Guys, sorry for asking again, but I can't find any documentation about the error I'm getting:
GPU1 The option -clKernel 3 is ignored for GPU1 (not enough VRAM)
It's happening with a RX580 8GB, epoc 402. I can only use clKernel 1.
Any ideas?
Have you increased your system V Memory I forget the equation but its like DAG Size * No. of GPU's + 16GB = xxx VRAM Size
FWIW: on my x10 RX580's rig I have set 90GB Virtual Memory / Page file.
Setting -eres 0 may help..
I went through a number of attempts to get all GPU's locked into clKernel 3 but in the end it worked out better clKernel 1 (for me, anyway !)
If i remember correctly, I had to config it to only apply to my GPU's which are running samsung memory. This may just be fluke but for some reason Hynix/Micron mem types were just not happy .
I got it to play ball, it went a little something like this:
clKernel = 3,1,3,3,1,1,3,3,3,1
Like on my my RX580 rig my GPU's are mapped according to memory timing/ASIC quality like this:- FAST,SLOW,MID,FAST,SLOW,FAST,FAST,FAST,SLOW,FAST (Samsung = fast) (Micron = Mid) (Hynx = Slow)
That is how I got clKernel to stick and run stable BUT after more testing clKernel 1 and setting the mt 2 = higher hashrate
Example below of my config I set a static -gt to get the rig to load up the DAG's.
AFTER I have a share then i go to work on my core and memory voltage and clock speeds in OverdriveNTool
Then I just hit the "Z" in PheonixMiner to auto-tune it. (make a seperate not of new -gt val's or revert back to old values. )
-cclock 1175,1150,1150,1150,1150,1150,1150,1150,1150,1150
-mclock 2150
-cvdcc 925,875,875,875,875,875,875,875,875,875 (Stock 1150)
-mvddc 925 (Stock 950)
-gpus 0123456789
-altinit
-mi 14
-gt 57,75,64,60,76,65,62,62,71,63
-sci 0
-clKernel 1 <<===== HERE
-clgreen 1
-clNew 1 <===== This Generates new Kernel use when testing clKernel 3 > 1
-clf 1 <== Syncs over gpu's
-minRigSpeed 285
-dagrestart 2
...
-powlim 0
-mt 2 <====== HERE
-leavemt
FYI: Pro Tip take a look at OverdriveNTool As it
slashed my power draw down easily by like 33% from ~120w - 130W / card. Now, down to like ~95w / card
plus I can push the memory clocks up between 2170 (on my SLOW GPU) 2230 (On My FAST Gpu's) but i generate to many HW Errors and stales/incorrect shares start to crop up so Im happy @ *:2150 ..
GPU0: 51C 24% 94W, GPU1: 46C 4% 92W, GPU2: 57C 44% 98W, GPU3: 54C 36% 100W, GPU4: 50C 20% 88W, GPU5: 53C 38% 87W, GPU6: 53C 26% 88W, GPU7: 53C 32% 91W, GPU8: 49C 16% 87W, GPU9: 53C 36% 98W
GPUs power: 921.7 W; total power: 1066.7 W; cost: 4.61 USD/day
~ 300Mh/s @ 1066 W + 140w over estimate = 1206w MAX
Eth speed: 297.516 MH/s, shares: 26486/2/0, time: 27:23
A More of a complete answer, however I hope this helps you get ideas on where to look for optimisation of your mining rig.
Just get it running stable clKernel 1 1st then get it to accept -mt 2 (maybe different for you clKernel 3 maybe better and mt 0 , test i guess..)
once, you happy with hashrate, head into OverDriveNTool to start scalping the power draws down / per GPU.
That is where i have made most progress is in regards to power draw. Multiple savings from 24/7 mining over multiple rigs you get the picture.
Thanks majika for such a detailed answer. Really appreciated. But my question is related to why I can't use Kernel 3. I've already fine tuned my RX580 and is running at 32.745 MH/s at 88W without any issues nor incorrect/rejected shares (
), and I could be breaking the 33 if I could use Kernel 3 (I've benched it with a lower epoc and is working)....but still wondering why I can't use kernel 3 on epoc 402 with this card...
I've tried adding more virtual memor...eres 0... etc etc....without luck.... idk why Phoenix is telling me I have no enough VRAM for running Kernel 3....and I cant find much on internet about it....