SRBMiner Cryptonight AMD GPU Miner V1.9.3 - native algo switching

[bruslie]

hy @doktor83 ,

gpu watchdog don’t work for me anymore since 1.6.8

on gpu fail miner don’t use reboot file , he try to restart miner and stuck ... can you plz take a look ?

https://imgur.com/a/LOmcyIb

https://imgur.com/a/EK5Fjcj

[1714981165]

1714981165

[1714981165]

1714981165

[1714981165]

1714981165

[1714981165]

1714981165

[1714981165]

1714981165

[1714981165]

1714981165

[vmozara]

If the miner is stuck when trying to reatart itself, and your rig can't restart either from windows, it means ypu gpu hard crashed and nothing can save it except haed reset of the pc (removing the power). Try to reduce overclock settings for that card

[doktor83]

hy @doktor83 ,

gpu watchdog don’t work for me anymore since 1.6.8

on gpu fail miner don’t use reboot file , he try to restart miner and stuck ... can you plz take a look ?

https://imgur.com/a/LOmcyIb

https://imgur.com/a/EK5Fjcj

I just checked, it is working :

Code:

{
"cryptonight_type" : "normalv8",
"intensity" : 0,
"double_threads" : true,
"reboot_script_gpu_watchdog" : "run-me-watchdog.bat"
}

Code:

[2018-10-28 07:45:35] watchdog: GPU0 [BUS: 3] hashing speed is 0 H/S
[2018-10-28 07:45:40] json_receive: {"jsonrpc":"2.0","method":"job","params":{"blob":"090994b6d5de053c408e072491052f414699215faf87c20ebd7d0b40ba3dd729a4a499cdbd6480000000005975b98061a7aec57fd05f418a5d774a78ea762a1abf6639fdd5b00109919f1205","job_id":"474","target":"cf8b0000"}}
[2018-10-28 07:45:40] pool_have_job: Pool sent a new job (ID: 474)
[2018-10-28 07:45:46] watchdog: GPU0 [BUS: 3] hashing speed is 0 H/S
[2018-10-28 07:45:57] watchdog: GPU0 [BUS: 3] hashing speed is 0 H/S
[2018-10-28 07:46:08] Starting reboot script run-me-watchdog.bat...
[2018-10-28 07:46:09] ocl_release: Device BUS_ID3 thread0 resources released
[2018-10-28 07:46:09] ocl_release: Device BUS_ID3 thread1 resources released
[2018-10-28 07:46:09] Stopping miner process

[bruslie]

hy @doktor83 ,

gpu watchdog don’t work for me anymore since 1.6.8

on gpu fail miner don’t use reboot file , he try to restart miner and stuck ... can you plz take a look ?

https://imgur.com/a/LOmcyIb

https://imgur.com/a/EK5Fjcj

I just checked, it is working :

Code:

{
"cryptonight_type" : "normalv8",
"intensity" : 0,
"double_threads" : true,
"reboot_script_gpu_watchdog" : "run-me-watchdog.bat"
}

Code:

[2018-10-28 07:45:35] watchdog: GPU0 [BUS: 3] hashing speed is 0 H/S
[2018-10-28 07:45:40] json_receive: {"jsonrpc":"2.0","method":"job","params":{"blob":"090994b6d5de053c408e072491052f414699215faf87c20ebd7d0b40ba3dd729a4a499cdbd6480000000005975b98061a7aec57fd05f418a5d774a78ea762a1abf6639fdd5b00109919f1205","job_id":"474","target":"cf8b0000"}}
[2018-10-28 07:45:40] pool_have_job: Pool sent a new job (ID: 474)
[2018-10-28 07:45:46] watchdog: GPU0 [BUS: 3] hashing speed is 0 H/S
[2018-10-28 07:45:57] watchdog: GPU0 [BUS: 3] hashing speed is 0 H/S
[2018-10-28 07:46:08] Starting reboot script run-me-watchdog.bat...
[2018-10-28 07:46:09] ocl_release: Device BUS_ID3 thread0 resources released
[2018-10-28 07:46:09] ocl_release: Device BUS_ID3 thread1 resources released
[2018-10-28 07:46:09] Stopping miner process

thx for check

but why in my log miner will do first a restart ?

can you share your reboot me .bat ?

thx !

