R-Pi Performance

[hippy]

From the FAQ ... "Overall real world performance is something like a 300MHz Pentium 2"

Does anyone have any performance ratio which compares with modern systems that people are more likely to have on their desks or laps to allow them to better appreciate what R-Pi performance can be expected ?

From http://www.roylongbottom.org.u.....esults.htm it seems a 300MHz P2 is about 10 times slower than a modern top-speed CPU but I don't know if that's fair as it's just one benchmark and may be an invalid or unfair comparison.

It would be useful to have some easily understandable indicator along the lines of; 'if it takes a couple of seconds on a desktop, it will take nearer half a minute on an R-Pi'.

[Joe Schmoe]

You know, the problem with all of this is that whenever I see things like "A 300 Mhz Pentium 2", with a note of derision like that's the equivalent of the Stone Age, I can't help but remember when we would have killed for that kind of computing power.  And we thought 10M of RAM was heaven on earth.  I remember the first machine that a friend of mine got that had 10M - wow!

Well, the point of all this is not to reminisce, but to note that maybe the core problem here *is* that we think we absolutely, positively, need a GUI.  Since we seem to be trying to re-create the 80s with this device, anyway, why don't we just go for it and put it forward as a command line system - you know, like the 80s micros that we think it should be?  I really think that the combination of the Windows-like GUI and web browsers is just killing us in terms of what this board is.

The command line + SVGAlib (whatever it is that gets you graphics without X in Linux) is a pretty good re-implementation of those 80s micros.

[shirro]

The thing about a device like the Pi is you arrive with a set of expectations based on your everyday experience and are initially frustrated and then as you use it you modify your habits until they match the capability of the hardware.

There is no point comparing software X running on a current model desktop and the same thing on a Pi, because you will probably use something lighter on a Pi or not use it at all. It is too slow to comfortably run many things and more than sufficient for others.

Straight cpu benchmarks don't really tell you how fast real usage is. If you have particular use in mind – add it here and I can time it on a Pi for you.

This is openssl speed for sha256 in 1000s of bytes per second give you a rough idea on integer performance. This doesn't really have much to do with real performance which is much more complicated. Different versions of openssl here so lies, damn lies and benchmarks applies. type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes1 sha256            1924.70k     4547.19k     8094.04k    10077.18k    10852.35k2 sha256            8593.84k    20022.73k    34365.70k    42853.02k    45673.13k3 sha256 14882.35k 36263.73k 67432.63k 86205.91k 93855.15k 1 is an 800Mhz Raspberry Pi (100Mhz overclock to cpu only)2 is a 1.6Ghz Atom N270 from a 2008 vintage MSI Wind3 is a 2.2Ghz Core 2 Duo from a 2009 vintage low end MacBook Pro

[The_Saint]

There was a useful* estmation of Pi performance the other day with the comparison with the Colossus

The second reply in the comments


I'm probably wrong here but if I've got this right then according to the link below Colossus had 0.000224MIPS. From what I gather from it's spec the ARM ARM1176JZF-S processor delivers 965 MIPS@772MHz. That should mean that it delivers over 4,308,035 times the processing capacity.


* Useful applied fairly loosely

[jojopi]

I actually still have a Pentium II 300MHz:sha256            1798.81k     4154.35k     7268.95k     8940.20k     9587.37k
It is now a headless server, but until late 2003 it was my main desktop.  It was never slow.  Its main deficiency at the time I replaced it was that it could not play DVD-quality video.

It always had X running.  (In fact, it used to manage an HP X terminal as well as its local display.)  But I generally ran lots of terminal windows and few true GUI apps.  I did run a web browser of course, though probably not one that would render today's pages very well.

I do not think we should have to lower our expectations too much to get useful work done on a RasPi, especially programming and exploration.  But certainly some modern GUI software is a poor fit.

[hippy]

Joe Schmoe said:


You know, the problem with all of this is that whenever I see things like "A 300 Mhz Pentium 2", with a note of derision like that's the equivalent of the Stone Age, I can't help but remember when we would have killed for that kind of computing power.


I recall that using a 20MHz 286 wasn't that bad at the time, but then I probably wasn't doing what I do today and I don't have any means to compare the two.

