Any word on amd vega hash rates?

[Shnikes101]

Hi, I have a weird problem with my vega64. (only 1 card)
I use the Vega64SoftPowerTableEditor to edit MorePowerVega64_142_Modded.reg,
Changed some value:
P1 to 991  / 810
P2 to 1084 / 820
P3 to 1138 / 830
P4 to 1170 / 840
P5 to 1240 / 860
P6 to 1320 / 880
P7 to 1408 / 910
After all settings, my vega64 can work at 1950+ h/s, but voltage is 0.988v in HwInfo64, not 0.910v.

The core / ram clock is correct.
Card: Silver Vega64 air cooler w/ stock BIOS
OS: Win10 64bit
Drv: AMD blockchain beta (Aug/23)
Reboot - did.
Reset wattman and set to manual and apply - did.
What I lost?    thanks.

Make sure your "memory" voltage is set at 910

[rednoW]

2 Hellae
No need to set memory 1100 or 1050mhz via reg mod. It can be done later with overdriveNtool. It is more safe

[dallase]

Getting 38.5 Mh/s UBIQ on Vega 56.

Pure open source..  rocm-opencl + sgminer-gm

pp_table being a bitch, so I'm OD'ing the mclk to where I want it, and setting the sclk to level 3.

# echo 17 > /sys/class/drm/card0/device/pp_mclk_od
# /opt/rocm/bin/rocm-smi --setsclk 3
# /opt/rocm/bin/rocm-smi --setmclk 3
# ./rocm-smi

====================    ROCm System Management Interface    ====================
================================================================================
 GPU  DID    Temp     AvgPwr   SCLK     MCLK     Fan      Perf    OverDrive  ECC
  0   687f   60.0c    120.0W   1269Mhz  936Mhz   88.63%   manual    0%       N/A     
================================================================================
====================           End of ROCm SMI Log          ====================


Once I get the pp_table working right, I'll be able to lower my voltage.

[cafemax]

Make sure your "memory" voltage is set at 910

OK, I fixed it. 

[l1xx]

Getting 38.5 Mh/s UBIQ on Vega 56.

Pure open source..  rocm-opencl + sgminer-gm

pp_table being a bitch, so I'm OD'ing the mclk to where I want it, and setting the sclk to level 3.

# echo 17 > /sys/class/drm/card0/device/pp_mclk_od
# /opt/rocm/bin/rocm-smi --setsclk 3
# /opt/rocm/bin/rocm-smi --setmclk 3
# ./rocm-smi

====================    ROCm System Management Interface    ====================
================================================================================
 GPU  DID    Temp     AvgPwr   SCLK     MCLK     Fan      Perf    OverDrive  ECC
  0   687f   60.0c    120.0W   1269Mhz  936Mhz   88.63%   manual    0%       N/A      
================================================================================
====================           End of ROCm SMI Log          ====================


Once I get the pp_table working right, I'll be able to lower my voltage.

If you are mining UBQ, you don't really need to use blockchain driver. And with the regular driver the wattman actually works, so a powertable mod is not needed as well ...
But this is a lot of wasted potential tbh.

[dallase]

Getting 38.5 Mh/s UBIQ on Vega 56.

Pure open source..  rocm-opencl + sgminer-gm

pp_table being a bitch, so I'm OD'ing the mclk to where I want it, and setting the sclk to level 3.

# echo 17 > /sys/class/drm/card0/device/pp_mclk_od
# /opt/rocm/bin/rocm-smi --setsclk 3
# /opt/rocm/bin/rocm-smi --setmclk 3
# ./rocm-smi

====================    ROCm System Management Interface    ====================
================================================================================
 GPU  DID    Temp     AvgPwr   SCLK     MCLK     Fan      Perf    OverDrive  ECC
  0   687f   60.0c    120.0W   1269Mhz  936Mhz   88.63%   manual    0%       N/A      
================================================================================
====================           End of ROCm SMI Log          ====================


Once I get the pp_table working right, I'll be able to lower my voltage.

If you are mining UBQ, you don't really need to use blockchain driver. And with the regular driver the wattman actually works, so a powertable mod is not needed as well ...
But this is a lot of wasted potential tbh.

My end goal is not to mine UBIQ, nor is my goal to use windows.

[BitTeo]

Hi all,

I've just posted a guide on reddit to help others. It will get more visibility and traction there, I hope.
There's some info I don't know like the V64 power consumption at .875 VDDC, so you're all invited to jump in and add your input.

My .reg file is there to download. Added another one for 1050 MHz if your card doesn't support the 1100.

https://www.reddit.com/r/MoneroMining/comments/74hjqn/monero_and_vega_the_definitive_guide/

great! but,vega 64  use only 130W?? i guess 140W or more.

Only 130 watt ? wow is very low result .

[espen83]

So.. does anybody think that there is an artificial throttle in the vega cards when it comes to mining?

[Hellae]

2 Hellae
No need to set memory 1100 or 1050mhz via reg mod. It can be done later with overdriveNtool. It is more safe

Thanks. On V64 the memory voltage can also be set at 905, sicne it is just a minimum mV and not actual voltage?