[doktor83]

thx for check

but why in my log miner will do first a restart ?

can you share your reboot me .bat ?

thx !

It could be anything in the .bat, as long as miner runs it I just had something echoed for the test.

[doktor83]

V1.6.9
- Algo switching without miner restart
- Benchmark everything from algos.txt and just paste the hashrates (--benchmarkalgos)
- Added number of algo switches to API and stats
- Added cmd parameter --resetmineronalgoswitch which forces miner to reset on algo change
- Changed minimum --algoswitchmintime from 3 min to 30 sec, default is still 600 sec
- Fixed API pool stats when not connected to a pool
- Added cmd parameters --cjobtimeout, --cmaxdifficulty, --ckeepalive
- When critical temp is reached, ADL temp is now logged before shutdown
- Notification on too many compute errors (set with --gpuerrorsalert value)
- Added a performance tuning parameter 'fragments' for gpu_conf and cmd
- Fixed a bug that could crash miner on shutdown


+ Algo switch without miner restart is here finally. This is now the default mode. If you want to use the 'old fashioned' switch where the miner restarts itself use --resetmineronalgoswitch in .bat.
The only reason i can see to use the switching with restart is for Vegas, because they tend to drop hashrate on algo switch, and only --resetvega can help to regain full speed again.
Check out here how to setup algo switching

+ Some people were too lazy to benchmark algo by algo to get the 'hashrate' values for algos.txt, so there is a --benchmarkalgos now , that benchmarks the algos set in algos.txt, and creates benchmark_results.txt, from where you just have to take and paste the values into algos.txt

+ The --algoswitchmintime minimum value can now be 30 seconds, altough Monero Ocean really likes to disregard this setting

+ There is on-screen and log notification now when you have too much compute errors (default is 15). You can set this value with --gpuerrorsalert value

+ (re)added fragments parameter for some performance tuning. This parameter was set internally, but my tests show it can REALLY be different for even the same type of GPU, but different vendor.
So now everyone can find his 'optimal' value for this parameter. It differs for algos and also for the amount of VRAM, so a value that is good for 4GB cards maybe isn't the best value for 8GB cards.

[GKumaran]

Thx for the new version.

You have added 2 new tuning param in last 2 versions:

• Bralock
• fragments

Can you pls help us as to the nature of the tuning ,its effects and sort of recommended values for vega cards.
This will help us fine tune our cards better.

Thx dude !! 

[poolerMF]

in folder "Config" there should be file "config-empty.txt"

[doktor83]

Thx for the new version.

You have added 2 new tuning param in last 2 versions:

• Bralock
• fragments

Can you pls help us as to the nature of the tuning ,its effects and sort of recommended values for vega cards.
This will help us fine tune our cards better.

Thx dude !! 

Fragments -> my tests show it can REALLY be different for even the same type of GPU, but different vendor.

So i can't give you a recommended setting for a vega, sorry. For my Gigabyte 56 with hynix mem the winner value is 4 for v8, and 1 for everything else.

Bralock -> this is something similar as worksize, you will also need to experiment with the values

[GKumaran]

Thx for the new version.

You have added 2 new tuning param in last 2 versions:

• Bralock
• fragments

Can you pls help us as to the nature of the tuning ,its effects and sort of recommended values for vega cards.
This will help us fine tune our cards better.

Thx dude !!  

Fragments -> my tests show it can REALLY be different for even the same type of GPU, but different vendor.

So i can't give you a recommended setting for a vega, sorry. For my Gigabyte 56 with hynix mem the winner value is 4 for v8, and 1 for everything else.

Bralock -> this is something similar as worksize, you will also need to experiment with the values

Hmm, k dude.

The miner shows fragments of 4 as defaults on my vega.

