding down the power button and hard-resetting your hung rigs every 6 minutes.

Not with my Asus dcII 290x..still stable with 1500mem   love my card

x15 testing and running 4080kh/s

4100kh/s  

Code:

sgminer 4.2.2 - Started: [2014-07-25 12:33:01] - [0 days 01:07:01]
--------------------------------------------------------------------------------
(5s):4.104M (avg):4.080Mh/s | A:3  R:0  HW:0  WU:0.059/m
ST: 0  SS: 1  NB: 107  LW: 4080  GF: 0  RF: 0
Connected to am02 x15 (stratum) diff 0.007 as user bullus.3
Block: 40a76a3e...  Diff:88  Started: [13:39:54]  Best share: 26
--------------------------------------------------------------------------------
[P]ool management [G]PU management [S]ettings [D]isplay options [Q]uit
GPU 0:  74.0C  85%    | 4.100M/4.084Mh/s | R:  0.6% HW:0 WU:0.060/m xI: 51
--------------------------------------------------------------------------------
GPU 0: 4.1 / 4.1 Mh/s | A:345  R:2  HW:0  U:8.51/m  I:0  xI:51  rI:0
73.0 C  F: 85%  E: 1060 MHz  M: 1500 Mhz  V: 0.000V  A: 100%  P: 10%
Last initialised: [2014-07-25 12:57:15]
Thread 0: 2.0 Mh/s Enabled ALIVE
Thread 1: 2.0 Mh/s Enabled ALIVE

[E]nable  [D]isable  [R]estart GPU  [C]hange settings
[I]ntensity  E[x]perimental intensity  R[a]w Intensity
Or press any other key to continue

edit
Final speeds Asus DCII 290X
@1060/1500 stock Voltage (ratio 1:1.415)
X11=6350
X13=4650
X15=4100

@1060/1425
X11=6305
X13=4585---> not much difference but with mem@1500 it does make difference
X15=4005---> not much difference

@1060/1350
X11=6200
X13=4585
X15=4000

@1050/1350
X11=6150
X13=4575
X15=3970

@1000/1415 (ratio 1:1.415) -> but can lower my voltage..let's see if hash/watt is better.
X11=6170
X13=4560
X15=3990

Code:

{
"profiles" : [
	{
		"name" : "x11",
		"algorithm" : "darkcoin-mod",
		"xintensity" : "64",
		"gpu-threads" : "2",
		"worksize": "64"
	},
	{
		"name" : "x13",
		"algorithm" : "marucoin-mod",
		"xintensity" : "51",
		"gpu-threads" : "2",
		"worksize": "64"
	},
	{
		"name" : "x15",
		"algorithm" : "bitblock",
		"xintensity" : "51",
		"gpu-threads" : "2",
		"worksize": "64"
	},
	{
		"name" : "nist5",
		"algorithm" : "talkcoin-mod",
		"intensity" : "16",
		"gpu-threads" : "2",
		"worksize": "64"
	}		
],
"default-profile": "x15",
"hamsi-expand-big" : "7",
"shaders" : "2816",
"gpu-fan" : "85-100",
"gpu-powertune" : "10",
"gpu-vddc" : "0",
"auto-fan" : true,
"failover-only" : true,
"expiry" : "1",
"gpu-dyninterval" : "7",
"hotplug" : "5",
"log" : "5",
"queue" : "0",
"scan-time" : "1",
"temp-hysteresis" : "2",
"shares" : "0",
"no-submit-stale" : true,
"no-restart" : true,
"failover-switch-delay" : "30",
"show-coindiff" : true,
"remove-disabled" : true,
"extranonce-subscription" : true
}

[rldep]

Raise that shit too high and you'll be holding down the power button and hard-resetting your hung rigs every 6 minutes.

