Raspberry Pi 2 vs C1 Benchmarks!

[schnip]

Hey Guys,



I'm really interested in the raw performance of the pi 2 vs the c1. I don't own any of these machines but I have searched for some benchmarks comparing them. I'm not really satisfied with my findings but maybe a forum member has some better results.

For a more general comparison take a look at this great article: http://www.cnx-software.com/2015/02/02/ ... omparison/



Both chips feature the ARMv7 instruction set. The Rpi2 however is a Cortex A7 whereas the C1 uses the more lightweight Cortex A5.


Wikipedia Specs for A7 (Rpi2)

Code: Select all

It has two target applications; firstly as a smaller, simpler, and more power-efficient successor to the Cortex-A8. The other use is in the big.LITTLE architecture, combining one or more A7 cores with one or more Cortex-A15 cores into a heterogeneous system.[1] To do this it is fully feature-compatible with the A15.


Key features of the Cortex-A7 core are:

Partial dual-issue, in-order microarchitecture with an 8 stage pipeline[2]
NEON SIMD instruction set extension
VFPv4 Floating Point Unit
Thumb-2 instruction set encoding
Jazelle RCT
Hardware virtualization
Large Page Address Extensions (LPAE)
Integrated level 2 Cache (0-1 MB)
1.9 DMIPS / MHz[2]

A5 (C1)

Code: Select all

Overview[edit]
It is intended to replace the ARM9 and ARM11 cores for use in low-end devices.[1] Compared to those older cores, the Cortex-A5 offers the advanced features of the ARM v7 architecture over the v4/v5 (ARM9) and v6 (ARM11) architectures e.g. VFPv4 and NEON advanced SIMD. It also allows devices to run current software, which is increasingly focusing on ARM v7 and dropping support for earlier architectures.

Key features of the Cortex-A5 core are:

Single-issue, in-order microarchitecture with an 8 stage pipeline[1]
NEON SIMD instruction set extension (optional)
VFPv4 floating-point unit (optional)
Thumb-2 instruction set encoding
Jazelle RCT
1.57 DMIPS / MHz

As you can see in the reference specification the A7 is inherently more powerful at 1.9 DMIPS / MHz whereas the A5 rates at 1.57 DMIPS / MHz only.



BUT the C1 clocks significantly higher than the Rpi2 (900mhz vs 1.5ghz)


I have found some Benchmarks for the Rpi2 here (Beware it's in german http://www.golem.de/news/raspberry-pi-2 ... 107-3.html)






And for the C1 benchmarks are provided by hardkernel. (http://www.hardkernel.com/main/products ... 1578608433)






Now when comparing these Benchmarks you can see that the Rpi2 Benchmarks are incomplete unfortunately. Also one can see that the Rpi 1 scores higher in the first picture compared to the Rpi 1 in the second picture. This seems weird to me. Does that mean that the Rpi 2 would score higher with the methodology of the second picture?

Supposed both of these Benchmarks are credible and comparable, the C1 is faster in every disclipline but not by a huge margin.. Also the Rpi2 might score higher with the methodology of the second benchmark. In that case the difference in performance would be negligible.

Also it would be very nice if someone could release Benchmarks for both machines with the exact same methodology.


Thanks in advance!
schnip

[memeka]

The C1 is faster when comparing the CPU because of the higher clock. I would be more interested in a power usage comparison (at stock clock levels).

[schnip]

The C1 is faster when comparing the CPU because of the higher clock. I would be more interested in a power usage comparison (at stock clock levels).

Well that sounds good. But isn't it possible that certain applications profit more from the feature set/architectural advantage of the Cortex A7 instead of higher clockspeeds?


Also the Rpi 2 according to the Pi foundation has some overclocking headroom. I don't know about the overclockability of the C1 though.

[mdrjr]

_I BELIEVE_ that C1 should be better on single threaded applications. (Most of them are)

As for Overclock current C1 kernel allows up to 1.72Ghz.

[memeka]

The C1 is faster when comparing the CPU because of the higher clock. I would be more interested in a power usage comparison (at stock clock levels).

Well that sounds good. But isn't it possible that certain applications profit more from the feature set/architectural advantage of the Cortex A7 instead of higher clockspeeds?


Also the Rpi 2 according to the Pi foundation has some overclocking headroom. I don't know about the overclockability of the C1 though.

Don't know the sizes of the L2 cache and the impact, but LPAE is useless (it's for >4gb ram), hw virtualization is useless (i don't think the target or these boards is running virtualization). The rest of the features are (almost) on par.
C1 is stable at 1.7GHz and mlinuxguy overclocked it to 1.9Ghz with active cooling. rpi2 can go I think up to 1.3Ghz, which would give similar results to the stock C1. So the edge will be similar when overclocked...

[crashoverride]

hw virtualization is useless (i don't think the target or these boards is running virtualization).

Even if hardware virtualization were an option on RPI2, it would be worthless. The VideoCore4 does not have an IOMMU on it like the ARM Mali on OdroidC1 does. This means you cant isolate virtual machines.

[Edit]
In short, that means you are no better of than running 'jails'/containers which would run faster since they dont have the overhead of a hypervisor.

