Repository navigation
Panel Benchmarks #134
Replies: 8 comments 1 reply
|
These figures make the difference much easier to understand. Calling both generations “Android wall panels” hides a very large gap; the FinalBenchmark result is much closer to the night-and-day difference you see in use. It also supports showing the SoC and CPU core generation in the profile-driven panel details, rather than leaving people to decode Rockchip product numbers. The new 5.5-inch panel from #129 is a good example of why that matters. If you still have the results page, could you add the exact FinalBenchmark 2 version and say whether these are single runs or averages? That would be enough context to preserve the comparison later. I would leave the 100 MHz difference open for now, as the firmware clock table, governor or thermal state could explain it as readily as manufacturing variance. |
|
FB2 version was 0.5.1 and I ran the "High Accuracy - Slow" version of the benchmark. I ran 3 times and these are the average (The GB scores are just ones), although there was less than a couple of points variation with each run. |
|
Oh, one last thing that I think just demonstrates the point - It takes longer for adb install to install FB2 apk on the NSPanel Pro 120 than it takes for the benchmark to run on the rk3576 panels. 🤣 |
|
I have a rk3576 based device and the performance is great for testing iterative builds but that performance is completely wasted for day to day use. Worse, the rk3576 Android 14 class devices start at £275 with tax and shipping from an unknown vendor in China vs a Sonoff device that can be delivered tomorrow from Amazon for £95. I have 8 Sonoff panels and they are easily capable of running complex dashboards with multiple live video streams and the hardware looks good too IMHO. It is certain that HA will continue to load functionality to the point the Sonoff panels are no longer viable but with the ha-paneld entity filter I no longer need to look for replacement faster panels for the foreseeable future. |
|
Oh, I love the NSPanel devices, but maybe I am doing something wrong - even a basic dashboard with <10 entities and a few cards runs like garbage. I don't need it to be a super responsive dashboard with cameras and stuff, just needs to have cards so i can turn lights on and off reliably :) What OS release are you running, and maybe I'll try to factory reset one of mine and start over. |
|
Hmm. That is curious and now I need to know more. Your experience could also be afflicting others and I am pretty confident I can fix it. Apologies for impromptu shaky camerawork but this is how responsive my Sonoff 86P panels are. My HA has about 4-5k entities and other than the climate tab (which has a really complicated custom UI element), each tab loads in under 500ms (yes those are 2 live video feeds) and touch latency is probably under 250ms IMG_25072.movMy lovelace dashboard resource list is not small either and looks like I used to run a secondary HA server just for panels with a smaller curated resource list but in my testing that only impacts first load and not ongoing performance. It's a public holiday tomorrow so I can take a look at this if you want to DM me on https://discord.gg/n342kY8mm |
|
Can you elaborate a bit about your camera setup, that looks quite fast! For me cameras still load slow and the delivery guy is already at the end of the street before my doorbell camera is shown 😉 Which cards do you use for the camera and which type of stream (webrtc?)? |
|
I exclusively use WebRTC Camera on wall panels and I keep Advanced Camera Card around but I value low latency and quick loading far more than all the ACC widgets that I will never use. The cameras are mostly fast because I run my own go2rtc and the WebRTC Lovelace card allows you to specify the exact WebRTC provider HTTPS URL. HA does a lot of invisible proxying to try to get push video streams to work over CasaCloud, CGNAT and HTTP and that costs a lot of setup time. WebRTC requires HTTPS so if your certificates are on a reverse proxy then the video stream goes that way for the initial connection. A direct UDP connection may eventually be negotiated under the right conditions but that part is almost impossible to debug. IPv6 and HTTPS certificates on my go2rtc server eventually delivered first time, every time, video for me from all devices on any network. |
|
I owe you the version you asked for. My 120 is on Android 8.1, with root added. I doubt the OS release is what separates our two experiences, though. Dashboard rendering is almost entirely WebView work, and that 120 runs WebView 138, which it only has because root let ha-paneld install a newer one. Check yours under Android settings, Apps, Android System WebView. If it is far older than 138, that is the first thing I would change, since an old WebView makes a panel feel exactly as you describe however few cards are on the page. rc3 can offer to install a known-good WebView itself where it has permission to install apps, which is worth trying before a factory reset. Your benchmark numbers are real and a 120 is a slow device, but 15 against 75 on the SoC does not explain ten entities being unusable on its own. |
Uh oh!
There was an error while loading. Please reload this page.
I have been frustrated with how slow the Sonoff NSPanel Pros are, but never have been able to articulate it. Since I just got a 5.5" rk3576 Android 14 to replace one of the NSPanels (which is night and day difference) I tried to run a few benchmarks.
For what it is worth neither PCMark nor Geekbench 5 even run on the NSPanel, so resorted to FinalBenchmark 2 for it. Also for some reason the cores on the 5.5 Android are clocked 100Mhz higher than the 10.5 - no idea why - Could be manufacturing variance.
NSPanel Pro 120 (US Version):
5.5 Android rk3576
10.5 Android rk3576s
All reactions