do not think the SunSpider benchmark needs no introduction. E 'has now been taken as a reference a bit' all the browser vendors.
Given that Microsoft has "joined" in this race who's got more 'long ... fast, I tried to make a comparison between IE9 tech preview 2 and Opera 10.53 and Chrome 5.0.375.29.
But the topic of the post is not 'the comparison of these browsers. What left me a bit 'and banned' behavior SunSpider on different machines / operating systems.
The first thing I noticed 'that the results have a variability' significant depending on the system running the test.
On the desktop with Windows 7 (Q9450 quad core 2.66 GHz) results are within a variability 'less than 5% and the best result I got and the' state of 399 ms, while I get on average 405 ms.
on MacBook with Windows 7 (P8400 2.4 GHz dual core) the best result and the 'state of 634 ms and a variable' greater than 10% (on average get 750 ms). 634 ms would also seem acceptable given the different characteristics of the CPU, but as we will not 'so'.
Actually 'and the SunSpider' composed of 9 groups of tests that have in turn in the air for a total of 26 individual tests. The first thing I wanted to understand was how many of them really take advantage of the presence of multi-core, so 'I run the test inside a virtual machine's desktop. I know that the VM uses a single core and the fact there are two of 26 tests which gave a result between 3 and 4 times worse than the quad core. All others were in a range of worsening between 0% and 10%.
It's all right up here, and the problem 'for this type of test perform a wide variety 'substantially negligible amount of code on a' data as insignificant, certain features of the CPUs are not being used, with the result that even a dual core laptop would stand up to a quad core. But in the case of the MacBook is not 'so', I expected a much better result of 634 ms. Above all, I can not explain why the deterioration from the quad core is equally distributed on all 26 tests and has not rather a bit of deterioration on the two tests that use the multi core.
To give you an example: Super PI and that 'single thread on a MacBook pays about 10% -15% worse than the quad core, not 50% or more' as it is for SunSpider.
Ho voluto provare sul Mac anche XP, pensando che magari fosse un problema di incompatibilita' tra Windows 7 e Boot Camp... risultato medio: 1.000 ms!
Ovvero molto peggio rispetto a Windows 7 (anche questo fatto non ha una spiegazione chiara).
La contro prova che Sunspider su un dual core dovrebbe ottenere risultati migliori l'ho avuta confrontando il risultato di Sunspider su OS X: sul MacBook ho ottenuto 426 che e' compatibile con le aspettative dopo aver letto su Arstechnica che su un Mac Pro quad core avevano ottenuto 389.
Per altro ho provato su uno Xeon 5355 2.66 GHz con Windows Server 2003 R2 ed ho ottenuto 330 che e' in linea con i dati di ArsTechnica (puo' essere che il Mac Pro da loro usato has clocked a hair lower).
I also tried the same test on the MacBook VM proven on the desktop, and the decrease in performance and 'was again unexpected best results with averages around 1,695 ms 2,000 ms and a profit even on to 3000 ms. It Does not Make Sense!
Why 'on the quad core machine to virtual real you pay only on the two test dates, while on the Mac so you pay so' flashy about all the tests? Why
'quadcore on the delta between the various tests and execute' limited to less than 5%, while the Mac will arrive in rising to 20%?
Given that the test of his commitment not so much the CPU, so it is not 'that other processes can significantly alter the result.
Whether it's a bug in this SunSpider and that is evident only with certain hardware / operating systems? What is some problem in Opera? That is a problem of the operating system? That is a problem with Boot Camp?
for now rule out is a problem in Opera, I tried Chrome and the% difference between quad core and dual core 'compatible with the results of operations: 349 ms and 686 ms on the desktop Mac
On the desktop I wanted try it under Ubuntu 4.10 but has not yet released the Opera 10:53, just try out. Meanwhile I tried Chrome that scored 399 ms and 7349 ms under Windows, values \u200b\u200bconsistent with normal error variance of this type of test and the fact that surely a little something changes between Windows and Linux.
If you have a dual core I'd be curious to know what is the best result you get (specify the frequency of CPU clock).
SunSpider For now confirms what I always thought: there are lies, big lies and benchmarks;)
Update: even finished publishing the post after another round of ten tests of eseguizioni Mac and 'popped an unexpected 499 ms. While everyone else had settled from 640 ms up '.
<$BlogFeedsVertical$>
0 comments:
Post a Comment