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

[doktor83]

Does anyone know the details of how the .85% fee is collected?  I looked through the logs and it was not clear to me.  If I keep restarting the miner during a tweaking period... is it the first period of time that it collects, what about when it auto restart the miner because it detects a GPU failure... does it meter it's collection based on that in any way?  I am sure there is a method (and it's OK for coders to get paid...), I just would like to know what it is.
Thanks.

There is no fee mined in the first 5 minutes, so you can do the tweaking without paying every time the fee.
On gpu failure miner restart or user kill/close miner process and start again, the fee timer is restarted.

[maximus63]

Verze 1.7.2 MEGA BEST !!!!

Test algo: Heavy

Asrock 570 8GB = 1156 H/s - 1200/2250 - 53C - 0.868 mV - 100W max.

VERY GOOD THANKS

Nitro+ RX570 8gb JCE Miner 1200/2240 = 1200 H/s - 98w max

[big_daddy]

Verze 1.7.2 MEGA BEST !!!!

Test algo: Heavy

Asrock 570 8GB = 1156 H/s - 1200/2250 - 53C - 0.868 mV - 100W max.

VERY GOOD THANKS

Nitro+ RX570 8gb JCE Miner 1200/2240 = 1200 H/s - 98w max

What type of memory are your tested cards?

[maximus63]

Verze 1.7.2 MEGA BEST !!!!

Test algo: Heavy

Asrock 570 8GB = 1156 H/s - 1200/2250 - 53C - 0.868 mV - 100W max.

VERY GOOD THANKS

Nitro+ RX570 8gb JCE Miner 1200/2240 = 1200 H/s - 98w max

What type of memory are your tested cards?

Micron

[doktor83]

Verze 1.7.2 MEGA BEST !!!!

Test algo: Heavy

Asrock 570 8GB = 1156 H/s - 1200/2250 - 53C - 0.868 mV - 100W max.

VERY GOOD THANKS

Nitro+ RX570 8gb JCE Miner 1200/2240 = 1200 H/s - 98w max

What type of memory are your tested cards?

Micron

You are doing something bad, on 470 8g with mem straps and clocks for eth, not monero :
1200/2250 is overkill

[goodminer]

Hi, I downloaded your last version of miner and I can't start mining on fast algo with new adrenalin 18.12.2 drivers (driver is crushing), but can't anymore mining on 1.6.9 version of your miner, the same problem. I tried to input 6 amd ocl files (amd_opencl32, amd_opencl64, amdocl, amdocl12cl, amdocl12cl64, amdocl64) from recommended drivers 18.5.2 and 18.6.1 into both versions of your miner and I can't start mining...What can be the problem and what is solution??
I have radeon r7 370 card and the same problem is on 470 cards.
Thank you forward.

[Uaciuganadu]

Hey Dok,

Have you played around with the new Adrenalin 2019 drivers? Do you plan on supporting them and optimize SRB on them??

[P00P135]

Hey Dok,

Have you played around with the new Adrenalin 2019 drivers? Do you plan on supporting them and optimize SRB on them??

I wouldnt touch these until they have fleshed out more bugs.  It's a crash fest while trying to tweak multiple GPU's.

[a_tom]

My rig (6x470) rented on MRR by CN-fast algo.
I dont no what coin mined by renter, but...
On version 1.7.2 miner - all shares STALE
On version 1.6.9 - OK. 1-2% stale
Miner show pool diff 1000, thats all what I can see

[doktor83]

My rig (6x470) rented on MRR by CN-fast algo.
I dont no what coin mined by renter, but...
On version 1.7.2 miner - all shares STALE
On version 1.6.9 - OK. 1-2% stale
Miner show pool diff 1000, thats all what I can see

The answer is there : diff 1000.
That's why a lot of stales.

[a_tom]

The answer is there : diff 1000.
That's why a lot of stales.

So version 1.6.9 is OK, but 1.7.2. - all stale. I don`t understand why. Other version dont check.

[doktor83]

The answer is there : diff 1000.
That's why a lot of stales.

So version 1.6.9 is OK, but 1.7.2. - all stale. I don`t understand why. Other version dont check.

1.7.2 is also ok, but he should use a pool with higher diff, 1000 is insanely low for 6 gpu's, he finds a ton of shares, and when new job arrives, those shares are still sent to the pool and they are stales then.
It can even happen that he gets rejects.

Just use much higher difficulty port on that pool, or a vardiff port.

[dazzmore]

Hi Dok, is there any way to find out which GPU (I have mix 56's & 64's with different memories) crashed the system? I've tried searching log but can't really tell. TIA

[doktor83]

Hi Dok, is there any way to find out which GPU (I have mix 56's & 64's with different memories) crashed the system? I've tried searching log but can't really tell. TIA

depends how the system crashed, if blue screen or freeze then you can't do much i guess.
If just a card dropped out, that will be in the log.

[lsHaKeRl]

Hi Dok, is there any way to find out which GPU (I have mix 56's & 64's with different memories) crashed the system? I've tried searching log but can't really tell. TIA

Hi Mate!

It´s easy to find out which Card is "damaged", just go to device Manager and just activate 1 GPU Card 1 by 1 and run the miner with just 1 ID, then let run the miner.

maybe it helps. 