Testing other fragments also generates the same 4, doesnt change to 8 or 2 etc.
My config file:

Code:

{
"cryptonight_type" : "normalv8",

"intensity" : 124,
"worksize" : 16,
"double_threads" : true,
"fragments" : 64,
"shutdown_temperature" : 80,

"timeout" : 5,
"retry_time" : 5,
"giveup_limit" : 5,
"main_pool_reconnect" : 300,

"min_rig_speed" : 11850,
"min_rig_speed_duration" : 250
}

[camchimonline]

edit same:

"reboot_script_gpu_watchdog" : "reboot-windows.bat",
"gpu_conf" :
[
   { "id" : 0, "intensity" : 58, "worksize" : 16, "threads" : 2, "fragments" : 8, "bralock": 16},
   { "id" : 1, "intensity" : 58, "worksize" : 16, "threads" : 2, "fragments" : 8, "bralock": 16},
   { "id" : 2, "intensity" : 58, "worksize" : 16, "threads" : 2, "fragments" : 8, "bralock": 16},
   { "id" : 3, "intensity" : 58, "worksize" : 16, "threads" : 2, "fragments" : 8, "bralock": 16},
   { "id" : 4, "intensity" : 58, "worksize" : 16, "threads" : 2, "fragments" : 8, "bralock": 16},
   { "id" : 5, "intensity" : 58, "worksize" : 16, "threads" : 2, "fragments" : 8, "bralock": 16},
]
}

Thx for the new version.

You have added 2 new tuning param in last 2 versions:

• Bralock
• fragments

Can you pls help us as to the nature of the tuning ,its effects and sort of recommended values for vega cards.
This will help us fine tune our cards better.

Thx dude !!  

Fragments -> my tests show it can REALLY be different for even the same type of GPU, but different vendor.

So i can't give you a recommended setting for a vega, sorry. For my Gigabyte 56 with hynix mem the winner value is 4 for v8, and 1 for everything else.

Bralock -> this is something similar as worksize, you will also need to experiment with the values

Hmm, k dude.

The miner shows fragments of 4 as defaults on my vega.

Testing other fragments also generates the same 4, doesnt change to 8 or 2 etc.
My config file:

Code:

{
"cryptonight_type" : "normalv8",

"intensity" : 124,
"worksize" : 16,
"double_threads" : true,
"fragments" : 64,
"shutdown_temperature" : 80,

"timeout" : 5,
"retry_time" : 5,
"giveup_limit" : 5,
"main_pool_reconnect" : 300,

"min_rig_speed" : 11850,
"min_rig_speed_duration" : 250
}

[bojan_s]

Best intensity value for Sapphire RX580 8Gb Nitro SE Samsung mem, 1250/2050?
I am getting around 9500-1100 hs

[doktor83]

Thx for the new version.

You have added 2 new tuning param in last 2 versions:

• Bralock
• fragments

Can you pls help us as to the nature of the tuning ,its effects and sort of recommended values for vega cards.
This will help us fine tune our cards better.

Thx dude !!  

Fragments -> my tests show it can REALLY be different for even the same type of GPU, but different vendor.

So i can't give you a recommended setting for a vega, sorry. For my Gigabyte 56 with hynix mem the winner value is 4 for v8, and 1 for everything else.

Bralock -> this is something similar as worksize, you will also need to experiment with the values

Hmm, k dude.

The miner shows fragments of 4 as defaults on my vega.

Testing other fragments also generates the same 4, doesnt change to 8 or 2 etc.
My config file:

Code:

{
"cryptonight_type" : "normalv8",

"intensity" : 124,
"worksize" : 16,
"double_threads" : true,
"fragments" : 64,
"shutdown_temperature" : 80,

"timeout" : 5,
"retry_time" : 5,
"giveup_limit" : 5,
"main_pool_reconnect" : 300,

"min_rig_speed" : 11850,
"min_rig_speed_duration" : 250
}

- Added a performance tuning parameter 'fragments' for gpu_conf and cmd

[MagicSmoker]

...

• Run 18.6.1 for less power consumption or 18.5.1 for higher hashrate (marginal differences both).
• Use MSI AB to "extend official overclock limits." Q: are you using MSI AB or ODT to control the fans? I greatly prefer using AB to set a custom fan curve.
• Bump up PL in AB to +5%
• Use ODT 0.2.5 to set highest P state for core/mem clock/voltage, trying 1180MHz/895mV for core and 2000-2100MHz for mem. Q: What about memory voltage?
• Set intensity manually in SRB try 23 to 24 initially. Q: What about worksize? SRB is setting it to 32 automatically (also, double threads true and K auto set to 1).

And that should bring my power consumption down to around 40W per card with a slight boost to hashrate, correct?

