All,
I'm looking for some help in reducing my wattage at the wall for my x8 RX580 rig.
I've recently moved from windows to SMOS and my x8 card rig is pulling 1649w at the wall, it was pulling about the same when i was using windows.
I have them set up as follows...
core =1130
Mem = 2100
Core UV = 975
Powerstage 4
TT = 60
MF = 65
Reading through the threads i should be able to knock quite a dent in this, but for some reason i can't get down.
Would appreciate some help Pls...
1 ·
Comments
https://www.overclock.net/forum/67-amd-ati/1604567-polaris-bios-editing-rx5xx-rx4xx.html
setx GPU_FORCE_64BIT_PTR 0
setx GPU_MAX_HEAP_SIZE 100
setx GPU_USE_SYNC_OBJECTS 1
setx GPU_MAX_ALLOC_PERCENT 100
setx GPU_SINGLE_ALLOC_PERCENT 100
EthDcrMiner64.exe -epool us1.ethermine.org:4444 -ewal wallet.rig -epsw x -mode 1 -ftime 10 -tstop 85 -tt 55 -fanmin 65 -fanmax 100 -cclock 1160 -mclock 2125 -cvddc 835 -mvddc 835 -asm 1 -dcri 8
https://cc.smos-linux.org/documents#panel2
Other options are ethos/hiveos, with their own techniques of undervolt/overclock.
You should be able to get Your rig consumption close to 1000W with 8x RX580. (depends on the PSUs efficiency).
it looks like SMOS 3.0 is using the amdgpu driver "built-in" method (through the sysfs) to undervolt/overclock the RX chips (available from kernel 4.17):
https://www.phoronix.com/scan.php?page=news_item&px=AMDGPU-Voltage-Sysfs
hiveos, is using Wolf's ohgodatool with a script that i posted in this forum few months ago (works with kernel 4.15 upwards):
https://forum.ethereum.org/discussion/comment/90717
if you prefer to run a standard linux distro (probably ubuntu), to use the sysfs method, you have to
- run kernel 4.17 or 4.18rc1/2
- change the amdgpu driver "ppfeaturemask" parameter:
$ cat /etc/modprobe.d/amdgpu.conf
options amdgpu vm_fragment_size=9
options amdgpu ppfeaturemask=0xffffffff
- set the desired vddc/vddci for each GPU, according to the amdgpu driver documentation: https://github.com/torvalds/linux/blob/master/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c#L459
//usage example
* set power_dpm_state
echo "performance" > /sys/class/drm/card0/device/power_dpm_state
* set pp_power_profile_mode
echo manual > /sys/class/drm/card0/device/power_dpm_force_performance_level
echo "5 1 0 0 0 1 0 0 0" > /sys/class/drm/card0/device/pp_power_profile_mode
* set pp_od_clk_voltage
echo "r" > /sys/class/drm/card0/device/pp_od_clk_voltage
echo "s 2 952 850" > /sys/class/drm/card0/device/pp_od_clk_voltage
echo "s 3 1041 850" > /sys/class/drm/card0/device/pp_od_clk_voltage
echo "s 4 1106 850" > /sys/class/drm/card0/device/pp_od_clk_voltage
echo "s 5 1168 850" > /sys/class/drm/card0/device/pp_od_clk_voltage
echo "s 6 1175 850" > /sys/class/drm/card0/device/pp_od_clk_voltage
echo "s 7 1175 850" > /sys/class/drm/card0/device/pp_od_clk_voltage
echo "m 2 2000 850" > /sys/class/drm/card0/device/pp_od_clk_voltage
echo "c" > /sys/class/drm/card0/device/pp_od_clk_voltage
You rebuilt my motivation to tweak the cards with impracticable (by me and my time) vrm controller.
https://forum.ethereum.org/discussion/18239/summer-tuned-settings-for-rx-580-cards#latest
If you have a Saphire or any other RX580 with Samsung memory you can flash it directly with this BIOS. But to be safe just download the latest version of Polaris Editor that has the one click timing feature built in. Hopefully you already have a backup your Stock BIOS saved somewhere. Open your stock BIOS in Polaris Editor. Apply the one click timing patch. Then open my BIOS in another session of Polaris Editor and arrange them side by side. Then just copy / transpose over the settings from my BIOS to the new BIOS you are creating. Leave the memory settings alone since the one click timing button took care of that.
I've flashed various brands of RX580 cards based on this method and they all delivered 31.5 - 31.7 mh/s.
