How to overclock Vega on Linux...?

this forum rocks, even when I was upset about the locked bios people didn’t respond with Hate even tho it might have been deserved.

thats great news to hear there is a power limit patch coming in a future kernel. I got some help from https://linuxconfig.org/overclock-your-radeon-gpu-with-amdgpu and now I can monitor gpu with a script containing:

while true; do
cat /sys/kernel/debug/dri/0/amdgpu_pm_info;
sleep 5;
done

and I can now also overlcock gpu core and memory via sudo su with
sudo echo “8” > /sys/class/drm/card0/device/pp_sclk_od
sudo echo “18” > /sys/class/drm/card0/device/pp_mclk_od

Next I just need to figure out how to increase the power usage to 100% in Ubuntu 18.04 with Kernel 4.17.12 if possible or even better replace my power play table with the one I use in windows:grin:
two weeks ago I water cooled this card so GPU core runs at 33c and hbm at 35c maximum when pulling 390watts running Furmark with -12mv UV, 1780mhz core, 1115mhz HBM using modified PP Table with 400A current limit and +142% power usage from windows10.

1 Like

Well, you can always patch the current kernel, as I do not see this patch included in the 4.19.

Also, you could do an OC with better granularity by specifying the frequency and voltage to the core:
echo "s 7 1209 900" > /sys/class/drm/card0/device/pp_od_clk_voltage
and memory:
echo "m 2 1850 850" > /sys/class/drm/card0/device/pp_od_clk_voltage
see: https://wiki.archlinux.org/index.php/AMDGPU#Overclocking

As for powertables, I do not know, if that is possible, but there was a hack (when vega arrived, no idea if there is a better workaround) that made it possible to load a custom bios after post but before the linux kernel was loaded.

As for monitoring, I recommend ROC_smi


just unpack the scripts (shell/python) and execute them in the command line with the
watch -n 1 /opt/rocm/bin/rocm-smi -a

A slightly simpler way of doing this is:

watch -n 5 cat /sys/kernel/debug/dri/0/amdgpu_pm_info

That’s how I’ve been monitoring my Vega, along with some XFCE sensor applets…

2 Likes

Read this:

https://wiki.archlinux.org/index.php/AMDGPU#Overclocking

AS far as i have heard, linux is the only OS where you can mod the VBIOS by overriding the VBIOS in RAM at Boot with your Mod.

As far as i understand the matter; BIOSes get read and put into main System RAM as a very early Step after PowerOn.

The Vega Bios Check seems to happen in Hardware before or at that loading time.
“Practically”, the Driver then loads that VBIOS from RAM and does it’s thing.

RadeonProfile is trying to fill the niesh of OC Software on linux.

Though the OC UI didn’t really work that well a few month ago.

2 Likes

Now, those changes seem to be included in the recent 4.20rc in arch (linux mainline) yet, I did not manage to make the power draw go over the limit set in bios (165W).
I can OC and undervolt just fine.

Are there any other required steps to make it work?

Hey I’m compiling the 4.20 mainline at the moment… hopefully I can change this here… Finding my card throttling from powerlimit and not thermals also.

EDIT:
Running at 300W powerlimit on Manjaro linux right now. Pretty sure this is a record for VerusHash aswell… 1600 Hs/s kicking the pants off any nVidia card I know of… will verify though… this is just PL increase no overclock, over STRIX factory OC.

Cant get powerlimit over 300W … seems to be limiting factor. Even setting clocks over 1700Mhz it caps out at 300w, and powercap is on 390W. :frowning:

Quite a few weird issues actually … Seems the best thing to do, is just use 4.20 mainline kernel and increase PL to 350w … touching anything else is bugging out and setting the card to p state 6 around 200w power draw.

2 Likes

How do you enable the power limit increase?
I have set the kernel params
radeon.cik_support=0 amdgpu.cik_support=1 amdgpu.ppfeaturemask=0xffffffff
to be able to overclock, yet if I try to set
echo 210000000 > /sys/class/drm/card0/device/hwmon/hwmon2/power1_cap
I get
bash: echo: write error: Invalid argument

What did I miss?

enter su?
sudo did not work for me… I copy pasted the number from powercap_max then changed from 39 to 35, to make sure the amount of 0s was correct… the actual OCing for me seems totally broken. Just compiled mesa-git over night to see if this helps any.

I am doing all of this as root, I can decrease the power cap but cannot increase it. What does you power_cap_max contain? Mine is 165

#!/bin/bash
echo “1” > /sys/class/drm/card0/device/hwmon/hwmon1/pwm1_enable
echo “167” > /sys/class/drm/card0/device/hwmon/hwmon1/pwm1
echo “350000000” > /sys/class/drm/card0/device/hwmon/hwmon1/power1_cap
echo “m 3 1100 1100” > /sys/class/drm/card0/device/pp_od_clk_voltage
echo “c” > /sys/class/drm/card0/device/pp_od_clk_voltage

is working for me … although its not a massive increase in performance anyway… core clocks wont budge for me, increasing them over stock just locks it at 200w :frowning: my max is 39xx …
I used the linux-mainline from AUR in manjaro package manager

1 Like

Frankly, I found the easiest thing to do was to flash a Vega 64 BIOS on to my 56. That increases the power limit to 225 watts and overclocks the HBM to 945MHz - but make sure you have Samsung memory if you try it.

tied thw wattman, its ok… but needs work… I already have a 64 STRIX cant really pdate the bios to anything higher wattage as I think its already the highest you can go. Will just wait for 4.20 stable… card is fast enough at stock tbh.

I’m currently also waiting for 4.20 stable as i’m currently stuck with undervolting my RX580 to get more mhz while being stuck in the current power limit.

everyone keeps saying that but is it really a hard requirement? My 56 Red dragon has hynix ram but I can clear benchmarks at 940 just fine (950 when I’m lucky) I run it daily at 930 right now. I’ve been wondering it the voltage increase from using a 64 bios might let me push it to 950 stable.

It seems I have no choice, as the vbios limit is still 165W, which is quite limiting most of the time. Especially considering the performance bump proton imposes, I do not need anything that would cripple my framerates even more.

unfortunately until we can raise the power limit increasing the clocks is pretty much useless, at least for my vega64 since the card just runs into power limit.

No, it’s not. It’s just that your odds of success are better than if you have Hynix.

Really? When I applied the Vega 64 BIOS, my power limit jumped to 225w in Linux and Windows.