How is Zig so fast?

[mattheww95]

After running all of the benchmarks In WSL2 running Debian, It appears that Zig has come out on top with quite a large lead and I was wondering if anyone noticed a similar outcome? If so I was hoping someone could provide a potential explanation for how Zig is so much faster than both C code and assembly code. I have attached the top results below and repeated the benchmarks yet identified the same outcome. If I have misinterpreted the results, please let me know.

I was going through the Zig code for solution 3, but I am inexperienced when it comes to C languages and I am struggling to fully understand how this performance was achieved. I did notice a pre-generation step was included in their code, and was wondering if anything is being calculated at compile time as the Zig compilations took up quite a bit of disk space.

I want to believe that the language is just that fast, but the difference in performance between Zig and languages like C and Assembly just feels a bit off especially when going against the optimized code compilers like gcc produce.

Just in case it is relevant I am using an AMD 3900x and running Windows 10.

Index	Implementation	Solution	Label	Passes	Duration	Threads	Algorithm	Faithful	Bits	Passes/Second
1	zig	3	ManDeJan&ityonemo-zig-single-bitSieve-u8-480of2310	68344	5.00009	1	wheel	yes	1	13668.55397
2	zig	3	ManDeJan&ityonemo-zig-single-bitSieve-u8-5760of30030	63058	5.00005	1	wheel	yes	1	12611.47389
3	zig	3	ManDeJan&ityonemo-zig-single-bitSieve-u64-5760of30030	47210	5.00017	1	wheel	yes	1	9441.67898
4	zig	3	ManDeJan&ityonemo-zig-single-bitSieve-u32-5760of30030	46142	5.00014	1	wheel	yes	1	9228.14161
5	zig	3	ManDeJan&ityonemo-zig-single-bitSieve-u64-480of2310	43012	5.00020	1	wheel	yes	1	8602.05592
6	zig	3	ManDeJan&ityonemo-zig-single-bitSieve-u32-480of2310	42058	5.00020	1	wheel	yes	1	8411.26355
7	zig	3	ManDeJan&ityonemo-zig-single-sieve-u8	33712	5.00011	1	base	yes	8	6742.25167
8	zig	3	ManDeJan&ityonemo-zig-single-bitSieve-u32	21682	5.00019	1	base	yes	1	4336.23522
9	c	2	danielspaangberg_48of210	21611	5.00004	1	wheel	yes	1	4322.16283
10	zig	3	ManDeJan&ityonemo-zig-single-bitSieve-u64	18998	5.00032	1	base	yes	1	3799.35684
11	c	2	danielspaangberg_8of30	18911	5.00007	1	wheel	yes	1	3782.15083
12	c	2	danielspaangberg_5760of30030_owrb	18738	5.00011	1	wheel	yes	1	3747.51905
13	zig	3	ManDeJan&ityonemo-zig-single-sieve-bool	18532	5.00050	1	base	yes	1	3706.0294

[mattheww95]

After going through the pull requests and re-running the bench marks. I see that the Zig entries are now marked unfaithful.

However, I am still amazed by the performance of the code from various languages that have been brought together.