Puppy Linux Discussion Forum

https://www.pine64.org/2023/04/01/march ... t-bonanza/
https://youtu.be/SujNfkegedM

PineTab2 will be available for pre-order on April 11th
Ships with DanctNix Arch Linux with KDE Plasma Desktop (software for early adopters)
CounterPillow talks about PineTab2’s performance: video decoding, watching Youtube, browsing the web and working with PDF documents
Tested on Plebian – a vanilla Debian for RK3566 devices .

you should be warned that it won’t include any critical new functionality (at least not out-of-the-box – improvements and additional enablement are obviously possible, and even likely, between the time units get flashed at the factory and when end-users receive their units).

I chatted briefly with Danct12 earlier this week and he wishes for you to know that the PineTab 2 software is still a work-in-progress, so you should always check for updates as things are likely to progress even quicker once people get their hands on the hardware. If you have any questions concerning the build and current status of the software then you are welcome to join the #archlinux-pinephone:kde.org channel on Matrix or @archmobile on Telegram. Danct12 also wanted me to let you know that the Wi-Fi is disabled on the PineTab2 by default due to driver stability issues, so you’ll have to run modprobe bes2600 to get it working. Now, the second, and perhaps the most important thing I wanted to write before CounterPIllow shares their insight is that the PineTab2 will be available on April 11th… bundled the unicorn with a mic on its head, of course.

[Quick note: I have taken the liberty to edit CounterPillow’s contribution for style solely – none of the written contents were altered. CounterPillow’s section starts below.]
I’ve seen some talk about how the RK3566 in the PineTab2 won’t perform well for its intended purpose. While it is true that the RK3566 system-on-chip is optimised for low power draw rather than raw performance, this does not mean it’s a bad SoC for a tablet. First off, yes, it is obviously slower than a RK3399 (Pinebook Pro, ROCKPro64) running at its full clock speed. The 4 Cortex-A55 cores in the RK3566 are in-order cores, meaning they do not reorder instructions for higher performance. This makes them comparable to the four Cortex-A53 cores found on the RK3328 (ROCK64) or A64 (PineTab1), but much faster and more efficient as it is a newer core.

But will it be enough for watching YouTube? Well, if we assume hardware decode is working and we’re rendering the video through hardware scalers, we can naturally reach the advertised 4K@30 video output. But on mainline kernels, i.e. kernels that are what sits in Linus Torvalds’ git tree, we’re currently missing a driver for the “rkvdec2” hardware decode block, and even if we had such a driver, the support for its v4l2-request user space API (the interface between programs and the kernel) is still not great.

So let’s look at pure software decoding. That is, video decoding performance without fixed function hardware acceleration, just what the CPU cores can do with their vector instructions. Anyone without a PineTab2 can do this, so long as they own some other RK3566 device. I ran these tests on a Quartz64 Model A, running Plebian. For this test, I kept things simple. I decoded 1080p and 720p videos fetched from YouTube with yt-dlp at the maximum speed the CPU could muster, using an FFmpeg build from Debian Bookworm’s repositories, also built against the libdav1d from Debian Bookworm. In essence, this means FFmpeg 5.1.2-3 and dav1d 1.0.0-2. The video used for the test were the various video streams YouTube offers a 25 fps video. Five minutes of the video were decoded as fast as possible with the command ffmpeg -i videofile -to 5:00 -f null – -benchmark and the achieved average framerate was calculated with 300 seconds / rtime * 25.

A more common use case is web browsing. For this test, I used Chromium with a few flags enabled (wayland, experimental performance stuff) on a 1080p display on a kwin wayland plasma session, again on a Quartz64 Model A, which has the same SoC as the PineTab2. You can see that we get quite a smooth web browsing experience from this video capture (Ignore the crunchy looking text rendering, that’s an artifact of my $9 HDMI capture device’s 4:2:2 downscaling.):

Firefox will be slower as it really doesn’t want to do GPU rendering on Panfrost and its JS engine isn’t as optimised for ARM. Compared to PineTab1, this is a night-and-day difference. You can actually browse the web!

Another task many want to use a tablet for is to browse through documents like PDF files. One particularly heavy workload for this is tabletop gaming rule book PDFs. They’re usually richly illustrated, taking a few seconds even on x86 computers to do a page flip with some reader software. Out of curiosity, I tried the “Warhammer Fantasy Roleplay” 2nd Edition core rule book, which is my heaviest PDF file, on the Quartz64 Model A. I can’t show you an HDMI capture of the experience for copyright reasons (Games Workshop, if you wanna allow this use hit me up, I swear we’ll treat your IP with dignity, unlike some game studios), but I can describe the results in text form.

Using Okular to browse through the PDF was terrible. While Okular works fine for richly illustrated scientific papers and other such content, the might of Sigmar brought it to its knees, resulting in 5+ second page loads. MuPDF, a more minimalist document viewer, copes well however. Flipping to the next page takes about 1 second from key press to the next page being rendered. This is more than usable enough for quickly browsing to the right page during a sweaty tabletop gaming night.

The RK3566 SoC found in the Quartz64 and SOQuartz line of devices has gotten fairly good mainline support in Linux. To showcase this, me (CounterPillow) and diederik banded together to create Plebian, an installer- free Debian image for Quartz devices based on Debian Bookworm, using its unmodified kernel package.