You are prikolist! :-))
(Prikolist is a man who like jokes)

[Singman33]

I'm testing latest GIT build of v5, and I have a few drop in hashrate compared to dedicated miners :
This is for a R9 280X Tri-X
Scrypt : v5 = 755 Kh/s, ref = 760 Kh/s, it's OK
X11 : v5 = 3,2 Mh/s, ref = 3,5 Mh/s, it's less than using sph-sgminer_x11mod (https://github.com/lasybear/sph-sgminer_x11mod)
X13 : v5 = can't compile .cl files at start, should be a bug introduced recently, ref = 2,45 Mh/s (same as X11)
Keccak : v5 = 380 Mh/s, ref = 370 Mh/s, OK
Nist5 : v5 = 7 Mh/s, ref = 13 Mh/s, it's VERY BAD DROP !

Maybe I'm missing some important tweaking (each miner have his own configuration when using different kernels) but using V5 drop hashrate on most cases.
If you have suggestions (config files), post them.

Thanks.

[Accordus]

I'm testing latest GIT build of v5, and I have a few drop in hashrate compared to dedicated miners :
This is for a R9 280X Tri-X
Scrypt : v5 = 755 Kh/s, ref = 760 Kh/s, it's OK
X11 : v5 = 3,2 Mh/s, ref = 3,5 Mh/s, it's less than using sph-sgminer_x11mod (https://github.com/lasybear/sph-sgminer_x11mod)
X13 : v5 = can't compile .cl files at start, should be a bug introduced recently, ref = 2,45 Mh/s (same as X11)
Keccak : v5 = 380 Mh/s, ref = 370 Mh/s, OK
Nist5 : v5 = 7 Mh/s, ref = 13 Mh/s, it's VERY BAD DROP !

Maybe I'm missing some important tweaking (each miner have his own configuration when using different kernels) but using V5 drop hashrate on most cases.
If you have suggestions (config files), post them.

Thanks.

Hmmmmm. I had much higher speeds.
See this post of my collected miners and configs: https://bitcointalk.org/index.php?topic=514242.msg7982772#msg7982772
My script was 725 kH/s. Other speeds:
4.3 MHs/card       (= x5.9)   for X11,  @1480W for the total 8 cards
3.05 MHs/card     (= x4.2)   for X13,  @1480W for the total 8 cards
2.955 MHs/card   (= x4.1)   for X15,  @1560W for the total 8 cards
413 MHs/card      (= x17.9)   for NIST5,  @1650W for the total 8 cards
6.2 MHs/card      (= x6.2)   for FRESH,  @1350W for the total 8 cards
0.375 kHs(!)/card   (= x0.00051)   for Cryptonote,  @1650W for the total 8 cards

[Singman33]

Hmmmmm. I had much higher speeds.
See this post of my collected miners and configs: https://bitcointalk.org/index.php?topic=514242.msg7982772#msg7982772

No, u dont understand. I already have better hashrate with custom or dedicated miners, I just want to have the best hashrate (and if possible the same) with sgminer-dev-v5. Look like this version is missing a lot of optimizations from others miners.

[badman74]

Hmmmmm. I had much higher speeds.
See this post of my collected miners and configs: https://bitcointalk.org/index.php?topic=514242.msg7982772#msg7982772

No, u dont understand. I already have better hashrate with custom or dedicated miners, I just want to have the best hashrate (and if possible the same) with sgminer-dev-v5. Look like this version is missing a lot of optimizations from others miners.

try looking though this and see if any of the changes help you
https://github.com/sgminer-dev/sgminer/issues/347

[bullus]

Which binaries build is using the optimized x11 bin? Or do I still need the "higher bin" from the other miner?

[Accordus]

Hmmmmm. I had much higher speeds.
See this post of my collected miners and configs: https://bitcointalk.org/index.php?topic=514242.msg7982772#msg7982772

No, u dont understand. I already have better hashrate with custom or dedicated miners, I just want to have the best hashrate (and if possible the same) with sgminer-dev-v5. Look like this version is missing a lot of optimizations from others miners.

Ahhh, OK, I see.

