Hi all,
I am pretty happy with my stock Reform, but am tempted by the A75 in the upgrade chip. How is software video decode in web browsers like Firefox? is 1080p achievable? This isn’t a dealbreaker for me but it is one of those little quality of life things I like about more computing power. On the stock chip MPV, Clapper, and VLC can push things along 1080 no problem, so I know those will work well.
What are people’s general impressions?
I also started out with the imx8mq and am familiar with its limitations. Right now, my Reform uses the a311d while my second board has an ls1028a. When I still had all three platforms around I was looking for benchmarks which would make it easier to decide for a platform. Unfortunately, this is not that simple. For example the phoronix-test-suite contains a benchmark that uses mpv to measure decoding framerates. But according to the mpv developers that benchmark is faulty: regression: --term-status-msg missing from --log-file since 2fd34889 · Issue #13038 · mpv-player/mpv · GitHub
So if you know a better benchmark, I could try to run it on a311d and ls1028a so that you have some real numbers to decide.
Until then, here is my “general impression”: it depends. 
So yes, the imx8mq can do 1080p (even in software) but it depends on the codec, the framerate and the encoder settings. It also very much depends on whether you do hardware decoding or not. If your material is 1080p h264 then the imx8mq can even decode 60 fps without any issues. Theoretically the hardware decoder (hantro) can also do h265 but there are still bugs on the software side which result in really weird artifacts and color shifts depending on the video you try playing. So on my imx8mq I was preferring h264 and when it had to be h265 then i had to reduce resolution to 720p because that had to be software decoded (mpv). And then there is also the issue that firefox will not utilize the hardware decoder of the imx8mq and even though mpv managed to do 1080p@30 in software (for h264), firefox is far from the performance of mpv.
The main videos I watch are twitch (1080p@60 h264), youtube (1080p@60 vp9) and some anime (1080p@23.976 h265 10bit) and the ls1028a has no hardware decoder. I’m also not much interested in watching youtube and twitch in a browser, so all my tests on ls1028a were with mpv and the yt-dlp backend. With that in mind, I can tell you that the ls1028a was not powerful enough to decode 1080p@60 h264 twitch streams, for example. I have some harder numbers for some cpu-loads which I documented here:
So indeed the A75 cores individually are faster than an individual core of the imx8mq but for loads that can utilize all cores (which video decoders are also able to do) the imx8mq is able to beat the ls1028a because it just has twice the number of (a bit slower) cores.
Finally, lets talk about the a311d bananapi. It’s my main platform right now and as vkoskiv already hinted in their response to my post above, its performance is far above that of imx8mq and ls1028a. I just tried it out and yes, you can play 1080p in firefox fullscreen without visible frame drops. If you send me a link, I can show you the CPU usage graphs so that you get an idea how taxing this is for the chip. As I said earlier, I’m mostly an mpv user and I have yet to find some material that the a311d is not able to handle (of course this means 1080p only – more doesn’t make sense with the display of the reform).
Bottom line: if you want raw performance, then there is not even a competition: a311d wins hands down over ls1028a – not only when it’s about software video decoding.
I fully agree with josch:
If you are looking for an upgrade both in CPU and GPU, go for the A311D:
or wait for the RK3588 
The later will bring the Reform to a whole new level - even when compared to the A311D - but it might still take some time (months) before this upgrade is ready.
Yeah, clearly the imx8mq does not hold a candle compared to the rk3588. But before too many people just wait for the rk3588, don’t forget, that while the design is mostly finished:
MNT doesn’t even have a physical adapter board yet (still to be manufactured) and all the bring-up work still has to be done. It would not be unexpected if weird or hard to diagnose problems show up with the adapter board (and multiple revisions are needed) or with the current state of the drivers in the kernel. A lot of work still has to be done and it’s impossible to say quite how much longer its still going to take and whether there will be some weird quirks in the final product or whether things like power consumption make the rk3588 less attractive than the a311d for certain use-cases.
The a311d is well understood, well supported and working for a lot of people today. Since (in contrast to imx8mq) one can even use javascript heavy websites without much of a fuzz I think it is also “good enough” for everyday tasks and that waiting for rk3588 is mostly something people should do if their current setup also is “good enough” for them and then they can “skip” a311d and upgrade to rk3588 directly once it becomes reality (if it becomes reality – lets hope it does).
I think I heard Lukas say that for MNT the best thing you can do, if you consider upgrading, is to spend the money today instead of waiting. In context of the Reform Next Lukas said
talking about it [Reform Next] now is probably Osborning our sales a bit (and we basically celebrate every order that’s coming in because we don’t have investors)
This is incredibly helpful. I’d read about the A311 on the shop page, and it’s crazy affordable as well. It looks like that might be the most practical choice.
I have an RPI I use as a media server, but I played with it for a while on my desktop; it’s very, very good, but not quite there, but it’s a basic 2Gb board. The BPI is much higher spec. I also own a Pinebook Pro which has decent power but I prefer the Reform in a lot of ways.
You are welcome! If you have a specific software to run or video to play and you’d like to know whether a311d can do it, I can try it out for you (if it’s not too much effort to set up). I’m having a really good time with the a311d and the only downsides to it that I noticed are:
taskset --cpu-list 2,3,4,5 so that only the four A73 cores will be used. Some background: https://bugs.linaro.org/show_bug.cgi?id=1443-j or similar option choosing parallelism is to choose more processes to run in parallel. Since the RAM is about the same as for imx8mq this means that I ran out of RAM twice so far when compiling big C++ projects. So I have to manually choose -j4 to limit memory usage.I remember back when I used to compile Firefox the linker running towards the end (I think it was the linker?) would often exhahe memory…
I don’t have any applications in mind but if I think of anything I will absolutely let you know. Thank you for the offer!