Benchmarks

Sunspider 0.9.1

These results were obtained using a laptop-class Intel Core i7 Q720.

Test Jurassic v1 Jurassic v2 Chrome 5 IE 8 Jint 0.8.9
3d-cube 1,904.3 426.0 21.4 197.8 14,212
3d-morph 3,917.0 210.5 19.4 194.2 9,474
3d-raytrace 1,899.1 430.8 17.6 236.6 FAIL
access-binary-trees 562.9 490.6 2.2 151.6 2,749
access-fannkuch 14,865.0 566.3 18.4 495.0 12,374
access-nbody 971.6 215.4 14.0 241.6 4,593
access-nsieve 7,962.8 430.5 5.6 133.4 6,663
bitops-3bit-bits-in-byte 167.7 30.4 2.8 127.6 4,088
bitops-bits-in-byte 167.7 25.3 9.6 137.2 5,938
bitops-bitwise-and 290.2 51.5 11.0 361.6 1,939
bitops-nsieve-bits 347.3 266.4 14.0 252.8 FAIL
controlflow-recursive 220.4 43.1 3.0 137.2 6,592
crypto-aes 5,429.9 460.6 10.6 171.6 FAIL
crypto-md5 995.9 326.8 7.8 111.8 FAIL
crypto-sha1 1,111.1 175.0 7.4 113.4 FAIL
date-format-tofte 5,149.1 2,676.2 22.6 211.6 FAIL
date-format-xparb 696.9 243.1 24.0 198.0 3,998
math-cordic 2378.7 157.9 13.6 263.6 7,146
math-partial-sums 190.0 76.7 18.8 172.4 5,501
math-spectral-norm 1,246.5 128.0 6.8 158.2 6,766
regexp-dna 769.1 724.8 13.8 205.2 FAIL
string-base64 1,338.1 711.6 14.2 174.6 FAIL
string-fasta 919.4 340.3 17.8 259.8 13,753
string-tagcloud 2,162.0 605.2 28.6 177.4 16,583
string-unpack-code 3,780.4 2,668.9 32.2 161.8 FAIL
string-validate-input 10,166.6 364.9 19.0 168.2 10,080
Totals 74,683.8 12,846.8 376.2 5,214.2


V8 Benchmark Suite v6

These results were obtained using a laptop-class Intel Core i7 Q720.

Test Jurassic v1 Jurassic v2
crypto 27,308.3 1,669
deltablue 551.4 372
earley_boyer 27,460.3 86,424
raytrace 2,341.0 2,089
regexp 3,041.2 4,063
richards 205.4 152
splay 6,873.0 5,589


Last edited Dec 27, 2010 at 4:55 PM by paulbartrum, version 23

Comments

Petoj87 Jun 27, 2014 at 11:40 AM 
I Would be nice if the tests where updated for Jurassic 2.2, Jint 2.1, IE11, Chrome 1000 or what ever version chrome currently is.

paulbartrum Sep 2, 2013 at 3:01 AM 
FlorianRappl: I would say "a lot". IMO the biggest gain would be getting more static typing. There are lots of techniques to do this. Method specialization would help a lot (i.e. generating different versions of a function based on the types of the parameters). Better type inference (propagating static types through loops and conditionals, try/catch, etc). Gathering statistical type information via hotspot-like profiling (with fallbacks in case types change).

FlorianRappl Sep 1, 2013 at 6:53 PM 
You already did a great job in optimizing from v1 to v2. The question that interests me: What (kind of performance gain) is still possible?

paulbartrum Apr 4, 2013 at 10:24 AM 
Numbers are milliseconds, and lower is better.

stonstad Apr 3, 2013 at 4:57 PM 
What unit are these numbers in? Is lower better or higher??