[DonBit]

Getting 38.5 Mh/s UBIQ on Vega 56.

Pure open source..  rocm-opencl + sgminer-gm

pp_table being a bitch, so I'm OD'ing the mclk to where I want it, and setting the sclk to level 3.

# echo 17 > /sys/class/drm/card0/device/pp_mclk_od
# /opt/rocm/bin/rocm-smi --setsclk 3
# /opt/rocm/bin/rocm-smi --setmclk 3
# ./rocm-smi

====================    ROCm System Management Interface    ====================
================================================================================
 GPU  DID    Temp     AvgPwr   SCLK     MCLK     Fan      Perf    OverDrive  ECC
  0   687f   60.0c    120.0W   1269Mhz  936Mhz   88.63%   manual    0%       N/A     
================================================================================
====================           End of ROCm SMI Log          ====================


Once I get the pp_table working right, I'll be able to lower my voltage.

So you're not using amdgpu-pro drivers?  Please could you share how you got OpenCL working in Linux? Sometime ago I've tried phoronix's guide using the hacked kernel and MESA + Clover + amdgpu stack with sgminer-gm but it didn't work.

[dallase]

Getting 38.5 Mh/s UBIQ on Vega 56.

Pure open source..  rocm-opencl + sgminer-gm

pp_table being a bitch, so I'm OD'ing the mclk to where I want it, and setting the sclk to level 3.

# echo 17 > /sys/class/drm/card0/device/pp_mclk_od
# /opt/rocm/bin/rocm-smi --setsclk 3
# /opt/rocm/bin/rocm-smi --setmclk 3
# ./rocm-smi

====================    ROCm System Management Interface    ====================
================================================================================
 GPU  DID    Temp     AvgPwr   SCLK     MCLK     Fan      Perf    OverDrive  ECC
  0   687f   60.0c    120.0W   1269Mhz  936Mhz   88.63%   manual    0%       N/A     
================================================================================
====================           End of ROCm SMI Log          ====================


Once I get the pp_table working right, I'll be able to lower my voltage.

So you're not using amdgpu-pro drivers?  Please could you share how you got OpenCL working in Linux? Sometime ago I've tried phoronix's guide using the hacked kernel and MESA + Clover + amdgpu stack with sgminer-gm but it didn't work.

No pro drivers.   No Mesa or Clover needed...  Dont need GUI, no OpenGL, etc. 

Code:

# ldd sgminer  | grep libOpenCL
	libOpenCL.so.1 => /usr/lib/x86_64-linux-gnu/libOpenCL.so.1 (0x00007f208a3d3000)

# ls -la /usr/lib/x86_64-linux-gnu/libOpenCL.so
lrwxrwxrwx 1 root root 40 Sep 29 10:27 /usr/lib/x86_64-linux-gnu/libOpenCL.so -> /opt/rocm/opencl/lib/x86_64/libOpenCL.so

# uname -a
Linux localhost 4.11.0-kfd-compute-rocm-rel-1.6-148 #1 SMP Wed Aug 23 12:00:35 CDT 2017 x86_64 x86_64 x86_64 GNU/Linux

# dpkg -l rocm* compute-firmware llvm-6* *amdgpu* hcc* hip* libclang-* linux*
in  compute-firmware                     <none>                  all                     (no description available)
ii  hcc                                  1.0.17312               amd64                   HCC: An Open Source, Optimizing C++ Compiler for Heterogeneous
ii  hip_base                             1.2.17305               amd64                   HIP: Heterogenous-computing Interface for Portability [BASE]
ii  hip_hcc                              1.2.17305               amd64                   HIP: Heterogenous-computing Interface for Portability [HCC]
ii  libclang-6.0-dev                     1:6.0~svn314833-0~x~pad amd64                   clang library - Development package
ii  libclang-common-6.0-dev              1:6.0~svn314833-0~x~pad amd64                   clang library - Common development package
ii  libdrm-amdgpu1:amd64                 2.4.83+git1709282014.1d amd64                   Userspace interface to amdgpu-specific kernel DRM services
ii  linux-headers-4.11.0-kfd-compute-roc 4.11.0-kfd-compute-rocm amd64                   Linux kernel headers for 4.11.0-kfd-compute-rocm-rel-1.6-148 on amd64
ii  linux-image-4.11.0-kfd-compute-rocm- 4.11.0-kfd-compute-rocm amd64                   Linux kernel, version 4.11.0-kfd-compute-rocm-rel-1.6-148
ii  llvm-6.0                             1:6.0~svn314833-0~x~pad amd64                   Modular compiler and toolchain technologies
ii  llvm-6.0-dev                         1:6.0~svn314833-0~x~pad amd64                   Modular compiler and toolchain technologies, libraries and headers
ii  llvm-6.0-runtime                     1:6.0~svn314833-0~x~pad amd64                   Modular compiler and toolchain technologies, IR interpreter
ii  rocm-dev                             1.6.148                 amd64                   Radeon Open Compute (ROCm) Runtime software stack
ii  rocm-device-libs                     0.0.1                   amd64                   Radeon Open Compute - device libraries
ii  rocm-opencl                          1.2.0-1430311           amd64                   OpenCL/ROCm
ii  rocm-opencl-dev                      1.2.0-1430311           amd64                   OpenCL/ROCm
ii  rocm-profiler                        5.1.6400                amd64                   The ROCm-Profiler is a command-line profiler for profiling 
ii  rocm-smi                             1.0.0-25-gbdb99b4       amd64                   System Management Interface for ROCm
ii  rocm-utils                           1.0.0                   amd64                   AMD ROCm utilities - Utilities for ROCm platforms

