TR Forums

OK, in response to a recent Phoronix article that purports to show how amazing the Panda Board is compared to a 2.5 year old single-core Atom (see TR thread here: http://techreport.com/news/23512/tuesday-shortbread) I have decided to fight fanboism with fact. I went out and installed the Phoronix Test Suite on my Arch Linux notebook and started doing some benchmarks of my own. I have preliminary results posted here and will update the thread as I get the chance to run more tests.

In one corner: The numeric results published by Phoronix from the article in question for the Panda Board (released late 2011 and still representing a very high-end ARM development board) using a dual-core Cortex A9 clocked at 1GHz a piece.
In the other corner: My four year old notebook (purchased August, 2008) with a Core2 Duo P8400 (dual core with 2.26GHz max clock but see explanation below below).

BUT WAIT YOU SAY THAT'S NOT FAIR TO ARM!!

Oh but it is.. you see while my notebook theoretically has two cores and a higher clock speed, I disabled one core and forced the CPU to a lower clock speed using the following commands:

1. Disable the second core:
sudo sh -c "echo 0> /sys/devices/system/cpu/cpu1/online"

2. Force the one remaining CPU core into permanent power-saving mode stuck at 800 MHz with no ability to boost back to full-speed:

sudo sh -c "echo powersave /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor"

Output of /proc/cpuinfo taken during the benchmark with single CPU pegged at 100%:
processor : 0
vendor_id : GenuineIntel
cpu family : 6
model : 23
model name : Intel(R) Core(TM)2 Duo CPU P8400 @ 2.26GHz
stepping : 6
microcode : 0x60c
cpu MHz : 800.000 <-- This is the actual Mhz of the single core taken during the benchmark
cache size : 3072 KB
physical id : 0
siblings : 1
core id : 0
cpu cores : 1
apicid : 0
initial apicid : 0
fpu : yes
fpu_exception : yes
cpuid level : 10
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx lm constant_tsc up arch_perfmon pebs bts rep_good nopl aperfmperf pni dtes64 monitor ds_cpl vmx smx est tm2 ssse3 cx16 xtpr pdcm sse4_1 lahf_lm dtherm tpr_shadow vnmi flexpriority
bogomips : 4524.72
clflush size : 64
cache_alignment : 64
address sizes : 36 bits physical, 48 bits virtual
power management:


Here is the first result from C-Ray (a multi-threaded benchmark that has favored the Panda Board), more to follow:

1. Phoronix published results: 875.76 seconds (14 minutes 35.76 seconds) to completion (less is better).
2. Single-core Core 2 at 800 MHz: 746.59 seconds (12 minutes 26.59 seconds) to completion (3 run average, 743.588 minimum, 751.047 maximum)
3. Winner: One core of the Core 2 at 800 Mhz Margin of win: 2 minutes 9.17 seconds (Core 2 is 14.7% faster compared to A9 posted times).


Before I hear how using a 4 year old single underclocked x86 core in multithreaded benchmarks against a higher-clocked multi-core ARM system is somehow cheating in Intel's favor:
1. This test was run from a full-blown KDE 4.9 desktop running on the notebook during the benchmarks because I was too lazy to kill the X-server and drop to the command prompt.
2. Unlike the Panda Board I slapped the Phoronix suite on a boring standardized Arch Linux system and was running the first benchmark in about 15 minutes. No running out trying to optimize every single library with some magical "hardfp" branch and trying to perform brain surgery with exotic GCC options for every test.
3. P.S. --> While I was running these benchmarks, I was also SSHing in to the laptop to collect data and I even took a screenshot (below). Try running KDE 4.9 on the A9 board and taking screenshots DURING THE BENCHMARK RUN and see what kind of timing hit you take.

Edit: Here is a link to a whole album of screenshots I'll be taking. The yakuake terminal is taken in the middle of the second c-ray run, the other two are screenshots of the official Phoronix test suite benchmark results in the web browser on the notebook:
https://picasaweb.google.com/1026166348 ... directlink

Here are the next results for the FLAC-encode benchmark:


1. Phoronix published results for dual-core A9 at 1GHz: 50.84 seconds.
2. Single-core Core 2 at 800 Mhz: Average: 40.29 seconds.
3. Winner: Core 2 (20.75% faster than the A9)

See the image link page above for result images.

EDIT: P.S. the high standard of deviation is mostly due to me running Firefox to open up TR's page.. during the bechmark run on a single-core setup.

EDIT 2: Officially uploaded to Phoronix test suite: http://openbenchmarking.org/result/1209 ... LAACCRIP86

Ok, I haven't been following the news on CPU front for a long time, but isn't Atom slower than Core clock to clock?

Even underclocked and limited to a single core, using a Core 2 as a stand-in for an Atom isn't valid. Atom is a crippled architecture (no instruction reordering, speculative execution, or register renaming), with a fraction of the IPC (somewhere between 1/3 and 1/2 I believe) of Core 2.

Should probably change the thread title to "Pseudo-Science Beats Gouda" or somesuch...

just brew it! wrote:

Even underclocked and limited to a single core, using a Core 2 as a stand-in for an Atom isn't valid. Atom is a crippled architecture (no instruction reordering, speculative execution, or register renaming), with a fraction of the IPC (somewhere between 1/3 and 1/2 I believe) of Core 2.

