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

[SmartOdissey]

finally i fixed issue with low hashrate on 580 with latest drivers.

Haven protocol with 1.6.2 now:
 RX 580 - 1100+-10 hashes
 Vega56 - 1450+-20 hashes

[DezmonDND]

finally i fixed issue with low hashrate on 580 with latest drivers.

Haven protocol with 1.6.2 now:
 RX 580 - 1100+-10 hashes
 Vega56 - 1450+-20 hashes

Core and memory clocks for your RX580 card? I have only 1040 hashes on RX580 8 GB (Micron), intensity 51. 1240/2230 without memory errors  

[zeef]

doktor,

this is happening mining bittubev2 in v1.6.2,
in other rigs is the same.. after a time i got an hasdropped.

Take a printscreen:

https://ibb.co/mnEFbo

Thanks

you need to crank up the fans to 3000rpm because hbm gets too hot that is why your hash is jumping.

Hi Doktor,

But on heavy or cryptonight doesn t occur that.  I have found  a sweet spot to avoid the throthling..

Just tested in a vega 56 putting the fans to maximum and in a vega 64 and its the same after one hour aproximally happens what i described.

With other miners doesn t ocurr that.

Could be something in your miner for bittubealgov2.. ?

Regards

[fellowbeing]

I can confirm that I am getting about 100 less hash/s on CN7 using the SRB Miner 1.6.2 instead of 1.6.1.

[RuMiner]

I can confirm that I am getting about 100 less hash/s on CN7 using the SRB Miner 1.6.2 instead of 1.6.1.

thanks. nobody believes me still using 1.6.1 because of that

[tf2addict]

My 1.6.2 is fine, same hash rate as 1.6.1.  Try playing with intensity.

[livada]

I can confirm that I am getting about 100 less hash/s on CN7 using the SRB Miner 1.6.2 instead of 1.6.1.

i have 300HR + in 1.6.2 with 6 vega card 112/16
and 400HR+ with heavy-tube algo 56/8

[doktor83]

vega got more hash on 1.6.2, rx series maybe a little less, but next version should get things back on track.

[SmartOdissey]

finally i fixed issue with low hashrate on 580 with latest drivers.

Haven protocol with 1.6.2 now:
 RX 580 - 1100+-10 hashes
 Vega56 - 1450+-20 hashes

Core and memory clocks for your RX580 card? I have only 1040 hashes on RX580 8 GB (Micron), intensity 51. 1240/2230 without memory errors  

XFX 580-8Gb Samsung memory, flashed with SRBPolaris bios. 1318/1100Mv Core, 2130/950mV Memory, -20% powerlimit
My friend have same cards XFX 580-8 with same BIOS, but he have 1100-1150 h/s! 1318/1150 Core, 2100/950 Mem and 0 powerlimit.

[fellowbeing]

vega got more hash on 1.6.2, rx series maybe a little less, but next version should get things back on track.

I have a rig of 5 Vega Frontier Editions. I have not tried on heavy algo but the CN7 is giving me about 100 less. Intensity is set to 120 with double threads. I am very happy with SRB Miner.

with XMR STAK, if I set intensity to 2048 for each thread then I would get closer to the hashrate of SRB but still 100 or so less. With SRB 1.6.1 I am getting a stable 10.9 KH/s, if I added another Vega FE card it would be 13 KH/s.


You can see photos of my rig and its making here: https://postimg.cc/gallery/ivrtjh24/

As you know Vega FE cards have 16 GB of HBM2. Is it possible to run more threads on these GPUs ? It would be awesome if there was a way to utilize this memory. I would appreciate if you could provide an answer to this question, as you are writing the code for SRB Miner and you know better how much memory can be used.

Thanks for the awesome SRB Miner

[doktor83]

vega got more hash on 1.6.2, rx series maybe a little less, but next version should get things back on track.

