So now they can tell us the world will LITERALLY end in 10 years due to climate change 1 year faster?
I predict the world will end sometime between 3,000 and 30,000 months from now.
-Paul Romer
MIT climate model using Fortran switched to Julia, would accept 3x code slowdown
-
This Julia benchmark report is indeed the most dishonest mischievous benchmark reported so far by JuliaComputing. They do not say if they spent any time updating the Fortran code. They do not say how much time and how many tricks they employed to achieve better performance in Julia. I can take a Julia code, write a Python equivalent that is 100X faster than the Julia version. Does this mean the Julia language is inherently slower? No, it only shows an insincere incorrect presentation of the Julia language.
Unfortunately, the Julia community and the for-profit JuliaComputing behind it desperately try to justify their existence by playing unfair games and presenting insincere flawed benchmarks and comparisons. The truth is, Julia does not really resolve the two-language problem. If you write Python-style code in Julia, it is as slow as Python. and if you want performant code, you have to write code as ugly and complex as C++. -
Oh, shut the fuk up, Zhang.
This Julia benchmark report is indeed the most dishonest mischievous benchmark reported so far by JuliaComputing. They do not say if they spent any time updating the Fortran code. They do not say how much time and how many tricks they employed to achieve better performance in Julia. I can take a Julia code, write a Python equivalent that is 100X faster than the Julia version. Does this mean the Julia language is inherently slower? No, it only shows an insincere incorrect presentation of the Julia language.
Unfortunately, the Julia community and the for-profit JuliaComputing behind it desperately try to justify their existence by playing unfair games and presenting insincere flawed benchmarks and comparisons. The truth is, Julia does not really resolve the two-language problem. If you write Python-style code in Julia, it is as slow as Python. and if you want performant code, you have to write code as ugly and complex as C++. -
This is wrong.
I had a simple simulation in Python and rewrote it Julia. 10x speedup immediately with a verbatim port. Probable reason: lots of loop overhead in Python.
This Julia benchmark report is indeed the most dishonest mischievous benchmark reported so far by JuliaComputing. They do not say if they spent any time updating the Fortran code. They do not say how much time and how many tricks they employed to achieve better performance in Julia. I can take a Julia code, write a Python equivalent that is 100X faster than the Julia version. Does this mean the Julia language is inherently slower? No, it only shows an insincere incorrect presentation of the Julia language.
Unfortunately, the Julia community and the for-profit JuliaComputing behind it desperately try to justify their existence by playing unfair games and presenting insincere flawed benchmarks and comparisons. The truth is, Julia does not really resolve the two-language problem. If you write Python-style code in Julia, it is as slow as Python. and if you want performant code, you have to write code as ugly and complex as C++. -
This Julia benchmark report is indeed the most dishonest mischievous benchmark reported so far by JuliaComputing. They do not say if they spent any time updating the Fortran code. They do not say how much time and how many tricks they employed to achieve better performance in Julia. I can take a Julia code, write a Python equivalent that is 100X faster than the Julia version. Does this mean the Julia language is inherently slower? No, it only shows an insincere incorrect presentation of the Julia language.
Unfortunately, the Julia community and the for-profit JuliaComputing behind it desperately try to justify their existence by playing unfair games and presenting insincere flawed benchmarks and comparisons. The truth is, Julia does not really resolve the two-language problem. If you write Python-style code in Julia, it is as slow as Python. and if you want performant code, you have to write code as ugly and complex as C++.Look at the code used in the benchmarks. No dirty tricks are being used in those benchmarks.
-
The following is the typical Julia community response (from the highest ranks to the lowest) when presented with facts.
Oh, shut the fuk up, Zhang.
Just look at the Julia Discourse community and HackerNews to see the same attitude and language toward any criticism or questions for clarity.
-
The following is the typical Julia community response (from the highest ranks to the lowest) when presented with facts.
Oh, shut the fuk up, Zhang.
Just look at the Julia Discourse community and HackerNews to see the same attitude and language toward any criticism or questions for clarity.
Link or didn't happen.
-
<div style="background-color: lime">
Yes, it DID happen, and it DOES happen. However, since the Julia community does not know how to speak, as evidenced above and below, I will leave that to you to find it (hint: look at some optimization packages in Julia).Julia does NOT resolve the two-language problem because one still has to write two versions of the code to trade elegance with performance. Julia had a good reason to exist in 2010, as its founder says in Julia's Discourse, but (un)fortunately, that need went away fast by the rapid advancements in the Python ecosystem and other languages.
So, what did the Julia community decide to do to justify their existence? They resorted to spreading all sorts of false information and benchmarks and attacks at other languages like Python for being slow, Fortran for being old and dumb, C++ for being complex, ...
This benchmark by Alan Edelman reporting 3X performance gain vs. Fortran culminate such efforts.
But why does the Julia community do this? Because if they do not, the for-profit company behind it goes bankrupt. That's a dirty business strategy.
</div>The following is the typical Julia community response (from the highest ranks to the lowest) when presented with facts.
Oh, shut the fuk up, Zhang.
Just look at the Julia Discourse community and HackerNews to see the same attitude and language toward any criticism or questions for clarity.
Link or didn't happen.
-
This Julia benchmark report is indeed the most dishonest mischievous benchmark reported so far by JuliaComputing. They do not say if they spent any time updating the Fortran code. They do not say how much time and how many tricks they employed to achieve better performance in Julia. I can take a Julia code, write a Python equivalent that is 100X faster than the Julia version. Does this mean the Julia language is inherently slower? No, it only shows an insincere incorrect presentation of the Julia language.
Unfortunately, the Julia community and the for-profit JuliaComputing behind it desperately try to justify their existence by playing unfair games and presenting insincere flawed benchmarks and comparisons. The truth is, Julia does not really resolve the two-language problem. If you write Python-style code in Julia, it is as slow as Python. and if you want performant code, you have to write code as ugly and complex as C++.There is a lot of truth in this; I myself ran some benchmarks to see if all the Julia hype was justified. Out of the box Julia is faster than pure Python for loop intensive code that may not be vectorized. However pure Python code is very rare as most tasks can be farmed out to library routines. In that case, pure Julia is SLOWER than Python. Eventually I decided to stick to my current setup with uses a mixture of Python and C/C++ for structural work.
-
Many economists without a dog in the fight were floored when they found out how much better their Julia code worked compared to Matlab and Python. Optimized Julia code compared to optimized Python code, it's not even close. Julia wins.
I know one person is talking to themselves over multiple IPs to try to bash Julia, so to the other readers out there, look at the benchmarks. Look at the implementation in each language. Look how much faster Julia is over the common code patterns.
-
all I did was make my global variables constants in julia and had a massive speed up. you can use numba but that is so weird to usee.
I'll be honest I actually don't like the julia multiple dispatch though. objects act as namespaces in python and that is very helpful.