Hi, you have to switch from gaming to compute. You can do it easily with using the AMD Software. Go unter Gaming and change from gaming to compute. Cheers CH
I'm using linux as my os, this reminds me of something I had done to fix a graphics issue the other day though. I'll give it a shot tonight and comment back the results. The issue I was having was the onboard vga connection wouldn't let me login so I reinstalled the amdgpu driver with the ./amdgpu-pro-install –compute switch to fix it. I've since reinstalled using the -y switch. Maybe I need to reinstall again using the compute switch since I'm accessing the box via ssh now...
I've got some Sapphire RX 570 Pulse ITX cards with Elpida memory and they are hashing at 29.5, running cool too at 62c. They are pulling 90w according to GPUz, at the wall I reckon its roughly 110w per card - thats including the riser card.
I have used the latest polaris bios editor which has the Ubermix memory straps pre-loaded. I'm underclocking the GPU to 1150 with 850mv and overclocking the memory to 2000 with 850mv. Not all my ITX pulse cards can achieve this, some are at 28.5 - but its still a great result.
My advice is to use windows 10, its more flexible than linux and the drivers are updated more often.
Have you verified what the effect of the memory voltage reduction is? @Wolf0 mentions in some other discussion that the memory voltage should not be reduced and that the windows tools do not really reduce the memory voltage.
Well I tried installing the amdgpu driver using the --compute switch but no luck.. It sounds like the only way to get these RX570s to hash higher is to mod them. I'm hesitant on doing that with a Linux based rig. I only used Linux because it was free vs buying a Windows 10 license. I'll have to stop being cheap and go buy one.
Is the amdgpu-pro used by your kernel? What do you get from modinfo?
# modinfo amdgpu | head filename: /lib/modules/4.10.0-42-generic/updates/dkms/amdgpu.ko version: 17.40.3.12 license: GPL and additional rights description: AMD GPU author: AMD linux driver team
If the kernel is too new it the amdgpu-pro will not compile. I have downgraded to kernel 4.10.0.
filename: /lib/modules/4.13.0-26-generic/kernel/drivers/gpu/drm/amd/amdgpu/amdgpu.ko
license: GPL and additional rights
description: AMD GPU
author: AMD linux driver team
Looks like 4.13. I tried loading 4.10 but didn't see it on the grub menu after I rebooted. Being a noob at this level of linux I'm sure I missed something. Is this how I would load 4.10?
I was able to get 4.10 loaded but I'm still showing 17mhs for all three of my cards. I might rebuild from scratch and load 14.10 from the get go.
filename: /lib/modules/4.10.0-28-generic/kernel/drivers/gpu/drm/amd/amdgpu/amdgpu.ko
license: GPL and additional rights
description: AMD GPU
author: AMD linux driver team
Have you verified what the effect of the memory voltage reduction is? @Wolf0 mentions in some other discussion that the memory voltage should not be reduced and that the windows tools do not really reduce the memory voltage.
yes I have, WattTool accurately measures voltages, I also see an almost 200w drop at the wall and a big fall in temerature
@chaffy Have you tried to rebuild the amdgpu pro using the command dpkg-reconfigure amdgpu-pro-dkms. This should build the driver for all installed kernels. If it faills for your 4.10 then it is not compatible for some reason.
If it works you should have an amdgpu.ko in /lib/modules/4.xxx/updates/dkms folder.
The boot to this kernel and you should get a version: 17.40.3.12 line in the modinfo...
Finally getting a chance to work on this a bit more. What hashrate are you getting with that script @Xcoder ? I went into my home office today and it was hot as hell. I didn't think 5 570s would produce so much heat. I've spread them out on the shelf they hang from and I keep the door open now. They run a lot cooler with this new physical setup. Hoping to get the last 3 cards added within the next month. One of my cards is running stock at 24mh/s and I'm trying to figure out which one without shutting down, unplug one, start up, and repeat until I figure it out. I just realized I hijacked my own thread! lol
Comments
- install amdgpu-pro driver version 17.40
- read VBIOS using atiflash for linux
- tune straps in VBIOS using PolarisBiosEditor and use the one click timing feature which worked great for my Elpida memory.
- write VBIOS with atiflash
- use ohgodatool to raise memory clock (--mem-clock option) and tune core clock (--core-clock option)
- use ohgodatool to reduce core voltage (--core-vddc-idx)
I have 28.5 MH/s @130W per card. There should be more room for tuning but I seem to still miss something...Cheers, Xcoder
I have used the latest polaris bios editor which has the Ubermix memory straps pre-loaded. I'm underclocking the GPU to 1150 with 850mv and overclocking the memory to 2000 with 850mv. Not all my ITX pulse cards can achieve this, some are at 28.5 - but its still a great result.
My advice is to use windows 10, its more flexible than linux and the drivers are updated more often.
If it works you should have an amdgpu.ko in /lib/modules/4.xxx/updates/dkms folder.
The boot to this kernel and you should get a version: 17.40.3.12 line in the modinfo...
777000000000000022AA1C00315A5B36A0550F15B68C1506004082007C041420CA8980A9020004C01712262B612B3715
Here you can find more hashrate of diffrent graphics card