Depending on what brand 560's you have, they may not be able to go down to 40W and retain 450 h/s, it all depends on how well they can undervolt. I think it's impossible with the 18.5.1 drivers you must use 18.6.1 or no dice. For me 18.5.1 read 40w on gpu-z if I had tested on the wall I'm sure it would actually be 44-45w. 18.6.1 read 36w on gpu-z and goes no higher than 39.5w on the wall for me. The tiny ~5 H/s gain on 18.5.1 is definitely not worth the higher power draw. If you're desperate for lower undervolt you could try lower core clock than 1180, but at this point every step down is negative for hashrate.
I just use ODT at fixed fan speeds, my 6x rx560 2gb rig is in my bedroom so I try to keep it quiet like 1600 rpm, as long as my cards stay 65C or under it's okay.
On v7 I got max hashrate with -10% power limit. On v8 I have to increase to +10% for max hashrate. PL doesn't really effect consumption, the core voltage effects it most.
On v8, my 560's need 865 mV on mem voltage to be 100% stable, at 860 mV they can crash or lose hashrate after many hours. I had it on 860 mV before because it was stable with older v7, so v8 needed +5 mV.
Try 8 worksize first and then 12/16 and see if it improves hashrate, doesn't make sense for a 2gb card to go any higher than 16.

Holy cannolis did I make myself a metric craptonne of work for myself yesterday. In directly violation of the scientific method - change one variable at a time - I DDU'ed 18.5.1 and installed 18.6.1, then moved the RX 570 from my main desktop over to my 4x RX 560 mining rig, then proceeded to spend the next 4 hours trying to get the mining rig to not crash running either SRB or xmr-stak. I finally gave up on the RX 570 but it wasn't a totally wasted effort: now I am getting 1840H/s out of the 4x RX 560 rig at the same 355W total from the wall (55W actual per GPU) running 18.6.1 and manually setting worksize to 8 (but letting SRB set the intensity, and still with double threads true).

And of course this morning I see there is a new version of SRB out, but I am too shell-shocked from yesterday to even contemplate messing around with this today.

Oh, and I didn't even get to the point of trying to tweak voltages with ODT; the miner needs to run for more than a few seconds to bother with that kind of optimization.

At any rate, thanks for help!

[GKumaran]

Thx for the new version.

You have added 2 new tuning param in last 2 versions:

• Bralock
• fragments

Can you pls help us as to the nature of the tuning ,its effects and sort of recommended values for vega cards.
This will help us fine tune our cards better.

Thx dude !! 

Fragments -> my tests show it can REALLY be different for even the same type of GPU, but different vendor.

So i can't give you a recommended setting for a vega, sorry. For my Gigabyte 56 with hynix mem the winner value is 4 for v8, and 1 for everything else.

Bralock -> this is something similar as worksize, you will also need to experiment with the values

Testing ALL the available fragment variables.
Best HR for Sapphire Vega 64 Reference card(8GB Samsung) is from a fragments size of 16. This is 10h/s above the default of 4 fragments

Code:

"gpu_conf" : 
[ 
	{	"id" : 0, "cryptonight_type" : "normalv8", "intensity" : 124, "worksize" : 16, "double_threads" : true, "fragments" : 16}
]

[playfast]

...

• Run 18.6.1 for less power consumption or 18.5.1 for higher hashrate (marginal differences both).
• Use MSI AB to "extend official overclock limits." Q: are you using MSI AB or ODT to control the fans? I greatly prefer using AB to set a custom fan curve.
• Bump up PL in AB to +5%
• Use ODT 0.2.5 to set highest P state for core/mem clock/voltage, trying 1180MHz/895mV for core and 2000-2100MHz for mem. Q: What about memory voltage?
• Set intensity manually in SRB try 23 to 24 initially. Q: What about worksize? SRB is setting it to 32 automatically (also, double threads true and K auto set to 1).

And that should bring my power consumption down to around 40W per card with a slight boost to hashrate, correct?

