My complete guide for setting up the ASRock BC-250. You can’t beat its price for performance, and it’s perfect for those who love projects.
TLDR (for those with BC-250 experience):
For those looking for a full guide, with all the juicy tips and tricks, please continue…
Chipset -> Integrated Graphics Controller -> Forces
Don’t bother with buying two fans. Only one really makes a difference. See very helpful Reddit post.
Out of the box, the BC250 GPU is locked to 1500MHz. The cyan-skillfish-smu GPU governor allows for dynamically scaling the GPU clock speeds to match demand. Highly recommended for both decreasing idle power usage, and dramatically increasing game performance.
Once installed, edit the config file: /etc/cyan-skillfish-smu/config.toml
These are extremely safe defaults, meant to work on everyone’s BC250. However, there’s a good chance you can squeeze out more performance with less power while also improving your system’s thermals.
Begin with updating the frequency-range. The max of 1850 MHz is a sweet spot once you enable the additional GPU CUs (compute units) in the next step. Going beyond it doesn’t improve performance much and only dramatically increases power-draw and heat. If you don’t enable extra CUs, then you’ll need to bump the max MHz to 2000 (safe) and up 2200 MHz if you can. But realistically there is no reason to do this once you enable the additional CUs.
We’ll start by lowering the minimum frequency-range. This will improve power-draw on idle and in less-demanding games. Follow the process below.
Keep repeating the below steps until your system becomes unstable and crashes:
I was lucky enough to have a system stable at 350MHz. But your milage may vary.
Next, I like to adjust the load-target values. This is entirely personal preference, feel free to skip. These are percentage values of GPU utilization. Upper defines at which percentage the GPU must be used to clock up. Lower defines the percentage of utilization for clocking down. The defaults are very unwilling to clock down, even in lesser demanding games.
These settings work perfect for me. It cranks the GPU just enough for less demanding games without giving it more power than it needs.
You’ll find safe-points defined for many GPU frequencies. Again, the voltages are very safe defaults, you can probably get away with lowering many of them to achieve better thermals and lower power without compromising performance.
Work through them one at a time. Drop it by increments of 50mV. Then test each frequency to ensure system stability. This example forces the GPU to a fixed frequency of 500 MHz.
To ensure that frequency applied run this command:
The current GPU frequency is the one with the (*). Continue decreasing the voltage by 50mV until the system becomes unstable; then revert to last working config and move on to the next safe-point voltage value. Once you are finished, you should have a stable highly optimized experience tuned to your BC-250.
Once you are done testing specific frequencies, disable performance mode.
Now that all your settings are dialed in, we are safe to have them persist on boot.
Make sure that your settings are stable before enabling, otherwise you’re gonna have a bad time if you get stuck crashing on every reboot.
By default the BC-250 ships with 24 CUs (compute units) on the GPU. In comparison the PS5 ships with 36 CUs. These binned crypto boards were stripped down, however the underlying hardware is still there–we just need to enable them.
Not all boards are created equally, some may work with all 40 CUs perfectly, others will only work with 38 or 36. Either way there’s lots of performance and better efficiency to be had by enabling these units.
BC250 Live Manager allows us to enable CUs on the fly.
Download the official script and make it an executable:
On Bazzite, it will ask you to install UMR, luckily it will run you through the installation process in the interactive UI.
On CachyOS install it from the AUR:
Once installed you can start the interactive UI again.
To quickly figure out if you won the silicon lottery, press ‘f’ to enable all CUs. You’ll know very quickly whether you did. If your system didn’t crash or freeze, you won! And you are safe to press ‘w’ to write the table and ‘i’ to install the service (to apply settings on boot).
If you weren’t so fortunate, you need to test each pair of CUs individually until you find the faulty one(s). Once you do just leave them blank and don’t enable them. After working your way through all the CUs and found a stable config, write the table and install the service.
Congradulations! Enjoy the free performance and power efficiency gains!
bc250-smu-oc tool allows users to overclock and undervolt their CPUs.
Bazzite (yes, there’s a lot more steps):
Navigate to fedora.pkgs.org. Search for “stress” and find the matching Fedora release (probably 44 or 43). Click the x86_64 rpm package to go to its page. Scroll to the bottom and find the “Download” section, then copy the “Binary Package” link.
Download the stress RPM package to a temporary folder:
Extract the RPM package to retrieve the underlying binary:
Copy the binary to the a writable location. If ~/.local/bin doesn’t exist, create it with mkdir -p ~/.local/bin
Run stress to ensure that stress is available and ready to use. You should see a list of available commands.
Now that the dependencies are installed, we an install bc250-smu-oc:
The script is hardcoded to read the stress binary at /usr/bin/stress since it doesn’t exist, the program errors out. To fix this, edit stress_helper.py.
Now that the bc250-smu-oc packages are installed, we are ready to begin overclocking & undervolting!
Stress will be used to stress-test the CPU and evaluate the most optimal overclock settings that are stable for your BC-250. It will also calculate how far it can undervolt and be stable.
If this command crashes, try re-running the command with –vid 1300. If it still is not stable, reduce the target frequency. To be easy on your system, you should stay below 1300 mV Vid.
If you wish not to overclock but only undervolt you may use these settings to stay at stock frequency.
After your detection completes, you will have a overclock.conf file. Give it a look! These are the results I got.
By default the system allows for up to 1GB of shader-cache to be compiled. After that it forgets old shader-cache and rebuilds the new ones on top of it. Heavy games will require more shader cache so I’ll increase the max shader-cache size.
Edit /etc/environment add this line:
I like to give it 5GB, but feel free to bump it up. This is just the max potential size of the shader-cache, so you aren’t wasting any space by raising it right now.
Source: Hacker News — This article was automatically imported from the source. Read full article at original source →