I'm actually coming from that direction; if a 20MHz 286, 4MB RAM, 10MB HDD was perfectly okay then an R-Pi with considerably more oomph should be perfectly acceptable now, but all these reports of it not being powerful enough or not having enough memory to run some apps and reports of it being slow are starting to worry me.

I suppose that's partly in not having one to play with and being able to judge for myself. Have I been so spoilt by high performance that I'll really notice and suffer from sluggish response ?

Perhaps we need some real world comparisons like "how long to render the http://www.bbc.co.uk homepage", how long to compile some particular application.

[nick.mccloud]

As a quick exercise I ran the SunSpider JavaScript benchmarks -

MacBook i5 2.3Ghz 8Gb memory in Safari - 203ms

Pi using std Debian - Iceweasel (FireFox) crashes out, Midori = 44000+ms

Clearly some limitations that prevent the benchmark running anywhere near sensibly on the Pi so not a test of real use to anyone.

As hippy suggests, some rather subjective but real world tests would be more appropriate.

[grumpyoldgit]

My 8 year old Dell Optiplex GX260; P4 2.4Ghz 1GB Ram; Firefox on Lubuntu 12.04 - 236.3ms

Pi overclocked to 800MHz and latest firmware; Chromium browser - 17162.9ms

[johnbeetem]

I don't know why people think GUIs require a lot of hardware performance.  After all, modern GUI expectations were originally set by the 1984 Apple Macintosh with an 8 MHz 68000 and 128KB (yes K with a K) of DRAM, with no hardware graphical acceleration.  Yes, QuickDraw had a very tight assembly language implementation and perhaps there's a lesson there.  And yes, when I first used X Windows on HP/Apollo 68030/040 workstations circa 1990 it ran rather poorly.  The excuse was that the inherent inefficiency of X Windows wouldn't matter because hardware would get faster.  Perhaps there's a second lesson there.

But I still say there's no inherent reason GUIs require more than the abundant graphical performance RasPi offers.

[hippy]

Grumpyoldgit said:


My 8 year old Dell Optiplex GX260; P4 2.4Ghz 1GB Ram; Firefox on Lubuntu 12.04 - 236.3ms

Pi overclocked to 800MHz and latest firmware; Chromium browser - 17162.9ms


Thanks for testing.

17 seconds feels like an eternity these days, not what I'd class as "a capable little PC" by today's standards. It would be interesting to see what the times are like on various iPad/iPhone and Android devices for comparison.

I don't think any performance issues are going to affect those using the R-Pi as primarily intended for the studying of computer science and programming, but it may impact on those looking at it as an application platform. I suspect my plans of using it as a TV web browser have gone out the window. Other plans using lighter software or needing less demanding processing should fare better.

[SN]

John Beetem said:


But I still say there's no inherent reason GUIs require more than the abundant graphical performance RasPi offers.


Unless you're microsoft with their very lovely but stupidly graphically overbearing Aero GUI - what was wrong with the Win95/96/NT4/2000 interface - even XP's veneer was a bit pointless - and of course the answer for them is - we have to change it to keep the money rolling in, not because its any better... (apply same logic to Office BTW)

[shirro]

John Beetem said:


I don't know why people think GUIs require a lot of hardware performance.  After all, modern GUI expectations were originally set by the 1984 Apple Macintosh with an 8 MHz 68000 and 128KB (yes K with a K) of DRAM, with no hardware graphical acceleration.  Yes, QuickDraw had a very tight assembly language implementation and perhaps there's a lesson there.


Mac bitmap display, 512x384 1bb = 24.5kB to write to update the screen

Typical Raspberry Pi 1920x1024 16bpp = 3,932kB to write to update the screen

So you need a machine 160x faster just to get to the same level of performance. A 1.2Ghz 68000?

No web browsers in 1984. These days you can find a single image in a web page that is bigger than the Mac's entire memory. Lots of these in a complicated document structure that can be manipulated and reflowed with an interpreted language attached. Suddenly you need something 100x faster again. It all adds up. If you want to run some xterms at a sensible size and a clock just like we did in the good old days, X works magnificently. Anyone who thinks it is going to handle a modern web site easily is innumerate.

