This only makes sense if you have an iGPU, but hear me out.

My system:

CPU: Intel Core Ultra 7 265K
GPU: AMD Radeon RX 9070
Motherboard: MSI PRO B860-P WiFi
RAM: G.Skill 64GB (32GBx2) DDR5-6000
PSU: Corsair RM850X
OS: Arch Linux KDE

I see people always telling others to connect their monitor directly to the dGPU instead of the motherboard/iGPU. I decided to test this with a wattmeter, and the results were interesting.

Power consumption (whole system):

Connected to dGPU:

Idle at 60/165Hz: 33-35.5W (constantly fluctuating)
YouTube 1080p fullscreen: 73-83W (constantly fluctuating)

Connected to iGPU (motherboard):

Idle at 60Hz: 31.8W
Idle at 165Hz: 32.5W
YouTube 1080p fullscreen: 40-44W (occasionally hitting 50-52W)

Not only while playing youtube , doing any light tasks like opening new tabs, moving windows, browsing, chatgpt, claude etc all these things consume about 25-40W more when connected to dgpu directly. Also the system gets to idle power quickly when connected to igpu. Whereas with dgpu, it takes noticeably longer to drop to lower power levels.

When doing GPU intensive tasks like gaming or running LLMs, the OS (at least on Linux, should be the same on Windows) automatically runs them on the dGPU. I get the same performance, or at worst within margin of error.

So, it makes no sense to connect directly to the dGPU unless you’re only gaming. If you have mixed workloads - work, browsing, movies, youtube AND gaming , then connecting to the iGPU saves significant power without sacrificing performance where it matters.

NextDNS analytics

It feel like we’re losing to Google, day by day. They aren’t killing AOSP directly, but they are making it useless step by step.

Now it’s Google Play Services, Play Integrity checks, installation source checks… more and more apps just refuse to run without GMS. Banking apps? Most of them don’t work. And it’s only getting worse. I run vanilla AOSP on my main profile, no Play Services. I keep GMS only in my work profile for the apps that absolutely need it. But now even some regular apps that don't need any play services won’t work on my main profile anymore. They simply block your from running , like le chat.

Maps is google's most important app there is no way to run without play services. Sure we can use webview or gmaps wv, but they don't provide turn-by-turn directions. Earlier maps used to work without play services, but two years ago, an update stopped it from working. Now that old version is out of date and no longer works.

Google is slowly making GMS very important to run. The problem with GMS is they require to run as system app and has to have all the permissions by default.

Hope EU puts pressure to make google allow apps to run independently without GMS or atleast install them as user apps(like graphene os sandboxed play services).

If we keep going on like this, AOSP can only run fdroid apps in the future.

Just wanted to share my setup and see if anyone has suggestions or feedback. Also share yours.

Phone : GrapheneOS(pixel 7a)

1. No google play service on my main profile. Rethink DNS (NextDNS DoH) blocks ads, trackers, and all Google & Facebook DNS (except WhatsApp).

2. Some FOSS apps like Aurora Store & NewPipe need Google servers, so I have excluded them in rethink dns.

3. Work Profile (with Island) with GrapheneOS’ sandboxed Play Services, but I use it maybe once or twice a month only for apps that absolutely need it. It stays turned off most of the time. If an app works on main profile without any issues, will use it. If not, will try to use it in firefox (as lack of play services doesn't matter). If only app is available (and not web version) and it doesn't work on main profile, will use it in work profile.

4. Hardened Firefox fork(Ironfox) for private browsing. Main Firefox for a few services where I have to stay logged in and don't have apps or want to use their apps.

5. Network & Sensor Restrictions: If an app works offline, I block its internet access. Also, disabled sensors for apps that don’t need them.

6. Mostly use foss apps from f-droid(droidify).

7. Email: moved from gmail to protonmail

PC/laptop: Arch linux kde on pc and fedora kde on laptop.

1. Not much to say. Most used apps are firefox and Zed. I allow data collection on kde as I want them to improve it.

Home Server: Raspberry Pi 4B

1. SSH hardening: Non standard ssh port(yes, I opened the port externally because I depend on my home server and need to access it remotely). SSH keys or password+totp, Fail2Ban, ufw.
2. Services running: Arr setup(jellyfin, prowlarr, radarr,sonarr, qbittorrent), pihole, Immich, Authelia(for now). All data sensitive services behind authelia with totp.
3. Nginx Geo-blocking: Only allows access from my country IPs
4. Weekly backups because data loss sucks.

Network & Router: OpenWRT (TP-Link)

1. Not much to say: Running default firewall rules with network-wide ad/tracker blocking via pihole and some ports opened.

I ran OpenSpeedTest from my PC to my Raspberry Pi 4B, both connected via LAN to my WiFi router. The left screenshot shows the speedtest via local_ip:3000, and I'm getting the expected 1 Gbps up/down.

The right screenshot shows the speedtest via https://speed.mydomain.com. I'm confident that the connection from my PC to my home server is routed internally and not through the internet because my lowest ping to the nearest Speedtest server (my own ISP) on speedtest.net is 6ms, and my internet speed is 100 Mbps up/down. So the traffic must be routing internally.

Is there typically such a massive difference between using http://local_ip:3000 and https://speed.mydomain.com?

Additional context: The speedtest server is running via Docker Compose. I'm using Nginx (native, not Docker) to access these services from outside my network.

Edit: you just have to be selective at your victim or do mass shooting.

Basically the title