CPU 20% slower than BeagleBoard?


22 posts
by lb » Fri Feb 10, 2012 7:09 pm
According to [1] the CPU of the Raspberry Pi is merely 20% slower than the BeagleBoard"s CPU. Isn"t that a bit overly optimistic? What is this number based on? I think it is important to give the public a realistic impression of performance, there are enough misconceptions about the Raspberry Pi going around already.

[1] http://www.linuxuser.co.uk/new.....
Posts: 193
Joined: Sat Jan 28, 2012 8:07 pm
by abishur » Fri Feb 10, 2012 7:13 pm
Do you mean what's benchmark tests did they run?  Or is it more a raw number deal 700 MHz vs whatever MHz CPU?

I guess what I'm trying to say is I'm not sure which piece of data you're considering to be overly optimistic :-P
Dear forum: Play nice ;-)
User avatar
Forum Moderator
Forum Moderator
Posts: 4265
Joined: Thu Jul 28, 2011 4:10 am
Location: USA
by lb » Fri Feb 10, 2012 7:23 pm
What benchmarks, and how they were done. The hardware is known and it's also pretty well known that Cortex-A8 is a massive improvement over ARM11 in performance. I think the benchmarks might be flaky and/or not reflect real-world performance.
Posts: 193
Joined: Sat Jan 28, 2012 8:07 pm
by stormy1 » Fri Feb 10, 2012 7:28 pm
720Mhz A8 vs 700MHz A6

20% in some benchmarks isn't unreasonable.

May be higher or lower in others.
Posts: 60
Joined: Fri Jan 06, 2012 3:44 am
by abishur » Fri Feb 10, 2012 7:43 pm
I don't know what benchmarks they used (maybe Gert, Liz, or Jamesh know?) but I do know the people making the claim.  I know that they're trustworthy people have been nothing but forthright with us as well as being extremely intelligent and tech savvy people.  As such if the r-pi team is the one that did the benchmarking (which the article seems to be implying) then you can have some modicum of assurance that these test aren't flaky ;-)
Dear forum: Play nice ;-)
User avatar
Forum Moderator
Forum Moderator
Posts: 4265
Joined: Thu Jul 28, 2011 4:10 am
Location: USA
by mkopack » Fri Feb 10, 2012 7:57 pm
And are they comparing the RPi to the Beagle BOARD or Beagle BONE? The Bone is around $80, the BOARD is something like $200 (IIRC)… So not exactly an apples to oranges comparison in price either… I would EXPECT the $200 board to be a lot faster!

(Edit, just looked up the Bone specs, yeah, they're talking about the BBone)... Still $90 vs $35, I would HOPE the Bone is faster given the higher price point.
User avatar
Posts: 242
Joined: Mon Nov 07, 2011 8:46 pm
by nullstring » Fri Feb 10, 2012 8:26 pm
mkopack said:


And are they comparing the RPi to the Beagle BOARD or Beagle BONE? The Bone is around $80, the BOARD is something like $200 (IIRC)… So not exactly an apples to oranges comparison in price either… I would EXPECT the $200 board to be a lot faster!

(Edit, just looked up the Bone specs, yeah, they're talking about the BBone)… Still $90 vs $35, I would HOPE the Bone is faster given the higher price point.



No, you're wrong. Please do not make such baseless assumptions.

Both the beagleboard and the beaglebone currently have a Cortex A8 running at 720mhz. And the beaglebone is far too new to be the one they used.

The clockspeed of the beagleboard has changed over the years. Only the new C4 revision has a 720mhz clock. See here: http://elinux.org/BeagleBoard#Clocking

My bet is that they are comparing a 600mhz Cortex A8 to their 700mhz armv11.

In this case, the RPI being 20% slower seems like it would be spot on.




EDIT: to add some context- Even though that article was released yesterday, I heard about those benchmarks when RPI was still in alpha status. The 720mhz version wasn't out then.
Posts: 178
Joined: Sun Oct 02, 2011 3:05 pm
by jamesh » Fri Feb 10, 2012 8:29 pm
I think about 20% is approx right - the Raspi is defeinitely slower, both is clock speed and because of the instruction set. It's never going to set the world alight in the processing stakes, but that's really not the point.

On the other hand, the Raspi has a 24GFlop GPU. The BBone doesn't.
Soon to be unemployed software engineer currently specialising in camera drivers and frameworks, but can put mind to most embedded tasks. Got a job in N.Cambridge or surroundings? I'm interested!
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 11545
Joined: Sat Jul 30, 2011 7:41 pm
by HansH » Fri Feb 10, 2012 9:00 pm
So.... if the cpu is idle for 80% of the time what use does that have.....

The Pi is fast enough for me... and for a very good price.
Posts: 212
Joined: Mon Sep 05, 2011 7:49 am
by Alchemy » Fri Feb 10, 2012 9:19 pm
Given the huge price difference maybe its fairer to benchmark VS bramble clusters.

