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

[GKumaran]

19.3.2/19.3.3 works really good - one day without crash - same settings as 18.6.1 = little bit more performance and i can little bit lower in mv with same speed

Vega56 @64
GPU_P7=1450;855 (was 865 with 18.6.1 to get it stable) -> hwinfo 1410MHz / 831mv -> 2090kHs

i think the new timing level for mem is doing great - same tight timings as with vega56 BIOS but with possibility to run 100Mhz mem

Can u redo the test after deleting the old kernal ?

[zeef]

I think it is v4 problem. 1.8.1 is better than 1.8.0 but still not 100% stable. It was no problem on other algos in srbminer, eth rock stable, mined beam for weaks rock stable, everything i was trowing at vegas just worked. Then it came v4 and it came problems. I just love srbminer cause it is best miner for vega cards, but when v4 cam and it was unstable, i tried other miners for v4, you have at lleast 4 of them, and they crash even faster, teamredminer crashed under 10min on same settings as srbminer. Tried  to underclock gpu,mem for stability and same. Problem is only on v4, other algos are fine Something is very bad with new algo, just put loop in bat file and it will at least run again when it crashes

The whole concept of how v4 works is bad on gpu's. Compiling code on every new block, then need to release that memory, but it is known that the amd cl_release api is buggy and it wont always work as it should.
For cpu's this concept of create code + compile is not a problem , but for gpu's it is.

I am trying out various things, hope for the best  

hi doc
v4 algor , designed for cpu not gpu.

do you test new amd driver ? it's very good. it's work for few day just few crash .

Haha, no, i test only on 18.6.1 and blockchain. Which new version do you refer to?

Radeon-Software-Adrenalin-2019-Edition-19.3.2-Mar14
you don't need soft power table . just overdriventool 0.2.8 . you can edit p0 to p7 and new fan setting. and memory  timing ....

I use it too with overdriventool and very good stability and more hashes than 18.6.1 U should consider look at it doktor xD

[doktor83]

I use it too with overdriventool and very good stability and more hashes than 18.6.1 U should consider look at it doktor xD

I will try it, maybe there is even room for some small optimisations too

[tvukoman]

I think it is v4 problem. 1.8.1 is better than 1.8.0 but still not 100% stable. It was no problem on other algos in srbminer, eth rock stable, mined beam for weaks rock stable, everything i was trowing at vegas just worked. Then it came v4 and it came problems. I just love srbminer cause it is best miner for vega cards, but when v4 cam and it was unstable, i tried other miners for v4, you have at lleast 4 of them, and they crash even faster, teamredminer crashed under 10min on same settings as srbminer. Tried  to underclock gpu,mem for stability and same. Problem is only on v4, other algos are fine Something is very bad with new algo, just put loop in bat file and it will at least run again when it crashes

The whole concept of how v4 works is bad on gpu's. Compiling code on every new block, then need to release that memory, but it is known that the amd cl_release api is buggy and it wont always work as it should.
For cpu's this concept of create code + compile is not a problem , but for gpu's it is.

I am trying out various things, hope for the best  

hi doc
v4 algor , designed for cpu not gpu.

do you test new amd driver ? it's very good. it's work for few day just few crash .

Haha, no, i test only on 18.6.1 and blockchain. Which new version do you refer to?

Radeon-Software-Adrenalin-2019-Edition-19.3.2-Mar14
you don't need soft power table . just overdriventool 0.2.8 . you can edit p0 to p7 and new fan setting. and memory  timing ....

I use it too with overdriventool and very good stability and more hashes than 18.6.1 U should consider look at it doktor xD

Interesting. On 18.6.1 too long time here...

Can you share pics of the settings on new overdriveNtool 0.2.8 without ppt?

[doktor83]

It looks like newer drivers (adrenaline 2019) slow down the crashing issue that occurs because the memory or page file grows too much.
Anyone else can confirm this?

Testing now 19.3.2 and ONT 0.2.8, the timing level whatever i put isn't changing the hashrate for me on Vega56. Is it just me or .. ?

First tests show that hashrate is worse than on 18.6.1 - tested on bittube and v4.
If you use the compiled kernels from 18.6.1 on 19.3.2 then the hashrate will be also good.

[MinermanNC]

I have SRB MINER CN 1.181 running perfectly on windows 7 pushing 2 AMD 7790's

Can not get it to run correctly on Windows 10 pushing 1 R9 270x. It starts and receives work from pool, but does not find any shares. Nor does it display any GPU readings "temp, fan speed etc.. shows all zeros there.

It does show hashing speed of 470, and it identify's the card when I first launch it and again it receives work from pool.  

I have tried the recommended driver 18.5.1 and even tried different compatibility settings. Still no luck getting a share.

I have a feeling its a W10 issue? as it runs fine on W7 with older cards. Any ideas? still testing other cards on different OS

Thanks  

[UnclWish]

It looks like newer drivers (adrenaline 2019) slow down the crashing issue that occurs because the memory or page file grows too much.
Anyone else can confirm this?

Testing now 19.3.2 and ONT 0.2.8, the timing level whatever i put isn't changing the hashrate for me on Vega56. Is it just me or .. ?

First tests show that hashrate is worse than on 18.6.1 - tested on bittube and v4.
If you use the compiled kernels from 18.6.1 on 19.3.2 then the hashrate will be also good.

I think it has long been necessary to update your miner for optimal work with newest drivers. 18.6.1 is too old already.

[doktor83]

It looks like newer drivers (adrenaline 2019) slow down the crashing issue that occurs because the memory or page file grows too much.
Anyone else can confirm this?

Testing now 19.3.2 and ONT 0.2.8, the timing level whatever i put isn't changing the hashrate for me on Vega56. Is it just me or .. ?

