Are There Any Raspberry Pi 2 Benchmarks You'd Like To See?

[darkbasic]

VC4 vs blob and soft float vs hard float.

[bcdonadio]

I was hitting a 4MB/s wall when writing on USB HDDs with ext4 on the RPi 1. I would like to see how much better is this figure with RPi2.

[Dick Palmer]

...and the performance of I/O access on the SD card.

I'd be interested in that too: Specifically a FS comparison to see how speed and CPU burden vary between FSs when accessing a nice fast SD medium on this relatively CPU limited platform... F2FS, XFS, FAT32, EXT4, EXT2 as a minimum. XFS and EXT4 with their journalling disabled would be interesting additions to that.

The question also touches on one of my Phoronix pet peeves... so, since it's been asked, here's a wee rant on the subject!..

As a general request for ALL benchmarks, could you please choose a representative selection of CPUs from through the ages and stick with them: forever... as "benchmarks" for all these benchmarks? Chips which were hugely popular in their day and which therefore linger on in ready availability and in the collective conciousness. A *static* frame of reference such as that would provide would be very, very useful... especially so as the benchmarking protocol itself as well as the bulk of the benchmark packages themselves seem to exist in an eternal state of upgrade flux.

Perhaps a good, useful, widely familiar list might look something like: A PIII (~800MHz coppermine perhaps?), Athlon XP (1600ish?), Pentium M (~1.6gHz?), Core2 (duo penryn?), Athlon64 X2 (4000+?), i5 (2500k?)

The (often huge) selection of contemporary chips is obviously academically interesting in the context of providing a snapshot of that instantaneous state of the market/industry - useful for those about to make a purchase, etc. but don't offer much of a *reference* since probably rather few of us maintain farms of diverse up-to-date chips. There doesn't strike me as much point comparing an oddity like the Pi with current mainstream CPUs. The only contemporary chips that seem even remotely interesting for a comparison might be a couple of generations of Atom, some Mediatek-Banana-Orangey Pi-alike and maybe a cheapo tablet. FireHDish thing perhaps?

/rant

[Ardje]

Why no XU4 benchmarking?

[Grawp]

I would rather say they just have utterly crappy power management. Even cheap chinese tablets CPUs can at least downclock like an order of magnitude, reducing voltage as well. This makes them cold and low-power during system being idle. But Pi uses some odd broadcom crap. It initially targeted DVD players or so, and so it does not cares about power management even a bit. Needless to say it is "owerpowered" in terms of consumed power.

But in no way it is overpowered in terms of, say, network. IIRC there is sill no native Ethernet, and usb-based Ethernet is, obviously, not great. Shame on Broadcom, $5 allwinner ICs are doing native gigabit these days.

Then it's not like Pi is good when it comes to storage. SD card and USB is all you get. So you would have hard time to complain it is too fast :P.

...so Pi would be, say, very crappy microserver. One can get MUCH better device for comparable price. It is also bad choice for installations where power consumption matters or where you want to have proper back-up power supply. Other boards allow to hook LiIon battery to their power manager IC output and it would handle charging, etc and automatically switch to battery, just like tablets and phones do. But for Pi it is still kinda troublesome, they can't get adequate power management. So their real achievement is marketing. But their engineering is rather varies from mediocre to crappy and easily loses to cheap chinese boards these days.


Still, quite a power hungry device and it can be like an order of magnitude better, especially if system is idle most of time. Everything related to power circuitry and power management is clearly not what RPi creators can consider as achievement. Actually most of comparable boards would have much better power management. Isn't it a shame cheap Chinese copy-pasters pwn these made-in-uk nuts and heavy-weight broadcom monster in terms of engineering? Welcome to XXI century! XD



Uhm, ESP8266 requires to deal with custom tool chain, unusual OS and other weird crap. Not to mention it is not compatible with anything else and you hit 100% vendor lock. Once you deploy some projects... what if Espressif would decide to, say, increase prices? Right: you already put your bet on it and can't easily migrate to another device. This is much easier with devices similar to Pi, which ensures there going to be decent competition, power and features would improve, while prices would tend to go down, etc. Not to mention Linux gives a shitload of options. I'm even scared to imagine how one supposed to implement web interface on ESPs. And on Linux ARM I can basically install usual httpd of my choice and do what I need. Obviously it is better to go for more or less lightweight one, like lighttpd, nginx, or even something like thttpd or openwrt's httpd, rather than slow and resource-hogging Apache. Not to mention it is kinda wrong idea to trust wireless security to some proprietary wireless stack. And wireless as method of communication is a really fragile thing, so it is unacceptable in places where inability to communicate with device can cause troubles. Generally speaking, it is really easy to send specific packets, bringing all wi-fi activity within transmitter's range to almost complete halt.

