have anybody here tested claymore with amdgpu-pro v17.50
on Ubuntu 18 beta LTS with latest kernel v4.16?

is there any improvement with "wattmann" functionalities as suggested on Phoronix? Undervolting?
https://www.phoronix.com/forums/forum/phoronix/latest-phoronix-articles/1001523-gpu-voltage-control-support-coming-to-amdgpu-driver

On phoronix they say some improvement are possible with next kernel versions
https://www.phoronix.com/scan.php?page=news_item&px=AMDGPU-Linux-4.17-Round-1

I've recompiled amdgpu module with this kind of change
https://www.phoronix.com/forums/forum/linux-graphics-x-org-drivers/amd-linux/918649-underclocking-undervolting-the-rx-470-with-amdgpu-pro-success
but in fact, I see that the numbers are already changed because of the bios power state, but that it have no effect. I could not continue investigation.

As far as I understand if one understand which memory have to be read/set to control the i2c, and understand the vbios tables (as read by AtomBiosReader), one could do what amdgpu do when interpreting the vbios pseudocode (x86 if I understand well). all normally is in the amdgpu driver, and as I suspect it could even be done with /dev/mem.
best of course would be to use sysfs peudo files calling subroutins in amdgpu, like it is already proposed for pp_*

On my Linux miners, the migration to v11.2 reduced performance. All was tuned automatically to optimise hashrate and consumption, especially the ETHI/DCRI... finally after reoptimizing, changing memfreq and intensities, it was faster and consuming less... If GPU are unstable, it may be because cards are near instability and were pushed too far with new algorithm. Review the frequencies and intensities...

I dont find b517 on Linux. Was the bug on Linux too, and how to check all is OK?


Beside that when upgrading I've seen performance (ETH only) down about 50kH/s on each RX4/580, but after relaunching my optimization process (Memfreq, dcri, ethi), I won 50kH/s. consumption seems slightly lower. I have to verify all, but it seems good on Linux+amdgpu v17.50.

Looks nice, same protocol as what I do manually...

By the way is there any way to undervolt in linux, without playing with offset and VoltageObjectInfo in the VBIOS?

@Claymore
"WattMan Support Coming For Radeon Polaris GPUs On Linux"
https://www.phoronix.com/scan.php?page=news_item&px=Auto-WattMan-Polaris
"Back in 2016 AMD introduced WattMan to their Radeon Software driver to allow for fine-tuning of GPU voltage, engine clocks, memory clocks and more. An incarnation of that is now approaching their open-source Linux driver. "

Not for now

I'll try to play with the js API, then see if I can understand what is happening...

Is there any support of java or C/C++ API for recent version : Ledger Nano S ?
I've seen Javascript and chrome embeded version, not well documented, but for C/C++ or java, it is only BTChip obsolete version...

Github have much projects, but documentation is brutal... It seems all i need would be a "Howto" setup the dev env,  a helloworld for the HID transport layer, and a reference documentation for the service (the app managing the secondary passphrase). BTChip java APi was good enought, so i could contact the Nano via WinUSB+usb4java, but even getFirmwareVersion is not compatible...

Maybe a good news on the horizon for Linux


Phoronix: GPU Voltage Control Support Coming To AMDGPU Driver

https://www.phoronix.com/forums/forum/phoronix/latest-phoronix-articles/1001523-gpu-voltage-control-support-coming-to-amdgpu-driver

with latest AMDGPU-PRO v17.50 no need of ROCm...
I've installed a blank ubuntu 16.04 with amdgpu 17.50 --compute... it works perfectly.
but with previous version ROCm was solving DAG limitation (activate big 2M pages).

I think it does already work with %NAMEOFVAR% (even on linux, this is DOs syntax)

for example in y epools.txt I have:
POOL: eth-eu2.nanopool.org:9999, WALLET: %ETHWALLETADDRESS%/%MINERNODE%/%MAIL%, PSW: x, WORKER: , ESM: 0, ALLPOOLS: 0
...

Hey, it's Sunday ! Even a miner maker have off-time.

"CUDA error - Cannot allocate big bugger for DAG. check readme for possible solutions"

I don't have Nvidia, but as said it shuld be documented in the readme.txt. I suspect
- "Set Virtual Memory 16 GB."
- need 4GB VRAM

on AMD I would suspect missing environement variables (GPU_FORCE_64BIT_PTR ...)

I use Ubuntu 16.04.3 and I have this kind of problem, linked to overclocking and similar instability problems.
The problem is that it is locking something hardware, and the kernel cannot stop the thread, nor even shutdown properly.
I used a USB watchdog, but recently i found that my Asrock H81 ProBTC include an iTCO watchdog that is just to be configured with linux watchdog.

for restart I use the restart.sh with a shutdown -r (check thay you have a restart.sh  that works, show trace to see if it is called).

if your kernel is just stuck while rebooting, try to configure the watchdog of your motheboard if there is on, or acquire a USB watchdog.

Maybe it is already configured under EthOS

The key should be to add an offset in VoltageObjectInfo (google it)
https://forum.ethereum.org/discussion/12970/undervolting-through-hex-editor-simplemining/p3
http://www.overclock.net/t/1621800/would-appreciate-help-in-finding-editing-the-voltage-offset-in-the-rx480-bios/0_100#post_26047272
https://bitcointalk.org/index.php?topic=1465128.20

I use V17.50 but for overclocking I use bios flashing, and I don't undervolt (too bad).

The amdgpu-pro v17 never worked for me on OC/UV. I've heard crimson drivers, but with older linux, did work... never tested.

On Linux, for me, the overclock and undervolt options don't work...
ADL is no more working with amdgpu-pro, and it was the lib who was doing the overclock/undervolt

some options works, because of sysfs entries, but I only see fan&temperature working.

if someone can tell me you can change clock or voltage from amdgpu, i'll be happy.

on my rig it produces
Probably you make a typo here, or else it is really an error.
can you see the cards with "atiflash -ai" ?
can you see your amdgpu cards with "clinfo"
try: clinfo|grep "Board name"
and: clinfo|grep "OpenCL"

mysterious for me

to know the package relative to amdgpu try
" dpkg -l| grep amdgpu|grep -v ubuntu16"
(the grep -v ubuntu16 is because some adgpu libs are from core ubuntu, not amdgpu-pro. check each name)
but there are also many other pakage related but without the name
then do somethin like
"dpkg --purge amgdpu-core amdgpu-pro-core  amdgpu-dkms ... and so on"
or if you are lazy and feel lucky (check the list generated before)
"dpkg --purge `dpkg -l| grep amdgpu|grep -v ubuntu| awk '{ print $2;}'` "
check if there are pachage with failed dependencies with
"apt-get install"
if there are, purge them too...

and then reinstall (--opencl=legacy and probably --headless unless you use X11 on your GPU).

I have a Sapphire RX580, so it should work...