[eshatohhin]

Here are my results from sysbench:

----- CPU -----
sysbench --test=cpu --cpu-max-prime=2000 run

Running the test with following options:
Number of threads: 1

Doing CPU performance benchmark

Threads started!
Done.

Maximum prime number checked in CPU test: 2000


Test execution summary:
total time: 148.7571s
total number of events: 10000
total time taken by event execution: 148.7042
per-request statistics:
min: 14.64ms
avg: 14.87ms
max: 31.24ms
approx. 95 percentile: 15.74ms

Threads fairness:
events (avg/stddev): 10000.0000/0.00
execution time (avg/stddev): 148.7042/0.00

----- Memory Write -----
sysbench --test=memory run --memory-total-size=2G

Running the test with following options:
Number of threads: 1

Doing memory operations speed test
Memory block size: 1K

Memory transfer size: 2048M

Memory operations type: write
Memory scope type: global
Threads started!
Done.

Operations performed: 2097152 (56877.18 ops/sec)

2048.00 MB transferred (55.54 MB/sec)


Test execution summary:
total time: 36.8716s
total number of events: 2097152
total time taken by event execution: 31.7923
per-request statistics:
min: 0.01ms
avg: 0.02ms
max: 12.94ms
approx. 95 percentile: 0.00ms

Threads fairness:
events (avg/stddev): 2097152.0000/0.00
execution time (avg/stddev): 31.7923/0.00

----- Memory Read -----
sysbench --test=memory run --memory-total-size=2G --memory-oper=read

Running the test with following options:
Number of threads: 1

Doing memory operations speed test
Memory block size: 1K

Memory transfer size: 2048M

Memory operations type: read
Memory scope type: global
Threads started!
Done.

Operations performed: 2097152 (63730.15 ops/sec)

2048.00 MB transferred (62.24 MB/sec)


Test execution summary:
total time: 32.9068s
total number of events: 2097152
total time taken by event execution: 27.7118
per-request statistics:
min: 0.01ms
avg: 0.01ms
max: 3.38ms
approx. 95 percentile: 0.00ms

Threads fairness:
events (avg/stddev): 2097152.0000/0.00
execution time (avg/stddev): 27.7118/0.00

[grumpyoldgit]

I installed sysbench for a comparison with a few arm_freq values.
I just used sysbench --test=cpu --cpu-max-prime=2000 run
I tried it with 700 800 and 900
At first the results were encouraging but after a while I realised that the first run after a reboot gave poor results and you needed to run the same test several times to get consistent results

700 195 239 147 146
800 128 208 127 128
900 178 149 113 113

For comparison I ran the same test with my Dell Optiplex GX260 running Lubuntu 12.04. It is 8 years old and is P4 2.4GHz with 1GB RAM.

4.9 4.7 5.9 5.9

My Optiplex is quite decrepit but certainly isn't 40 times faster than the Pi.

However, as long as you run the test several times, you can see that over clocking the Pi does get an improved score. How that relates to real life I am not sure.

[Nr90]

Grumpyoldgit said:


My 8 year old Dell Optiplex GX260; P4 2.4Ghz 1GB Ram; Firefox on Lubuntu 12.04 - 236.3ms

Pi overclocked to 800MHz and latest firmware; Chromium browser - 17162.9ms


Thanks for testing.

17 seconds feels like an eternity these days, not what I'd class as "a capable little PC" by today's standards. It would be interesting to see what the times are like on various iPad/iPhone and Android devices for comparison.

I don't think any performance issues are going to affect those using the R-Pi as primarily intended for the studying of computer science and programming, but it may impact on those looking at it as an application platform. I suspect my plans of using it as a TV web browser have gone out the window. Other plans using lighter software or needing less demanding processing should fare better.

Sunspider 0.9.1 Javascript
My ipad 3 with iOS 5.0.1 and the standard safari browser: 1810.5 ms

[eshatohhin]

to me it looks like the memory is a big bottleneck here.
62/55 MB/sec read/write from the test is real bad.
(i hope its doesn't work over USB too lol