First tests show that hashrate is worse than on 18.6.1 - tested on bittube and v4.
If you use the compiled kernels from 18.6.1 on 19.3.2 then the hashrate will be also good.

I think it has long been necessary to update your miner for optimal work with newest drivers. 18.6.1 is too old already.

Newer is not always better, as still the best hashrate can be achieved on 18.6.1 ... old or not

[doktor83]

I have SRB MINER CN 1.181 running perfectly on windows 7 pushing 2 AMD 7790's

Can not get it to run correctly on Windows 10 pushing 1 R9 270x. It starts and receives work from pool, but does not find any shares. Nor does it display any GPU readings "temp, fan speed etc.. shows all zeros there.

It does show hashing speed of 470, and it identify's the card when I first launch it and again it receives work from pool.  

I have tried the recommended driver 18.5.1 and even tried different compatibility settings. Still no luck getting a share.

I have a feeling its a W10 issue? as it runs fine on W7 with older cards. Any ideas? still testing other cards on different OS

Thanks  

When you try out different drivers every time clear the cache directory. R9 270x cards work fine on W10 and 18.6.1 drivers.

[chimaeraa]

It looks like newer drivers (adrenaline 2019) slow down the crashing issue that occurs because the memory or page file grows too much.
Anyone else can confirm this?

Testing now 19.3.2 and ONT 0.2.8, the timing level whatever i put isn't changing the hashrate for me on Vega56. Is it just me or .. ?

First tests show that hashrate is worse than on 18.6.1 - tested on bittube and v4.
If you use the compiled kernels from 18.6.1 on 19.3.2 then the hashrate will be also good.

i F8 .

no i think something in your setting is wrong . B.C  i got 2300 H/S . i seen some screen shot before.


i use new new compiled keener and it's good.

[Iamtutut]

DOKTOR83, there is a memory pill for AMD

Could you implement it on windows ? Right now it's linux only.

Here: https://bitcointalk.org/index.php?topic=5123724.0

[doktor83]

DOKTOR83, there is a memory pill for AMD

Could you implement it on windows ? Right now it's linux only.

Here: https://bitcointalk.org/index.php?topic=5123724.0

No, im not interested. Someone else will probably do it.

[JuanHungLo]

Using latest AMD driver, 19.3.3, win10, SRB 1.8.1, stock intensity, NO PPT, latest OverdriveNTool beta,
Vega 56@64 1552/852, 1100/900 getting H/s = 1994 average, stable for Monero.  Not sure of power draw.

[igotek]

i have just download the latest version.
It works well but there is a problem if i set gpus manually.

No problem on all 3 gpu

Code:

{
"cryptonight_type" : "stellitev8",
"intensity" : 0,
"double_threads" : true
}

mining not started and bat file close immediately.

Code:

{
"cryptonight_type" : "stellitev8",
"intensity" : 0,
"double_threads" : true

"gpu_conf" :
[
{ "id" : 0, "intensity" : 0 , "worksize" : 16, "threads" : 2},
{ "id" : 1, "intensity" : 0 , "worksize" : 16, "threads" : 2},
{ "id" : 2, "intensity" : 0 , "worksize" : 16, "threads" : 2}
]
}

Is there any chances to set gpus manually ?

[doktor83]

i have just download the latest version.
It works well but there is a problem if i set gpus manually.

No problem on all 3 gpu

Code:

{
"cryptonight_type" : "stellitev8",
"intensity" : 0,
"double_threads" : true
}

mining not started and bat file close immediately.

Code:

{
"cryptonight_type" : "stellitev8",
"intensity" : 0,
"double_threads" : true

"gpu_conf" :
[
{ "id" : 0, "intensity" : 0 , "worksize" : 16, "threads" : 2},
{ "id" : 1, "intensity" : 0 , "worksize" : 16, "threads" : 2},
{ "id" : 2, "intensity" : 0 , "worksize" : 16, "threads" : 2}
]
}

Is there any chances to set gpus manually ?

yes, put a comma after the double thread true line

[igotek]

Code:

{
"cryptonight_type" : "stellitev8",
"intensity" : 0,
"double_threads" : true,

"gpu_conf" :
[
{ "id" : 0, "intensity" : 0 , "worksize" : 16, "threads" : 2},
{ "id" : 1, "intensity" : 0 , "worksize" : 16, "threads" : 2},
{ "id" : 2, "intensity" : 0 , "worksize" : 16, "threads" : 2}
]
}

it is the same. not working. The strange thing is older versions work with same config.

[anvl]

 Some bug in the miner. On my systems with two CPUs one card is recognized as BUS: -125 by the miner whereas the system says it's 131.    Seems like not enough bits somewhere. As a result this card doesn't report ADL information.
 doktor83, could you please correct this error in your great miner?

[iRonNuke]

Why am I getting a lot of rejected shares? "Pool rejected result 0x000074E1 for job[2] [low difficulty share]" By a lot I mean some, most shares are ofcourse accepted but still, quite frequent.

Pool difficulty is set at 120000 by default. May I change it for better results?

The pool I'm using is nanopool.org btw.

[doktor83]

Why am I getting a lot of rejected shares? "Pool rejected result 0x000074E1 for job[2] [low difficulty share]" By a lot I mean some, most shares are ofcourse accepted but still, quite frequent.

Pool difficulty is set at 120000 by default. May I change it for better results?

The pool I'm using is nanopool.org btw.

You get low diff shares when your gpu returns a bad result. So loosen on your OC , or depending of your card, maybe you use 'bad' mem. straps.

[RuMiner]

DOKTOR83, there is a memory pill for AMD
Could you implement it on windows ? Right now it's linux only.

Cryptonights are not so sensitive to memory straps, it's good for ETH