[mcobit]

Does the C1 have a NEON unit?
The Raspberry 2 has it.

[mikronauts]

C1 has NEON according to the data sheet.

Does the C1 have a NEON unit?
The Raspberry 2 has it.

[crashoverride]

Does the C1 have a NEON unit?

It has 4 of them.

[Matt]

The biggest thing to consider when choosing between the two is that the RPi2-B is the same thing as the RPi-B+, except that there is more memory and a much better CPU. That means that you are still limited by the fact that most peripherals run through the USB2 controller. The RPi2 still gives you a camera connector, so it would be good for anyone who wants to use one of the familiar RPi camera modules, rather than a USB-attached webcam.

The C1's biggest advantages are the option of using an eMMC module and the Gigabit Ethernet port (which is connected via a dedicated GbE bus, not through the USB2 controller, if I'm not mistaken). Using an eMMC module and the GbE port gives you a significant advantage if you are running an IoT application that needs a fast network connection and a small -- but fast -- bit of storage. Otherwise, I think the two boards are fairly comparable at their price points.

So far, it appears that the RPi2-B is going to cost $10 more than the C1 for the board itself (Adafruit is selling the RPi2-B for ~$45).

[crashoverride]

Now that there is actual competition in the $35 single board computer market, consumers can afford to be picky about other things like how a company treats its customers:
http://www.raspberrypi.org/forums/viewtopic.php?f=63&t=98693

[meveric]

I like this one:

There's not a lot more to say, apart from perhaps giving some background in to the previous statements about when the Pi2 might be coming out.

Two years ago, 2017 would have been the best date for a new Pi, but the technology world changes quickly. Things like the ODROID-C1 mean that for the same price there was much more competent device, and something needed to be done. Broadcom had been working on the 2836 and even though the team was small, this work came to fruition much faster than expected (the chip worked first time!) , so there was an opportunity to catch up with the other SBC's out there, earlier than expected. Also worth noting that although the Foundation is a charity, they are still a business, and need to keep going as a business. With the education team running full tilt, and some expensive engineers to pay for, they need a steady income stream, and the P2 is how that revenue stream is maintained.

So the reason to bring this out was to compete with HardKernel.. they had no plans to release this board until HardKernel came up with the C1.
Well it was speculated that the RPi Foundation told Broadcom to discontinue selling their chips to HardKernel and therefore the ODROID-W is discontinued.. Guess they are not so friendly when it comes to business afterall

[tank]

so the new broadcom chip and the layout of the rpi2 was already done a long time ago just to be prepared when something like the c1 is coming out?
i mean the odroid c1 was announced in decembre and i would have said it needs a bit longer to design, produce, test,... a product like the rpi2?

[crashoverride]

the odroid c1 was announced in decembre and i would have said it needs a bit longer to design, produce, test,

It wasn't until 6 months ago that they actually ironed out one of the major hardware bugs of the Raspberry Pi: until B+, inserting a USB device would reset the board.

Every ARM board out there has its "dirty little secrets". These are developer boards, not retail products so they under go much less testing. Its a world of "release early, release often" now.

[crashoverride]

Every ARM board out there has its "dirty little secrets"

http://www.businessinsider.com/raspberry-pi-2-camera-flash-glitch-turns-off-2015-2

One of the chips on the exposed computer's circuit board seems to be sensitive to light and poorly shielded, so when the camera's flash triggers, it causes the entire device to crash.

[robroy]

One of the chips on the exposed computer's circuit board seems to be sensitive to light and poorly shielded, so when the camera's flash triggers, it causes the entire device to crash.

That is hilarious!

[cutterjohn]

One of the chips on the exposed computer's circuit board seems to be sensitive to light and poorly shielded, so when the camera's flash triggers, it causes the entire device to crash.

That is hilarious!

Apparently just Xenon flash, although it should be more of a question of frequency.

Anyways, I ordered one of the rpi2-b, unfortunately it won't be here for at least 2w and that's IF I make it into newark's 1st batch AND it's on time. rpi's been flakey about being on time in the past, so I'm not holding my breath to have it before mid-March, and I have zero idea how many have been ordered and where I'm at in the queue anyways... may not be until May before it gets.

Newark only listed when they expected shipments and how many would be in each shipment and that was it, so it was useless info. AFAICS they only provide a general order # and they do sell a crapload of other stuff given that they really are primarily an industrial supplier, and just kinda sorta tacked on the rpi and then a few others for onesies/twosies v. 100s or 1000s and POs. As a matter of fact for my first several orders from them they created a "fake" PO apparently just to get their billing/shipping system to work, but my latest order is lacking this "feature". }:D

[corb555]

benchmark by oversun here:

http://habrastorage.org/files/7e9/6c7/3 ... 3cdd41.jpg

C1 beats RPi2 by quite a bit on most items. Crushes it on network throughput. RPi2 faster on execl and process creation which seems odd when C1 is faster on almost everything else.

[mikronauts]

Thanks corb555.

That agrees with my findings http://www.mikronauts.com/raspberry-pi/ ... -review/6/

Regardomg execl and process creation, I wonder if the RPi2B kernel may have some custom tuning?