I have a rig of 5 Vega Frontier Editions. I have not tried on heavy algo but the CN7 is giving me about 100 less. Intensity is set to 120 with double threads. I am very happy with SRB Miner.

with XMR STAK, if I set intensity to 2048 for each thread then I would get closer to the hashrate of SRB but still 100 or so less. With SRB 1.6.1 I am getting a stable 10.9 KH/s, if I added another Vega FE card it would be 13 KH/s.


You can see photos of my rig and its making here: https://postimg.cc/gallery/ivrtjh24/

As you know Vega FE cards have 16 GB of HBM2. Is it possible to run more threads on these GPUs ? It would be awesome if there was a way to utilize this memory. I would appreciate if you could provide an answer to this question, as you are writing the code for SRB Miner and you know better how much memory can be used.

Thanks for the awesome SRB Miner

I don't have FE vega, but try this for V7 algo:

"gpu_conf" :
[
   { "id" : 0, "intensity" : 126, "worksize" : 8, "threads" : 2}
]

[doktor83]

I can confirm that I am getting about 100 less hash/s on CN7 using the SRB Miner 1.6.2 instead of 1.6.1.

thanks. nobody believes me still using 1.6.1 because of that

Who said nobody believes you ? 
Which gpu's are we talking about? Rx550 or Rx580 ?

[yaya79]

my result vega64 1550/920, 1100/920 heavy algo.
Is it possible to make more than 1700hs?
https://preview.ibb.co/fUdLCT/image.jpg

[roma__11]

my result vega64 1550/920, 1100/920 heavy algo.
Is it possible to make more than 1700hs?

Driver?

win10-64bit-radeon-software-adrenalin-edition-18.6.1-june13
https://thumb.ibb.co/diHN98/img_2018_07_09_12_10_23.png

[yaya79]

my result vega64 1550/920, 1100/920 heavy algo.
Is it possible to make more than 1700hs?

Driver?

win10-64bit-radeon-software-adrenalin-edition-18.6.1-june13
https://thumb.ibb.co/diHN98/img_2018_07_09_12_10_23.png

Thx! With 18.6.1 driver - 1730hs (1550 core).

[doktor83]

V1.6.3
- More speed on Vega56 cards on every algo (probably on Vega64 and VegaFE too, i don't have them to test)
- Added parameter --enableduplicategpuid (used in start.bat), that enables to use the same gpu id in gpu_conf multiple times
- Removed useless kernels 3 and 4
- Removed useless CPU opencl support

+ Vega cards should get even more speed than in previous version (tested with adrenaline 18.3.4)

+ Added parameter --enableduplicategpuid (in .bat) which basically allows you to do something like this :

"gpu_conf" :
[
   { "id" : 0, "intensity" : 120, "worksize" : 8, "threads" : 1},
   { "id" : 0, "intensity" : 120.5, "worksize" : 8, "threads" : 1},
]

So instead of using the same settings for every thread on same GPU, you can define different intensity for example for one thread.
This way maybe you can squeeze out a little more hash from your GPU.

On initialisation this is displayed as DIFFERENT GPU'S, but on hashrate display threads of same gpu are merged, so a per gpu hashrate is still displayed, not a per thread speed.

+ Removed a few useless things like kernels 3&4, and CPU OpenCL support, nobody used these but they just took me more time when implementing something new


Don't copy new version of miner over old one, or if you do make sure you delete .srb files before running it!

[mk111]

V1.6.3
- More speed on Vega56 cards on every algo (probably on Vega64 and VegaFE too, i don't have them to test)
- Added parameter --enableduplicategpuid (used in start.bat), that enables to use the same gpu id in gpu_conf multiple times
- Removed useless kernels 3 and 4
- Removed useless CPU opencl support

+ Vega cards should get even more speed than in previous version (try with adrenaline 18.3.4 or adrenaline 18.6.1)

+ Added parameter --enableduplicategpuid (in .bat) which basically allows you to do something like this :

"gpu_conf" :
[
   { "id" : 0, "intensity" : 120, "worksize" : 8, "threads" : 1},
   { "id" : 0, "intensity" : 120.5, "worksize" : 8, "threads" : 1},
]

So instead of using the same settings for every thread on same GPU, you can define different intensity for example for one thread.
This way maybe you can squeeze out a little more hash from your GPU.

