Solve the SunSpider mystery: post your results.

[flexy]

Asus Z87-Pro, 4770K at 4.4Ghz, 8GB Memory DDR-2000 10-11-10-1T

Cyberfox 35.0.2 32Bit:

RESULTS (means and 95% confidence intervals)
--------------------------------------------
Total: 5910.2ms +/- 0.5%

***

IE 11.0.9600:

RESULTS (means and 95% confidence intervals)
--------------------------------------------
Total: 6602.0ms +/- 0.4%

***

Chrome:

N/A

For some reason Chrome doesn't give me any results

 

[Makaveli]

I don't run test's on my hardware just cause someone posted a link.........

If you are that paranoid about it run it in a VM.

 

[DrMrLordX]

I don't run test's on my hardware just cause someone posted a link.........

Ah cmon, it's Hans. Do you really think he's going to try and hijack our browsers? If it was some guy with low post count then maybe (or at least let someone else run it and report on behavior first). And of course . . .

If you are that paranoid about it run it in a VM.

. . . right, what you said.

Personally I trust Hans de Vries to not hijack my browser with sketchy javascript.

Btw, Hans, what exact data are you looking for here? I'm guessing you want to see the contrast between the "normal" Sunspider and the "100x loop" Sunspider on different browsers? I see what you are getting at.

IE11 obviously handles Sunspider the most quickly, until the internal test is looped 100 times. Then IE is the slowest.

For me, Firefox 64-bit under Linux was the fastest in handling the 100x loop version, but the slowest in standard Sunspider. Very interestink.

 

[Hans de Vries]

You can always download the benchmark itself and run it locally

Results never showed for me when I did this.

Chrome seems to have an issue if you run the test from hard disc
instead of a website. Some of the test get a 'undefined' result. You
can see this in the result URL. The original code without the x100
for loops has the same issue.

FireFox, Internet Explorer and Safari work fine from hard disc with me.

 

[Hans de Vries]

Ah cmon, it's Hans. Do you really think he's going to try and hijack our browsers?

Thanks

Besides that, anybody can do a diff between the original SunSpider
source code and the modified source code with the for-loop lines

for ( var x100 = 0; x100 < 100; ++x100 )

Btw, Hans, what exact data are you looking for here? I'm guessing you want to see the contrast between the "normal" Sunspider and the "100x loop" Sunspider on different browsers? I see what you are getting at.

IE11 obviously handles Sunspider the most quickly, until the internal test is looped 100 times. Then IE is the slowest.

For me, Firefox 64-bit under Linux was the fastest in handling the 100x loop version, but the slowest in standard Sunspider. Very interesting.

Yes, It would also be interesting for instance to see results from a
Nexus 9 with the Denver NVidia K1. It has a rather low SunSpider result
but that is understandable now:

The sub-benchmarks run only for a few milliseconds and Dever not only
does a JIT compile from javascript to Arm-64bit but also a JIT compile
from Arm-64bit to Native code. No wonder.

Denver should do much better with the 100x version.

 

[imported_ats]

The sub-benchmarks run only for a few milliseconds and Dever not only
does a JIT compile from javascript to Arm-64bit but also a JIT compile
from Arm-64bit to Native code. No wonder.

Denver should do much better with the 100x version.

Yes, but I'm rather unsure if that would actually be a more "correct" result.

 

[Stuka87]

Chrome seems to have an issue if you run the test from hard disc
instead of a website. Some of the test get a 'undefined' result. You
can see this in the result URL. The original code without the x100
for loops has the same issue.

FireFox, Internet Explorer and Safari work fine from hard disc with me.

Ahh ok, will run it from IE tonight on my PC. Or maybe the site will be a bit faster and let me run it from there. My Mac ran it fine from the site, but the PC wouldn't.

 

[Stuka87]

Dell Precision M4600 - i7-2620 @2.7GHz
IE11

============================================
RESULTS (means and 95% confidence intervals)
--------------------------------------------
Total:                 10426.5ms +/- 1.2%
--------------------------------------------

  3d:                   1034.2ms +/- 2.9%
    cube:                372.8ms +/- 1.2%
    morph:               140.0ms +/- 2.7%
    raytrace:            521.4ms +/- 5.8%

  access:               1153.0ms +/- 0.9%
    binary-trees:        193.7ms +/- 1.0%
    fannkuch:            471.0ms +/- 1.5%
    nbody:               321.7ms +/- 1.0%
    nsieve:              166.6ms +/- 2.7%

  bitops:                808.0ms +/- 1.5%
    3bit-bits-in-byte:    66.1ms +/- 4.7%
    bits-in-byte:        131.9ms +/- 6.1%
    bitwise-and:         215.9ms +/- 2.2%
    nsieve-bits:         394.1ms +/- 1.0%

  controlflow:           158.5ms +/- 3.3%
    recursive:           158.5ms +/- 3.3%