Should probably change the thread title to "Pseudo-Science Beats Gouda" or somesuch...

Oh you aren't getting the point of this thread (and there are more benchmarks to come). The point isn't to say that my Core 2 should be slower than an Atom.. it most certainly should be substantially faster than a single core Atom at the very minimum. Instead, this post is to show the "ARM is about to takeover the Desktop" crowd how a massively crippled 4 year old Core 2 notebook performs compared to high-end dual core ARM systems from the 2011-2012 timeframe.

I doubt anyone is thinking that ARM will overtake the desktop in a few months. But if you ask me, they are closing the gap pretty quickly, and with all the R&D dollars going into ARM these days because of the gadgety obsessions, they soon might be just good enough.

MORE RESULTS TO COME TOMORROW!
Don't worry guys, I think I'll find something that the A9 can win at. I'm betting that the Graphics-Magick tests, where the A9 was about twice as fast as the Atom, could give ARM a win or two. I can pretty well guarantee that OpenSSL will be a bloodbath since Atom already wins there. One other thing: The N-Queens benchmark appears to be a downloaded binary that was not re-compiled on my system. That means there is good chance it is not well optimized for this particular box and could give interesting results.... to be continued.

Madman wrote:

I doubt anyone is thinking that ARM will overtake the desktop in a few months. But if you ask me, they are closing the gap pretty quickly, and with all the R&D dollars going into ARM these days because of the gadgety obsessions, they soon might be just good enough.

The problem isn't ARM overtaking the desktop- it's that the desktop is a stationary target. Quad A15's should be able to run WinRT and whatever Office MS puts on it with the same snappiness as a standard business desktop, and if you add in the screen size and input options, ought to outright replace it.

The very integrated nature of the ARM platform is likely more amenable to the office environment and makes ITX look enormous. No wonder Intel went out of their way to design a full-scale x86 form factor that's roughly half the size. People are going to stop looking at x86-based computers as a necessity for every day computing and relegate them to performing tasks which amount to actual work, not just plodding on a browser or application.

And don't forget us gamers .

chuckula wrote:

Oh you aren't getting the point of this thread (and there are more benchmarks to come). The point isn't to say that my Core 2 should be slower than an Atom.. it most certainly should be substantially faster than a single core Atom at the very minimum. Instead, this post is to show the "ARM is about to takeover the Desktop" crowd how a massively crippled 4 year old Core 2 notebook performs compared to high-end dual core ARM systems from the 2011-2012 timeframe.

Is anybody actually saying that ARM is "about to take over the desktop"? In 5 years I could see it making substantial inroads... maybe. But not anytime soon. The Atom was barely competent as a netbook CPU, let alone a desktop; so the Phoronix article is not implying that ARM can replace a traditional desktop CPU.

Sure, for people who only check e-mail or browse the web it could make sense. But that isn't "the desktop". That's one slice (the low end) of the desktop segment.

Seems to me you're fighting a non-existent enemy.

I'm not so sure "**fill in arch here** is taking over the desktop" is literally based around the arch. What people are talking about is the fact that it almost doesn't make any difference what you build for your mother, father, sister or brother. Whatever is inside the case it will be fast enough for their needs. If you paired an Atom, Zacate, Pentium G or even a Sempron with an SSD you would be hard pressed to tell the difference in everyday workloads.

Because of this other archs are becoming viable alternatives because they generally can deliver the same experience. Running Atom, ARM or Core2Duo through Cray isn't going to prove much. Everyday users don't use those type of workloads. The average person doesn't even transcode video. All ARM, Atom, et al need to do is generally provide decent performance when end users check their email, or surf the web. That's it.

Yeah, it really does depend on how you define "desktop". There's a broad spectrum there, from web terminal to power-user workstation.

Getting back to the OP: I still don't see how Phoronix is "full of it". They compared the performance of an ARM CPU to an Atom. So...?

OK: I've gotten fancier with the benchmarking and I ran the entire set of tests that Phoronix used in its article. The full result set is online here: http://openbenchmarking.org/result/1209 ... 208261BY38

Basically, the Core 2 running one core at 800 Mhz won all the benchmarks with the following exceptions:

1. Graphics-Magick*:
-- The HWB Color Space and Local Adaptive Thresholding tests were a tie.
-- The Resize, Sharpen, and Blur operations were wins for the ARM chip.

* This one is using OpenMP that would normally be a good thing, but I'm interested in re-testing with OpenMP disabled since this is only a single core CPU and OpenMP may actually be slowing it down in this situation.

2. HMMER: the ARM board was 13 seconds faster on average

3. N-Queens: ARM still wins here, but remember this was a strange benchmark that did not actually compile on my notebook. I need to investigate this one more.

Where my Crippled Core 2 wiped the floor with ARM:

LZMA;
OpenSSL;
DCRAW;
HIMENO;
FLAC & MP3 ENCODING.

Don't forget that your laptop has a massively faster storage system in it (presumably a 2.5" HD) Vs what that ARM board used (SDC flash) in the phoronics benchmarks. While your HD might have no trouble keeping up with reading an audio or video file and saving the encoded version back to the same drive a flash card could become a bottleneck.

I'm not saying this is the cause of the ARMs being slower just that it muddies the waters sufficiently to make this whole thing a waste of time especially as phoronics doesn't make it clear what Sandisk SDC is being used.