I am attaching my summer and winter AMD settings for Claymore in a ZIP file. The winter settings will deliver 31.7 mh/s @ around 140 watts. The summer settings will deliver 30.5 mh/s @ around 115-120 watts. I also have Nvidia cards in the same rig with the RX580 cards and have a different Claymore batch file for the Nvidia cards. I isolate the cards via the -d switch so my six RX580 cards in a rig are called out as -d012345. The seven Nvidia cards in a rig are called out as -d6789abc. I removed these flags from the batch files attached in the zip file in this post because I don't know if you have Nvidia and AMD cards in the same rig. You don't need the -d flag if you only have AMD cards in the rig.
Note the winter settings are very aggressive and the only way to get the cards to mine successfully is to connect to the rig via RDP. If you try via VNC, Logmein etc. it will error out. But via RDP it works fine and I mined for months that way. Only drawback is when you connect via RDP to start mining Claymore will not report the AMD GPU temperatures / fan speeds for some reason. Nvidia cards are not affected by this and properly report the settings. The summer settings will work via VNC / Logmein and the temperatures / fan speeds are reported for the RX580 cards.
Make sure you have the latest AMD driver for RX580 installed and Windows 10 Build 1709. Do not install Build 1803 it is full of problems. Your cards should also all be set to Compute Mode. There is a tool that will set multiple cards via a single click - search for it on Google.
- you are right, claymore's miner undervolt capability is limited, his miner is unable to set the vddc/vddci below(around) 900mV. to be honest, i don't understand why everyone is using claymore's miner for ether, if there is a better and open-sourced miner available https://github.com/ethereum-mining/ethminer
@rmh
- right, the VBIOS surgery to insert in VRM offset and then the endless cards flashing to find the correct offset for each GPU is really a PITA
@asusrig
- to use Windows for mining nowadays when Linux does everything better ? maybe for Vega, otherwise - no way!
They are all MSI cards 7 are Hynix and the other Samsung. I brought the rig of a friend already up and running, but he's left the place I work now. I'm led to believe the cards have already been flashed.
I've just powered the rig back up in Windows to check the memory using GPU-Z and have started minestat. See attached pic of claymore settings... Thanks for this....
They are all MSI cards 7 are Hynix and the other Samsung. I brought the rig of a friend already up and running, but he's left the place I work now. I'm led to believe the cards have already been flashed.
I've just powered the rig back up in Windows to check the memory using GPU-Z and have started minestat. See attached pic of claymore settings power usage meter.
I have tried many other claymore settings to reduce the power usage but it's like it just ignores them and the usage stays high. I also have tried afterburner but to no avail....
https://bitcointalk.org/index.php?topic=1433925.0
My experience with stable mining in Windows 10:
1080 ti = up to 56 mh/s
1070ti = up to 32.5 mh/s
1070 = up to 32.5 mh/s
RX580 = up to 31.7 m/s
If you're not beating what we are getting with Windows 10 I fail to see how it is better.
i'm using only AMD RX cards, so I can't comment Nvidia.
each RX580 with 8GB memory can do 31 mh/s, especially with Micron/Samsung memory chips regardless of the OS you use. (with modified memory straps off course).
Linux advantages (just a few):
- a server OS designed to run 24/7
- you can run it from a 5$ USB flash drive
- many Linux distributions customized for mining are available, you only need to write the OS image to a USB key and configure the miner through a WebUI within a few minutes.
- much better logging/debugging capabilities
As I mentioned before, Claymore's ethereum miner undervolt implementation on Windows is limited to a certain level(s) of vddc/vddci voltages. You can try to set Your desired voltage to 800mV, but it never goes below 900mV (check the GPU-Z output).
I can push the Polaris GPU voltages close to 800mV on Linux, setting the core clocks to/or below 1100MHz mining ethash.
Regarding the power consumption, in You previous post You reported these numbers:
31.7 mh/s @ around 140 watts.
30.5 mh/s @ around 115-120 watts.
These numbers are from GPU-Z (which reports only a GPU power draw) or calculated from a rig overall consumption ?
The below is only necessary if you have a sub-3gb AMD card, remove it otherwise.
setx GPU_FORCE_64BIT_PTR 0
setx GPU_MAX_HEAP_SIZE 100
setx GPU_USE_SYNC_OBJECTS 1
setx GPU_MAX_ALLOC_PERCENT 100
setx GPU_SINGLE_ALLOC_PERCENT 100
Hope this helps.