  crypto:                729.5ms +/- 2.5%
    aes:                 329.3ms +/- 2.8%
    md5:                 202.2ms +/- 2.5%
    sha1:                198.0ms +/- 3.3%

  date:                 1878.3ms +/- 4.2%
    format-tofte:        682.4ms +/- 4.6%
    format-xparb:       1195.9ms +/- 4.0%

  math:                  940.8ms +/- 2.4%
    cordic:              196.4ms +/- 5.8%
    partial-sums:        606.2ms +/- 2.6%
    spectral-norm:       138.2ms +/- 1.2%

  regexp:                454.1ms +/- 1.6%
    dna:                 454.1ms +/- 1.6%

  string:               3270.1ms +/- 1.0%
    base64:              236.8ms +/- 1.8%
    fasta:               700.9ms +/- 1.9%
    tagcloud:            464.9ms +/- 1.2%
    unpack-code:        1277.7ms +/- 1.3%
    validate-input:      589.8ms +/- 1.8%

Opera 27

============================================
RESULTS (means and 95% confidence intervals)
--------------------------------------------
Total:                  9055.6ms +/- 1.0%
--------------------------------------------

  3d:                   1341.5ms +/- 1.3%
    cube:                289.9ms +/- 2.0%
    morph:               780.2ms +/- 1.7%
    raytrace:            271.4ms +/- 2.8%

  access:                729.5ms +/- 2.9%
    binary-trees:         81.9ms +/- 9.6%
    fannkuch:            380.0ms +/- 2.0%
    nbody:               119.3ms +/- 4.0%
    nsieve:              148.3ms +/- 3.2%

  bitops:                843.1ms +/- 2.1%
    3bit-bits-in-byte:   113.1ms +/- 2.1%
    bits-in-byte:        309.3ms +/- 6.6%
    bitwise-and:         172.3ms +/- 3.0%
    nsieve-bits:         248.4ms +/- 1.8%

  controlflow:           129.3ms +/- 10.1%
    recursive:           129.3ms +/- 10.1%

  crypto:                451.3ms +/- 4.3%
    aes:                 192.2ms +/- 7.4%
    md5:                 150.8ms +/- 5.5%
    sha1:                108.3ms +/- 3.4%

  date:                 1418.6ms +/- 2.0%
    format-tofte:        767.0ms +/- 1.8%
    format-xparb:        651.6ms +/- 2.7%

  math:                  819.8ms +/- 0.9%
    cordic:              196.0ms +/- 1.8%
    partial-sums:        534.4ms +/- 1.2%
    spectral-norm:        89.4ms +/- 1.7%

  regexp:                606.0ms +/- 0.5%
    dna:                 606.0ms +/- 0.5%

  string:               2716.5ms +/- 0.7%
    base64:              247.4ms +/- 1.4%
    fasta:               583.6ms +/- 0.9%
    tagcloud:            293.3ms +/- 2.5%
    unpack-code:        1216.1ms +/- 1.3%
    validate-input:      376.1ms +/- 2.0%

 

[Stuka87]

Yes, but I'm rather unsure if that would actually be a more "correct" result.

Well, what it would mean is MS may be doing some trickery for the quick test. Since those times include the JIT compiler and such. So running the JS is slower (as the long running tests are showing), but all the stuff that happens before the code is actually run is much faster.

So in a real world, its debatable if its noticeable or not. But, MS has been advertising that IE11 is "The worlds fastest Browser" based on benchmarks like this one. When in reality it may not actually be the fastest.

 

[imported_ats]

Well, what it would mean is MS may be doing to trickery for the quick test. Since those times include the JIT compiler and such. So running the JS is slower (as the long running tests are showing), but all the stuff that happens before the code is actually run is much faster.

So in a real world, its debatable if its noticeable or not. But, MS has been advertising that IE11 is "The worlds fastest Browser" based on benchmarks like this one. When in reality it may not actually be the fastest.

Oh certainly, we are basically in the Time To Solution(TTS) dilemma here. It all depends on how much JS the actual webpages load vs how much they use. Fast JIT/slow code or slow JIT/fast code.

 

[DrMrLordX]

Yes, It would also be interesting for instance to see results from a
Nexus 9 with the Denver NVidia K1. It has a rather low SunSpider result
but that is understandable now:

The sub-benchmarks run only for a few milliseconds and Dever not only
does a JIT compile from javascript to Arm-64bit but also a JIT compile
from Arm-64bit to Native code. No wonder.

Denver should do much better with the 100x version.

Sadly, I do not have one available, though someone in the Mobile Devices and Gadgets forum might be able to help you there.

Based on what you and Stuka are saying, it certainly looks as though Microsoft has focused on a different aspect of javascript execution. One wonders if MS' solution to JIT compilation times is fundamentally incompatible with quick execution of the actual js code itself.