This is all very true, RPi is a piece of crap but it's the only board in the price category that will get OpenGL (not ES) support && it won't be fixed to a particular kernel version. Or is it not? I know about Qualcomm's Adreno drivers however the cheapest board with Adreno GPU is Dragonboard which is about 80$, isn't it?

[brent]

VC4 vs blob

Seconded.

and soft float vs hard float.

Everyone uses hard float. And the result would be pretty obvious. So this is hardly interesting.

[SystemCrasher]

... but it's the only board in the price category that will get OpenGL (not ES) support && it won't be fixed to a particular kernel version. Or is it not?

I would agree here, if one REALLY needs it, to degree it outweighs everything else. Though I do not really get what final device supposed to look like. Some HTPC-like thing? it still got relatively weak CPU and crappy IO for this application. Most control and automation devices can live without 3D, etc. Microserver-like networked things, etc just do not need GPU at all.

In no way it is balanced like PC. It is very niche thing which is mostly GPU. And everything else is kinda secondary. First version had utterly crappy CPU. Chinese boards started to offer like 2-4 of far better CPU cores, so in version 2 Pi they HAD to improve CPU or they could lose their market share. Now they've got better CPU, sure, but... AFAIK there is still same crappy IO, still same crappy power management, same weird boot sequence where you MUST have blobs and FAT32, and so on. For me it matters in many cases.

I know about Qualcomm's Adreno drivers however the cheapest board with Adreno GPU is Dragonboard which is about 80$, isn't it?

TBH most of tasks where I use similar devices are not demanding in terms of OpenGL graphics, etc.

...and btw, IIRC, mainline kernel support of these broadcom SoC's isn't great. When I've checked it for last time, several months ago, it has been almost completely missing in mainline. And when it comes to using custom out of tree kernel, IMHO it is not a big deal if it fixed version or not, it is a major trouble on the way. Maybe it somehow improved, etc, I've not checked it recently. And it's not like if I want some Crapbian instead of real Debian and custom out of tree crap kernel, etc.

For me, safest option is proven to be to go for mainline u-boot loader, recent mainline kernel and then usual "generic" Debian, debootstrapped myself from armhf debian packages. This way I can be sure how it performs, it performs in expected ways, it lacks troublesome "goodies" and "improvements" I do not need or want, etc. I like u-boot way of doing things, mainline kernel way of doing things, debian way of doing things, etc. Where you can expect to get software which is only does what advertised, does not attempts to pwn or vendorlock you, and there're clear licensing terms, so it is okay to go for commercial applications as long as you're okay with fairly simple terms (mostly coming to availability of sources). And I have idea who is in charge of various components and related bugs, etc. Which is a good thing for sure.

[Michael]

Why no XU4 benchmarking?

Because I don't have one? Nor really extra funds to buy every board out there...

[riklaunim]

If someone wants some SD cards benchmarks: http://www.rkblog.rk.edu.pl/w/p/test...-raspberry-pi/

Other RPi v1 benchmarks:

http://openbenchmarking.org/result/1407062-BY-1407013BY79

http://openbenchmarking.org/result/1406224-SO-1406227RA12

[andreano]

For benchmarks, you can compare Raspbian vs Android, X vs Wayland, and of course RPi vs cheap x86.

But more than benchmark numbers, I want to know what remains to be fixed, in terms of drivers, before RPi is ready for the Linux desktop. What still sucks?

When RPi was new, in 2012, there was this OpenGL benchmark that rendered to a framebuffer overlay on top of the non-accelerated desktop. And that was it, nothing else was accelerated. How useful.

Reasons I gave up RPi for desktop:
1. Firefox scroll latency
2. Fullscreen video in Firefox
3. Composited desktop? (didn't even try)