Really undecided about the RK3588, any advice?

So, I use my Pocket Reform quite a lot, usually at University for group projects (were I can’t do everything alone at my desktop at home) and just a general purpose mobile computer. And I love so many things about it (form factor, look, how it feels)

Currently I’m running the i.MX8M+ that came with the machine and it’s … not great. It does most of the low performance jobs like reading PDFs and checking email just fine but for example when I need to compile an R package I can forget it, it takes hours sometimes and that can sometimes block me from working but I can plan around it and try to install everything I need the night before at home which is cumbersome.

Battery life and heat is a problem, along with the WiFi issues that are well known at this point. When I bring a power bank, battery life is no longer a problem so that’s fine. But the rest remains.

The A311D is not really an option because I do need the SD slot and the multi-display capability.

I think an RK3588 upgrade would solve most of my remaining issues but it’s quite an investment and I’m really not sure it it’s worth it.

I would love to hear experiences from people who have upgraded.

1 Like

Worth it or not would boiled down how long pocket reform gonna stay with you and how often do you use your pocket reform, if pocket reform will be your only on-the-go device then the upgrade is definitely worth it.

From your description I would said just go for 16GB ram version of rk3588 is plenty enough.(if you don’t need extra ram to compile stuff)

The rk3588 upgrade kit also come with independent Wi-Fi card(AsiaRF M27612-BU3) since rk3588 module itself doesn’t provide Wi-Fi functionality.(people also report that AsiaRF M27612-BU3 worked really well)

2 Likes

Hey, I can absolutely give my experience here. I’ll try to touch on my use cases and who I am and what I want it to be so that you can take it with the appropriate pinch of salt.

So I loved the device already, warts and all, when I had the imx8mp, but that’s because a) I got it with a view to hacking on it, personalizing it and hopefully contributing improvements and fixes, b) place a lot of my identity into the above and work deeply with Linux as an OS professionally and want to learn more about the electronics and hardware aspect as a hobbyist c) I just really love the form factor of the device itself. I’ve had it in pieces several times to do improvements and fixes, all of which was the expected (and welcomed) cost of entry for me, so there’s a degree of wanting this involvement with the device that absolutely colours my perspective.

Now I have the RK3588 I’m really genuinely in love with it and I can’t really put it down. It’s much snappier for all of the day-to-day tasks you’re talking about, but with the performance increase you should find it’s able to handle more stressy workloads without grinding to a halt for hours. That’s a very subjective thing to say though - I’m wondering if there’s an example of an R package that was problematic for you that I could have a go at compiling and timing on my pocket? Let me know if you want to try this out.

WiFi is hugely improved with the AsiaRF. Again, very subjective but before I had to use my external Alfa to get any reception on the end of the desk where my pocket sits, now I don’t bother. Perhaps worth ordering this separately first to see if it gets you more of the way toward something you’re happy with? You’d need it regardless for the RK3588.

Subjectively I think it runs slightly cooler (other people have alluded to this) when idling, and when I was compiling OpenOCD the other day the internal sensors reached ~81 degrees and the MNT logo was not unbearably hot. I still think this could be improved by increasing the surface area of the exposed copper on the backplate.

Another option here is to look at the RK3588 compared to other processors that you’ve felt were acceptably fast on benchmarking sites (if that’s a reference you have). Rockchip RK3588 vs nxp IMX8MP [cpubenchmark.net] by PassMark Software shows a ~5x performance increase over the imx8mp (which is a very coarse measure but at least it’s illustrative) so if you have another computer that performs sufficiently it’s possibly worth doing the same comparison.

Anyway, I hope some of that helps you to find your way forward to a decision.

5 Likes

A typical R package that doesn’t come pre-compiled in Debian is MBESS, maybe give that a try. It pulls in quite a lot of dependencies too and has binary components that need to be compiled.

So:

  1. install R with sudo apt install r-base r-base-dev
  2. run R and in there install.packages("MBESS")
1 Like

Just been doing a trial run (I’m pretty far from the router so I didn’t want the download times to disproportionately affect the amount of time it took) and it’s running just fine but I’ve noticed that it’s actually only building on one core by default which is going to take significantly longer and not reflect the performance gains of the RK3588 fully. Have you been doing anything to ensure multi-core builds (like export MAKE="make -j8"?). Just want to make sure that the result is going to be a helpful comparison for you.

1 Like

No custom options, just the standard install.

I think R doesn’t even support multi-threaded builds, at least not from inside RStudio

Not sure whether this is valid, but I just looked up multithreaded compile/install with R, and it may be supported.

Ah, thanks. I’ll give that a try on the i.MX8M+ then to see what the baseline is.

It think doing this whole R thing is not a good benchmark so I just compiled a simple C library: ncurses 6.1 (ftp://ftp.gnu.org/gnu/ncurses/ncurses-6.1.tar.gz)

After ./configure, on the i.MX8M+ a run of time make -j 4 reports:

real    1m59.490s
user    6m20.046s
sys     0m46.221s
1 Like

I did similarly on my i.MX8M+ Pocket Reform for ncurses 6.3 (https://invisible-island.net/datafiles/release/ncurses.tar.gz), and got the following similar result.

real 2m2.717s
user 6m20.290s
sys 0m44.916s

I did the same on my RK1 32GB RAM from TuringPi running ubuntu and have got the following result.

real 0m39.143s
user 1m37.720s
sys 0m25.267s

When I ran time make -j 8 on the same RK1, the result was:

real 0m31.130s
user 2m12.493s
sys 0m26.310s
2 Likes

-j8:

real    0m24.168s
user    2m10.685s
sys     0m16.718s

and for completeness
-j4:

real    0m30.443s
user    1m34.469s
sys     0m15.511s

:fire:

Was interesting to see in htop that -j4 primarily got scheduled on the performance cores but not exclusively and didn’t max them out.

I think that was a good choice of workload to test :slight_smile:

1 Like

Thank you, that’s very helpful.

1 Like

I’d say it’s worth it if it’s your daily driver. Got my chip installed a few days ago and it’s been nice to be able to not run out of ram. Haven’t done anything too compute intensive, but even things like web browsing are much more responsive. Overall very pleased and would recommend.

4 Likes

I ended up ordering one :slight_smile:

8 Likes

cool. I will be interested in how you find it

I am getting tempted by one for my reform.

Maybe we can start collecting some very basic benchmarks about rk3588 against other SoMs concerning the battery runtime? People keep (rightfully) asking about battery life in comparison to other SoMs. I am aware that benchmarking battery life is really tricky given all the things that eat power but I believe that having something is better than having nothing.

When Jeff Geerling reviewed the MNT Reform, this Youtube video was played in firefox in fullscreen with full display brightness while on internet over wifi.

My A311D classic Reform managed 4:45 hours.

@selfawaresoup maybe you can do a similar benchmark for your current imx8mplus and later for the rk3588 once you have it? That might help other users a bit. :slight_smile:

And, for everybody testing this, don’t forget to disable upower or otherwise your device will soft-shutdown when it thinks the batteries are empty (as it might’ve happened in Jeff’s video) and not hard-shutoff when the cells are really depleted according to the charging circuit.

2 Likes

I’ve already got the upgrade, but still waiting for the Pocket, so thinking I should really try it without first, then with, but impatience may get the better of me…

1 Like

Probably need to set screen brightness to a standard value too (50% seems fair). Also probably best to state your sysctl firmware version and whether you’ve got any hardware mods.