Depending on what brand 560's you have, they may not be able to go down to 40W and retain 450 h/s, it all depends on how well they can undervolt. I think it's impossible with the 18.5.1 drivers you must use 18.6.1 or no dice. For me 18.5.1 read 40w on gpu-z if I had tested on the wall I'm sure it would actually be 44-45w. 18.6.1 read 36w on gpu-z and goes no higher than 39.5w on the wall for me. The tiny ~5 H/s gain on 18.5.1 is definitely not worth the higher power draw. If you're desperate for lower undervolt you could try lower core clock than 1180, but at this point every step down is negative for hashrate.
I just use ODT at fixed fan speeds, my 6x rx560 2gb rig is in my bedroom so I try to keep it quiet like 1600 rpm, as long as my cards stay 65C or under it's okay.
On v7 I got max hashrate with -10% power limit. On v8 I have to increase to +10% for max hashrate. PL doesn't really effect consumption, the core voltage effects it most.
On v8, my 560's need 865 mV on mem voltage to be 100% stable, at 860 mV they can crash or lose hashrate after many hours. I had it on 860 mV before because it was stable with older v7, so v8 needed +5 mV.
Try 8 worksize first and then 12/16 and see if it improves hashrate, doesn't make sense for a 2gb card to go any higher than 16.

Holy cannolis did I make myself a metric craptonne of work for myself yesterday. In directly violation of the scientific method - change one variable at a time - I DDU'ed 18.5.1 and installed 18.6.1, then moved the RX 570 from my main desktop over to my 4x RX 560 mining rig, then proceeded to spend the next 4 hours trying to get the mining rig to not crash running either SRB or xmr-stak. I finally gave up on the RX 570 but it wasn't a totally wasted effort: now I am getting 1840H/s out of the 4x RX 560 rig at the same 355W total from the wall (55W actual per GPU) running 18.6.1 and manually setting worksize to 8 (but letting SRB set the intensity, and still with double threads true).

And of course this morning I see there is a new version of SRB out, but I am too shell-shocked from yesterday to even contemplate messing around with this today.

Oh, and I didn't even get to the point of trying to tweak voltages with ODT; the miner needs to run for more than a few seconds to bother with that kind of optimization.

At any rate, thanks for help!

Awesome hashrate but still, that's disappointing power consumption. The fact that you can't get the 570 stable suggests an issue with the rig outside of software. Only other thing to try is a different motherboard/PSU or your 560's are just plain power hogs. From experience the best AMD vendors are MSI and Gigabyte, usually have best asic quality and good undervolt results with them. Sapphire is a mixed bag. The worst are PowerColor and XFX, they always seem to use cheap reference design and struggle with undervolt.

[haliavshik]

i have 3 farms on rx570 one of them is with just 1950 mem (thats on pc that i am gaming on)
when it switch algo on 1.6.9 i get
[2018-10-28 13:55:50] pool_have_job: Pool sent a new job (ID: ezf4jX58YooLZjABQrqsX7CgVRKD) [haven]
[2018-10-28 13:55:53] compute_error_alert: Too many compute errors on GPU BUS_ID[2] (15)

it spams errors on all cards no shares on pool and after it switches back to v8 it starts mining again on fast algo its the same pattern

with --resetmineronalgoswitch it works fine on 1.6.8 aswell

[UnclWish]

Best intensity value for Sapphire RX580 8Gb Nitro SE Samsung mem, 1250/2050?
I am getting around 9500-1100 hs

You forgot to point algo.

[Bwtmn7]

"double_threads" : true,
"gpu_conf" :

I see i had one extra " in the pool.txt

[MagicSmoker]

Holy cannolis did I make myself a metric craptonne of work for myself yesterday...

Awesome hashrate but still, that's disappointing power consumption. The fact that you can't get the 570 stable suggests an issue with the rig outside of software. Only other thing to try is a different motherboard/PSU or your 560's are just plain power hogs. From experience the best AMD vendors are MSI and Gigabyte, usually have best asic quality and good undervolt results with them. Sapphire is a mixed bag. The worst are PowerColor and XFX, they always seem to use cheap reference design and struggle with undervolt.

I was having the same problem trying to get the RX 570 to work with SRB on the main desktop - crashing the miner soon after logging in to the pool - but that was with 18.5.1 (hence why I changed to 18.6.1 when I moved it to the dedicated mining rig). All of my AMD cards are XFX, and yes, they do struggle with undervolting. Decent hashrates when you get them dialed in, though. I know they aren't the best, but they were what was available when got into mining last November (not the best time, but at least I didn't keep buying GPUs all through January...).

Once I get 1.6.9 running on the 4x RX 560 rig I'll try fooling with voltages in ODT to get the power usage down, but I'm trying to keep it all in perspective: the extra 60W per day amounts to around $0.17 in additional energy cost and now the rigs acting as heaters is a good thing, rather than something I was trying to minimize.