Also. I really want to know bramble cluster benchmarks?
Posts: 92
Joined: Mon Jan 30, 2012 2:10 pm
by johnbeetem » Fri Feb 10, 2012 10:19 pm
Here are some other considerations for comparing BeagleBoard's Cortex-A8 Ti OMAP 3530 and RasPi's ARM11 BCM2835 performance:

The ARM Processor guide shows 2.07 DMIPS/MHz for the superscalar Cortex-A8 and 1.25 DMIPS/MHz for the ARM1176JZ-S.

The OMAP 3530 has 256 KB of L2 cache.  I think I read that RasPi's BCM2835 also has that much L2 cache, but by default it's assigned to the GPU.  L2 cache can make a pretty big difference for some applications.

The Cortex-A8 has NEON SIMD.

The OMAP 3530 has a DSP for video applications.  I don't know how it compares to the BCM2835 GPU for released binaries.  However, I believe the TI DSP is a published architecture so you can write your own DSP applications instead of being limited to released binaries.

The OMAP 3530 has POWERVR SGX 3-D graphics.  I don't know how it compares to the BCM2835 GPU for released binaries.

I suspect performance depends a lot on the application and how it is partitioned between processors.  YMMV.  OTOH, RasPi has ample processing power for its intended purpose and beats BeagleBoard hands down on the "processing power for a $25/$35 board" benchmark.
User avatar
Posts: 942
Joined: Mon Oct 17, 2011 11:18 pm
Location: The Coast
by reiuyi » Sat Feb 11, 2012 12:29 am
It's a good thing this thread specifically says "CPU".

In practice; when using both CPU and GPU for computations, the Raspberry Pi would most likely outrun the Beagleboard and it'd be an unfair comparison. GPU accelerations are still gaining popularity in programming due to nearly everyone having a huge GPU which isn't doing anything useful 99% of the time (when not gaming). I don't know anything about hardware acceleration, but apparently GPUs can harness much greater computational force compared to CPUs.
User avatar
Posts: 165
Joined: Sun Oct 09, 2011 4:59 pm
by Andre_P » Sat Feb 11, 2012 12:45 am
In a slightly belligerent mood so please excuse me.

So there is a board out there with a faster throughput core.

And the point is ?

If you want a faster processor then you have several options.

1. go buy the faster board.

2. Design a faster processor, get it into an FPGA and then use that.

3. Get a license from ARM and put it into an FPGA.

4. Get a Xilinx/Altera FPGA that already has a Cortex in there and use that.

For points 2,3 and 4 add the entire engineering flow.

I fear that there is a loss of vision as to what the aim of this entire project is all about when I see this kind of post. It's about getting kids to programme again.
Posts: 241
Joined: Sun Aug 28, 2011 7:57 am
by Jessie » Sat Feb 11, 2012 3:59 am
The other thing to note about the beaglebone aside from it's more than double price tag is that it has no video output.  That is supposed to be released later as a "shield."  So more money.

I'm not completely knocking the BB I may in-fact buy one, but I doubt it.  The Pandaboard ES appeals to me more even with its super heafty price.  All these ARM boards serve a different purpose.  The R-Pi will do many things just as good as the Beaglebone and do them at $35, and if you need that 20% extra performance well by all means drop $90 on one.  I think I could get 20% more (may be more) out of the R-Pi with some overclocking for free (won't know till I have one though.)
User avatar
Forum Moderator
Forum Moderator
Posts: 1352
Joined: Fri Nov 04, 2011 7:40 pm
by Benedict White » Sat Feb 11, 2012 12:26 pm
Jessie said:


I think I could get 20% more (may be more) out of the R-Pi with some overclocking for free (won't know till I have one though.)


Or just buy 2 Raspberry Pi's when you can?
Posts: 225
Joined: Sat Dec 24, 2011 12:24 am
by lb » Sat Feb 11, 2012 5:00 pm
If the 20% is against the first revision of the original BeagleBoard, it does indeed make sense, since it's clocked at 500 MHz by default. However the current revision of the BeagleBoard is clocked at 720 MHz, and I thought that was the base for the 20% claim. Anyway, would be nice to see this clarified. :)
Posts: 193
Joined: Sat Jan 28, 2012 8:07 pm
by johnbeetem » Sat Feb 11, 2012 9:14 pm
Jessie said:


The other thing to note about the BeagleBone aside from it's more than double price tag is that it has no video output.  That is supposed to be released later as a "shield."  So more money.