On initialisation this is displayed as DIFFERENT GPU'S, but on hashrate display threads of same gpu are merged, so a per gpu hashrate is still displayed, not a per thread speed.

+ Removed a few useless things like kernels 3&4, and CPU OpenCL support, nobody used these but they just took me more time when implementing something new

Great stuff!
On heavy algo I am getting 40H/s extra per Vega 56 with V1.6.3

[livada]

my result vega64 1550/920, 1100/920 heavy algo.
Is it possible to make more than 1700hs?

Driver?

win10-64bit-radeon-software-adrenalin-edition-18.6.1-june13
https://thumb.ibb.co/diHN98/img_2018_07_09_12_10_23.png

WOW- this is nice- how much does it spend on the wall?

gpu mv? mem mv? pls SS overdrive setup

[UnclWish]

V1.6.3
- More speed on Vega56 cards on every algo (probably on Vega64 and VegaFE too, i don't have them to test)
- Added parameter --enableduplicategpuid (used in start.bat), that enables to use the same gpu id in gpu_conf multiple times
- Removed useless kernels 3 and 4
- Removed useless CPU opencl support

+ Vega cards should get even more speed than in previous version (try with adrenaline 18.3.4 or adrenaline 18.6.1)

+ Added parameter --enableduplicategpuid (in .bat) which basically allows you to do something like this :

"gpu_conf" :
[
   { "id" : 0, "intensity" : 120, "worksize" : 8, "threads" : 1},
   { "id" : 0, "intensity" : 120.5, "worksize" : 8, "threads" : 1},
]

So instead of using the same settings for every thread on same GPU, you can define different intensity for example for one thread.
This way maybe you can squeeze out a little more hash from your GPU.

On initialisation this is displayed as DIFFERENT GPU'S, but on hashrate display threads of same gpu are merged, so a per gpu hashrate is still displayed, not a per thread speed.

+ Removed a few useless things like kernels 3&4, and CPU OpenCL support, nobody used these but they just took me more time when implementing something new

What about cn-v7 algo on RX 5xx cards? You wanted to return speed from 1.6.0. On 1.6.1 and 1.6.2 cn-v7 algo on RX 580 is slower...

[doktor83]

V1.6.3
- More speed on Vega56 cards on every algo (probably on Vega64 and VegaFE too, i don't have them to test)
- Added parameter --enableduplicategpuid (used in start.bat), that enables to use the same gpu id in gpu_conf multiple times
- Removed useless kernels 3 and 4
- Removed useless CPU opencl support

+ Vega cards should get even more speed than in previous version (try with adrenaline 18.3.4 or adrenaline 18.6.1)

+ Added parameter --enableduplicategpuid (in .bat) which basically allows you to do something like this :

"gpu_conf" :
[
   { "id" : 0, "intensity" : 120, "worksize" : 8, "threads" : 1},
   { "id" : 0, "intensity" : 120.5, "worksize" : 8, "threads" : 1},
]

So instead of using the same settings for every thread on same GPU, you can define different intensity for example for one thread.
This way maybe you can squeeze out a little more hash from your GPU.

On initialisation this is displayed as DIFFERENT GPU'S, but on hashrate display threads of same gpu are merged, so a per gpu hashrate is still displayed, not a per thread speed.

+ Removed a few useless things like kernels 3&4, and CPU OpenCL support, nobody used these but they just took me more time when implementing something new

Great stuff!
I am getting 40H/s extra per Vega 56 with V1.6.3

That is what i measured too, increase on V7 about ~35HS.
But listen i just installed 18.6.1 and got even more HS! +30 LOL

So nice, this is a huge boost