I have the m-bab 4.12.14+ kernel installed also, but was having issues detecting my GPUs, even though lspci showed them, amdkfd didnt see them.. and because of that clinfo would segfault.   I tried rx470, rx570, and rx vega, and no luck.    I sort of wanted to play with the 4.12.14+ with the DC/DAL patches for a full mesa opengl wayland, but I can wait for v4.15 for all that.  Its not necessary for mining.

[rednoW]

2 Hellae
No need to set memory 1100 or 1050mhz via reg mod. It can be done later with overdriveNtool. It is more safe

Thanks. On V64 the memory voltage can also be set at 905, sicne it is just a minimum mV and not actual voltage?

Yes, just make it equal to gpu voltage.

[Mashy81]

To all the contributors to this thread. Thanks soo much.
I am a noob at mining so I just stuck to the basics.
My setup is vega 56 just 1 because its my pc.
Wattman settings from p26.
HBCC
Blockchain driver.
NO reg mods and original veha 56 bios
I got 1980h but couldn't keep it stable.
In the end I played around with settings.
1320 gpu
950 HBM
3500rpm fan
-10 power
HBCC max (mine is 16g)
No stable 1800-1840hs.

I did try the lastest amd driver and it was shit compaired to blockchain

[STT]

So you're not using amdgpu-pro drivers?  Please could you share how you got OpenCL working in Linux? Sometime ago I've tried phoronix's guide using the hacked kernel and MESA + Clover + amdgpu stack with sgminer-gm but it didn't work.

I'd be happy just using amdgpu-pro drivers.   For some reason I fail to get Vega recognised, is there some easy guide to do it from scratch obviously I'm making some basic mistake.  

So.. does anybody think that there is an artificial throttle in the vega cards when it comes to mining?

Only 130 watt ? wow is very low result .

130w on vega use is feasible if you are down to 900mv it would be about normal.   The memory goes all the way to 1100mhz for me on this volt and that seems to be the throttle.  I'm interested to hear if anyone got their memory to 1150 or higher but it seems not possible at 900mv at least.   I think its capped quite heavily beyond that anyway

[Shnikes101]

130w on vega use is feasible if you are down to 900mv it would be about normal.   The memory goes all the way to 1100mhz for me on this volt and that seems to be the throttle.  I'm interested to hear if anyone got their memory to 1150 or higher but it seems not possible at 900mv at least.   I think its capped quite heavily beyond that anyway

I've yet to see anything higher than 1100 on mem on any of the drivers

[ChipDiode]

130w on vega use is feasible if you are down to 900mv it would be about normal.   The memory goes all the way to 1100mhz for me on this volt and that seems to be the throttle.  I'm interested to hear if anyone got their memory to 1150 or higher but it seems not possible at 900mv at least.   I think its capped quite heavily beyond that anyway

I've yet to see anything higher than 1100 on mem on any of the drivers

Only if you apply more voltage, otherwise not yet possible with undervolt.

[Mashy81]

I have notice on my card its a waste of time trying to get 1100 on the memory. It doesn't give any increase in hash. Just makes system more unstable. I got it to 1050 with no increase around 950 is the sweet spot
I have now improved a little on my settings they are now.
1312 gpu  just min and max on state 4
940 on memory
-10 power

Thats it and now i am getting 1850h/s stable
About 140w power

[yobil]

Hi everybody.
Firstly, thx for this thread.

I have a mb x370 gaming pro carbon and a Vega 56 (sapphire)
I manage to get a steady 1800h/s 1850h/s with the tuning from hallae

Now I bought a second card,vega 56 (gigabyte)
I can't manage to get 1800h/s on both. (I have 3100h/s)

I disable crossfire, and deactivate/reactivte HBCC each time I start.
On both card i'am with intensity 2016/1600

In wattman i notice one thing, on the second card I can't go upper that 920Mhz for the memory. ( but the first one handle the 950Mhz).
Do you think is a GPU default?

Thx for your answers!

[Django69]

flash bios

[Ursul0]

Anyone has any FEs here? How's your experience?
(and yeah... now I know it was really poor investment on my part)
I'm getting really weird behavior with both existing drivers: it barely holds its mem clock, slight decrease in voltage or whatever changes I introduce makes it hang or in the best case hugely underperform.
Any thoughts/advice, besides "through this beautiful piece of crap into garbage"