5 ms·

I'm increasingly intrigued by Rust. Can anyone recommend a good analysis / benchmarking of Rust compared to C/C++ code? Curious how it fares on a performance ba

I'm increasingly intrigued by Rust. Can anyone recommend a good analysis / benchmarking of Rust compared to C/C++ code? Curious how it fares on a performance basis.

bluejekyll 7y ago
This is by no means scientific, but it’s a fun benchmark, there’s some debate about the fastest Rust options submitted (ie I believe there were faster than C versions that were rejected). C: https://benchmarksgame-team.pages.debian.net/benchmarksgame/fastest/rust.html https://benchmarksgame-team.pages.debian.net/benchmarksgame/... Clang (for common llvm backend between rust and c): https://benchmarksgame-team.pages.debian.net/benchmarksgame/fastest/rust-clang.html https://benchmarksgame-team.pages.debian.net/benchmarksgame/... C++: https://benchmarksgame-team.pages.debian.net/benchmarksgame/fastest/rust-gpp.html https://benchmarksgame-team.pages.debian.net/benchmarksgame/... Rust is pretty much on par with both C and C++.
- rumanator 7y ago
  > This is by no means scientific, It's actually very scientific (testing hypothesis, measuring results, etc) but might not be generalizable.
  - tasubotadas 7y ago
    What's missing in these tests is that they do not capture the required knowledge by the developer to achieve those results. If anybody has actually checked C/C++ examples, one could see that they use non-trivial techniques/libraries to achieve superior performance. What would be REALLY interesting to see is breakdown of the performance by the years of experience of people that submitted working code examples. I have a feeling that a developer with one year of experience in Java, will probably beat developers with two years of experience in C/C++.
    - igouy 7y ago
      Do you have an example of program performance measurements that do "capture the required knowledge by the developer" ? How can you tell that the knowledge was actually "required" rather than incidental ? Wouldn't it depend exactly what experience they gained in that one year and how relevant it was to the task at hand?
- igouy 7y ago
  > I believe there were faster than C versions that were rejected Believe but do not know? Shade?
- pkolaczk 7y ago
  On par is an understatement. It beats C++ in 8 benchmarks and loses in 2. Quite impressive. Rust has more potential for better optimisation than C++ because the compiler has full information on pointer aliasing.
  - igouy 7y ago
    Always look at the source code (or even the CPU load). One of those C++ comparisons is currently multi-core vs sequential. Several C++ programs were recently removed, which upends things. You might say — "there were faster than [Rust] versions that were rejected".
    - pkolaczk 7y ago
      One is multicore, but the other 7 winning have the same parallelism. As for that one winning by parallelism - maybe it was just easier to provide a correct parallel solution in Rust than in C++? This game didn't appear there yesterday and C++ developers had much more time to play it. If that's the case, that benchmark still counts.
      - igouy 7y ago
        Maybe it was just that I contacted the Rust programmer and suggested they might like to update their program :-) (The changes were in the last month or two.)
damnyou 7y ago
You can expect it to be in the same ballpark as C or C++ for single-threaded code, especially if you use idiomatic Rust (combinators instead of manual iteration). Where Rust shines is that it often takes jaut a few minutes to turn a single-threaded program into a multithreaded program — and the type system guarantees that the program won't have data races.
pornel 7y ago
Technically it's about the same speed, but idiomatic Rust programs are structured differently, which has its ups and downs. I wrote an overview of Rust's performance characteristics: https://kornel.ski/rust-c-speed https://kornel.ski/rust-c-speed

I'm increasingly intrigued by Rust. Can anyone recommend a good analysis / benchmarking of Rust compared to C/C++ code? Curious how it fares on a performance basis.

bluejekyll 7y ago

This is by no means scientific, but it’s a fun benchmark, there’s some debate about the fastest Rust options submitted (ie I believe there were faster than C versions that were rejected). C: https://benchmarksgame-team.pages.debian.net/benchmarksgame/fastest/rust.html https://benchmarksgame-team.pages.debian.net/benchmarksgame/... Clang (for common llvm backend between rust and c): https://benchmarksgame-team.pages.debian.net/benchmarksgame/fastest/rust-clang.html https://benchmarksgame-team.pages.debian.net/benchmarksgame/... C++: https://benchmarksgame-team.pages.debian.net/benchmarksgame/fastest/rust-gpp.html https://benchmarksgame-team.pages.debian.net/benchmarksgame/... Rust is pretty much on par with both C and C++.

rumanator 7y ago
> This is by no means scientific, It's actually very scientific (testing hypothesis, measuring results, etc) but might not be generalizable.
- tasubotadas 7y ago
  What's missing in these tests is that they do not capture the required knowledge by the developer to achieve those results. If anybody has actually checked C/C++ examples, one could see that they use non-trivial techniques/libraries to achieve superior performance. What would be REALLY interesting to see is breakdown of the performance by the years of experience of people that submitted working code examples. I have a feeling that a developer with one year of experience in Java, will probably beat developers with two years of experience in C/C++.
  - igouy 7y ago
    Do you have an example of program performance measurements that do "capture the required knowledge by the developer" ? How can you tell that the knowledge was actually "required" rather than incidental ? Wouldn't it depend exactly what experience they gained in that one year and how relevant it was to the task at hand?
igouy 7y ago
> I believe there were faster than C versions that were rejected Believe but do not know? Shade?
pkolaczk 7y ago
On par is an understatement. It beats C++ in 8 benchmarks and loses in 2. Quite impressive. Rust has more potential for better optimisation than C++ because the compiler has full information on pointer aliasing.
- igouy 7y ago
  Always look at the source code (or even the CPU load). One of those C++ comparisons is currently multi-core vs sequential. Several C++ programs were recently removed, which upends things. You might say — "there were faster than [Rust] versions that were rejected".
  - pkolaczk 7y ago
    One is multicore, but the other 7 winning have the same parallelism. As for that one winning by parallelism - maybe it was just easier to provide a correct parallel solution in Rust than in C++? This game didn't appear there yesterday and C++ developers had much more time to play it. If that's the case, that benchmark still counts.
    - igouy 7y ago
      Maybe it was just that I contacted the Rust programmer and suggested they might like to update their program :-) (The changes were in the last month or two.)

damnyou 7y ago

You can expect it to be in the same ballpark as C or C++ for single-threaded code, especially if you use idiomatic Rust (combinators instead of manual iteration). Where Rust shines is that it often takes jaut a few minutes to turn a single-threaded program into a multithreaded program — and the type system guarantees that the program won't have data races.

pornel 7y ago

Technically it's about the same speed, but idiomatic Rust programs are structured differently, which has its ups and downs. I wrote an overview of Rust's performance characteristics: https://kornel.ski/rust-c-speed https://kornel.ski/rust-c-speed