[badman74]

Which binaries build is using the optimized x11 bin? Or do I still need the "higher bin" from the other miner?

try these
https://github.com/badman74/Binaries/archive/V5_0_mod.zip

[ebliever]

edit
I notice a hashrate drop to below 6mh/s and rise back again with xI=65.
Drop to xI=64 (6274kh/s) and is not dropping anymore.(10kh/s or so)

Edit2
As u can see the Accepted shares are calculate fine in sgminer, but the above value isn't.
Is it hard to change the above value to the value sgminer is showing with "G".?

edit3
Still stable after 3 hours.

Restarted and see if the other algos also be stable...6.346kh/s

Sigh... every time I get close someone has to leap farther ahead.  

Looks like you encountered the same thing I did with the x-65/x-64 threshold. I'm running x-68 since last night, averaging 5.96 Mhash/GPU across 4 290X's over 12 hours. Stable apart from a renter trying to load dead pools for a while, LOL. I get up to around 6.17 Mhash on one of the GPU's but they keep dropping off and don't hold consistently at the highest speeds. But I'm still at 1020/1350 on the clocks so that explains much of my shortfall, I'm sure.

On scrypt I had all 4 of my GPU's individually tuned to four different engine and memory speeds for the fastest possible stable speed. I'm going to try the latest builds and settings you and badman74 helpfully shared, and then see if I can push individual GPU's higher. Thanks again!

EDIT: Dumb question - I've never messed with voltage settings before as I didn't want to fry anything. But I see in bullus' profile that he has vddc set to 0. Does that mean "go with the default setting"? As I understand it the reference voltage on a 290X is 1.0 volt. So if I wanted to try cutting that by 10%, should I put "0.9" in the vddc line, or "-0.1"?  Reading the docs would have me trying the former, but bullus' example makes me think the latter is right.

[KiloWatts]

Keep it up guys - people like you ensure GPU-mining will never die 

[bullus]

edit
I notice a hashrate drop to below 6mh/s and rise back again with xI=65.
Drop to xI=64 (6274kh/s) and is not dropping anymore.(10kh/s or so)

Edit2
As u can see the Accepted shares are calculate fine in sgminer, but the above value isn't.
Is it hard to change the above value to the value sgminer is showing with "G".?

edit3
Still stable after 3 hours.

Restarted and see if the other algos also be stable...6.346kh/s

Sigh... every time I get close someone has to leap farther ahead.  

Looks like you encountered the same thing I did with the x-65/x-64 threshold. I'm running x-68 since last night, averaging 5.96 Mhash/GPU across 4 290X's over 12 hours. Stable apart from a renter trying to load dead pools for a while, LOL. I get up to around 6.17 Mhash on one of the GPU's but they keep dropping off and don't hold consistently at the highest speeds. But I'm still at 1020/1350 on the clocks so that explains much of my shortfall, I'm sure.

On scrypt I had all 4 of my GPU's individually tuned to four different engine and memory speeds for the fastest possible stable speed. I'm going to try the latest builds and settings you and badman74 helpfully shared, and then see if I can push individual GPU's higher. Thanks again!

EDIT: Dumb question - I've never messed with voltage settings before as I didn't want to fry anything. But I see in bullus' profile that he has vddc set to 0. Does that mean "go with the default setting"? As I understand it the reference voltage on a 290X is 1.0 volt. So if I wanted to try cutting that by 10%, should I put "0.9" in the vddc line, or "-0.1"?  Reading the docs would have me trying the former, but bullus' example makes me think the latter is right.

That's funny cause 1 renter told I was selling a 4.2mh/s rig but I have 9.8mh/s.(4x7950)
At the graph at betarigs u see fluctuating hashrates but can't see that at sgminer like i saw it with xI=65.(290x)
Average is still 9.6mh/s, a bit lower than normal, but no way 4.2mh/s. Is there a tweak for flat hashrates? Decreasing intensity?



gpu-vddc is set at 0 ,because I set it with msi afterburner for testing. At github guide u should set it like "1.000"

[bullus]

Which binaries build is using the optimized x11 bin? Or do I still need the "higher bin" from the other miner?

try these
https://github.com/badman74/Binaries/archive/V5_0_mod.zip

Will try those, thanks.

[platinum4]

Keep it up guys - people like you ensure GPU-mining will never die 

Not dead; a hobby right now, it seems.