Greetings
Patrick

[doktor83]

V1.7.3
- Bringing back support for GCN1 cards [pitcairn, tahiti ...]
- Heavy_mode 2 is now an improved 3 with a little less hash and less errors than 3
- Small OCL optimisations for Vegas
- Added parameter 'off_temperature' to gpu_conf, to protect GPU from overheating
- Added parameter 'old_mode' to gpu_conf, which creates the kernel with the old method
- Added new algo 'Hycon'
- Added new algo 'Upx'
- Added new algo 'Stellitev8'
- Renamed algo 'freehaven' to 'swap'


+ Even if unintentionally, but SRB became slower and slower on old GCN1 cards because optimisations that were good for newer gpu's were a kill for those oldes.
I managed to borrow a tahiti card (280x) for a few days, and create a new kernel that is now working OK with these oldies. Speeds should be acceptable now

+ As you know there are 3 heavy modes available. This is how it goes now :
1. Slowest but shouldn't produce any compute errors
2. Faster than 1, it is a re-worked 3, a little bit slower than 3, but with less errors
3. The fastest but it can produce the most compute errors

I think 2 could be the winner here because my tests of 12hr show :
1. 0 errors
2. 1 error
3. 4 errors

Of course the default one is still 1, because it is the most stable mode.

+ Added a gpu_conf only parameter 'off_temperature'. I think this will be very useful. What it does is it turns off that one particular overheated GPU until it cools down by 15c, then it's turned back on.
Default is 85c, so that means if a GPU reaches 85c, miner turns it off automatically (while the other GPU's are still working), and it waits until it cools down by 15c, in this case that is 70c.
When temperature is < 70c the GPU is automatically turned back on and it continues to work.
The '15c' difference can't be changed.

+ Added a gpu_conf only parameter 'old_mode'. With the changes i made in 1.7.2, that allow the creation of kernels on drivers newer than 18.6.1, algos that use a scratchpad <=1MB (litev7, mox, dark, upx) suffer from a hashrate drop on Vegas (maybe others too, i didn't test). So if you set this parameter to true, it will create kernel the old way, and the speed on these algos should be back.
Don't forget, if you use old_mode : true, and compile the kernel on a driver newer than 18.6.1, you probably won't be finding any shares!

+ Added 3 new algos :

- upx [ https://bitcointalk.org/index.php?topic=5058404.0 ]
- hycon [ https://hycon.io ]
- stellitev8 [ https://bitcointalk.org/index.php?topic=2813261.0 ]

Stellitev8 will be active when stellite forks.

[GKumaran]

Thank you for the new version Dok !! 
Restores 1mb algo speeds to their original high speeds !!

For vega users :  the key word being GPU_Conf "ONLY"

Code:

"gpu_conf" : 
[ 
	{ "id" : 0, "intensity" : 0, "worksize" : 8, "fragments" : 1, "double_threads" : true, "old_mode" : true }
]

[doktor83]

Thank you for the new version Dok !! 
Restores 1mb algo speeds to their original high speeds !!

For vega users :  the key word being GPU_Conf "ONLY"

Code:

"gpu_conf" : 
[ 
	{ "id" : 0, "intensity" : 0, "worksize" : 8, "fragments" : 1, "double_threads" : true, "old_mode" : true }
]

Yes, i should use font size 300, because even if there are separate sections for gpu_conf and other parameters in the readme, and on the first page, somehow people dont see it 

[zeef]

V1.7.3
- Bringing back support for GCN1 cards [pitcairn, tahiti ...]
- Heavy_mode 2 is now an improved 3 with a little less hash and less errors than 3
- Small OCL optimisations for Vegas
- Added parameter 'off_temperature' to gpu_conf, to protect GPU from overheating
- Added parameter 'old_mode' to gpu_conf, which creates the kernel with the old method
- Added new algo 'Hycon'
- Added new algo 'Upx'
- Added new algo 'Stellitev8'
- Renamed algo 'freehaven' to 'swap'


+ Even if unintentionally, but SRB became slower and slower on old GCN1 cards because optimisations that were good for newer gpu's were a kill for those oldes.
I managed to borrow a tahiti card (280x) for a few days, and create a new kernel that is now working OK with these oldies. Speeds should be acceptable now

+ As you know there are 3 heavy modes available. This is how it goes now :
1. Slowest but shouldn't produce any compute errors
2. Faster than 1, it is a re-worked 3, a little bit slower than 3, but with less errors
3. The fastest but it can produce the most compute errors

I think 2 could be the winner here because my tests of 12hr show :
1. 0 errors
2. 1 error
3. 4 errors

Of course the default one is still 1, because it is the most stable mode.

+ Added a gpu_conf only parameter 'off_temperature'. I think this will be very useful. What it does is it turns off that one particular overheated GPU until it cools down by 15c, then it's turned back on.
Default is 85c, so that means if a GPU reaches 85c, miner turns it off automatically (while the other GPU's are still working), and it waits until it cools down by 15c, in this case that is 70c.
When temperature is < 70c the GPU is automatically turned back on and it continues to work.
The '15c' difference can't be changed.

+ Added a gpu_conf only parameter 'old_mode'. With the changes i made in 1.7.2, that allow the creation of kernels on drivers newer than 18.6.1, algos that use a scratchpad <=1MB (litev7, mox, dark, upx) suffer from a hashrate drop on Vegas (maybe others too, i didn't test). So if you set this parameter to true, it will create kernel the old way, and the speed on these algos should be back.
Don't forget, if you use old_mode : true, and compile the kernel on a driver newer than 18.6.1, you probably won't be finding any shares!

+ Added 3 new algos :

- upx [ https://bitcointalk.org/index.php?topic=5058404.0 ]
- hycon [ https://hycon.io ]
- stellitev8 [ https://bitcointalk.org/index.php?topic=2813261.0 ]

Stellitev8 will be active when stellite forks.

Thanks for new version.

When could we see support for a more recent amd drivers?

Thanks

[doktor83]

Thanks for new version.

When could we see support for a more recent amd drivers?

Thanks

What do you mean?
It works with the latest drivers also