BeagleBone add-on boards are actually called "capes", a reference to that famous beagle Underdog :-)   The advantage of BeagleBone is that you can have any kind of video output you like by adding the appropriate cape, so you don't get page after page of "I want VGA" whingeing.  Or if you want to build a tablet from a BeagleBone you have a better chance of being able to interface to the LCD panel.  But it comes at a pretty steep price tag compared to RasPi, but still a whole lot cheaper than most professional development boards.
User avatar
Posts: 942
Joined: Mon Oct 17, 2011 11:18 pm
Location: The Coast
by liz » Sat Feb 11, 2012 9:29 pm
According to Eben, he used something called Nbench, and a Beagleboard "I had lying around", so I'm not sure how old it was. (Sorry; I realise this isn't very helpful, but I'm sure as soon as they're out in the wild there will be all the benchmark data you can eat.)

If you're interested in overclocking, Dom has some more to say about it in this thread.
--
Head of Comms, Raspberry Pi Foundation
User avatar
Raspberry Pi Foundation Employee & Forum Moderator
Raspberry Pi Foundation Employee & Forum Moderator
Posts: 4037
Joined: Thu Jul 28, 2011 7:22 pm
by reiuyi » Sun Feb 12, 2012 12:01 am

I have nothing more constructive to add to this thread; I just want to say that the Raspberry Pi is 20% cooler regardless of how fast the BeagleBoard CPU is

User avatar
Posts: 165
Joined: Sun Oct 09, 2011 4:59 pm
by ukscone » Sun Feb 12, 2012 12:53 am
i stumbled across fio http://freecode.com/projects/fio a few days ago and pointed someone who has an alpha/beta board to it. hopefully they will build it and run it a few times on a raspi and a few other devices and post a few results
User avatar
Forum Moderator
Forum Moderator
Posts: 2764
Joined: Fri Jul 29, 2011 2:51 pm
by Chromatix » Sun Feb 12, 2012 7:35 am
Some relevant points:

1) Many "general purpose" applications make absolutely zero use of NEON.  So the Cortex-A8 gets a boost from that *only* with specially optimised applications or libraries.

2) Both Cortex-A8 and the R-Pi have a VFP of similar design.  Don't expect any vast difference in floating-point performance between the two.  However, an R-Pi running with hardfloat ABI could probably outperform an A8 with softfp ABI under some conditions.  Cortex-A9 has a better VFP design and could beat both, but my own tests indicate that this doesn't always happen either.

3) Nbench is a general-purpose synthetic benchmark, dating back to the era when a Pentium Classic was still commonly available in the shops.  As such it is a useful indicator rather than a rigorous study.

4) Anything that does graphics will be influenced by the quality of the graphics stack including the GPU.  AFAIK, R-Pi currently uses the CPU for graphics when in X11, while BeagleBoard has at least some 2D acceleration from the TI GPU, or else gets to use NEON (via the Pixman library which does have some NEON optimisations).  However if we use the GPU properly, which the R-Pi can do from the console via OpenGL ES or OpenVG, then the R-Pi will probably have a distinct advantage.
The key to knowledge is not to rely on people to teach you it.
User avatar
Posts: 430
Joined: Mon Jan 02, 2012 7:00 pm
Location: Helsinki
by lb » Sun Feb 12, 2012 2:22 pm
Chromatix said:


Some relevant points:

1) Many "general purpose" applications make absolutely zero use of NEON.  So the Cortex-A8 gets a boost from that *only* with specially optimised applications or libraries.

2) Both Cortex-A8 and the R-Pi have a VFP of similar design.  Don"t expect any vast difference in floating-point performance between the two.  However, an R-Pi running with hardfloat ABI could probably outperform an A8 with softfp ABI under some conditions.  Cortex-A9 has a better VFP design and could beat both, but my own tests indicate that this doesn"t always happen either.

4) Anything that does graphics will be influenced by the quality of the graphics stack including the GPU.  AFAIK, R-Pi currently uses the CPU for graphics when in X11, while BeagleBoard has at least some 2D acceleration from the TI GPU, or else gets to use NEON (via the Pixman library which does have some NEON optimisations).  However if we use the GPU properly, which the R-Pi can do from the console via OpenGL ES or OpenVG, then the R-Pi will probably have a distinct advantage.


1) The same is true for SSE/AVX on x86, yet it"s still a very powerful tool. And then there"s autovectorization, which can be quite useful and recently started to work better in gcc. Plus, note that NEON is also useful for scalar code.

2) Well, ARM always sucked at FP, and still sucks at it, compared to other architectures. However, I wager Cortex-A8 is going to outperform ARM11 at FP due to the various improvements in VFPv3, RunFast and of course NEON, if used or enabled.

4) That remains to be seen. The SGX540 is no slouch either, and well supported by software. Both GPUs are perfectly capable of typical 2D graphics acceleration. But without a capable X driver, the power of the VideoCore IV will be largely unused.
Posts: 193
Joined: Sat Jan 28, 2012 8:07 pm