Suggestions for Benchmarks

[MAA]

Hello,

When the switch to version 8 was made, I expected updated benchmarks; however, that didn't happen, and CheckMate remains the only new benchmark to date.
I had already suggested updating certain benchmarks, but I’ve now consolidated those ideas into a single post.

CPU:
CheckMate    - OK
PhotoWorxx    - should be replaced, for example, with the VVC/EVC/AV2 encoding benchmark
Zlib    - must be replaced with a faster implementation, for example Zlib-ng https://github.com/zlib-ng/zlib-ng
AES    - OK
+However, it is also necessary to add the ChaCha20 (used in TLS 1.3) benchmark, which is relevant for processors without AESNI https://en.wikipedia.org/wiki/ChaCha20-Poly1305
+In addition, it is desirable to add the new AEGIS benchmark (proposed for future use with TLS) https://cfrg.github.io/draft-irtf-cfrg-aegis-aead/draft-irtf-cfrg-aegis-aead.html
SHA3    - OK
+However, it is also necessary to add a SHA-2 benchmark capable of utilizing SHA instructions https://en.wikipedia.org/wiki/SHA_instruction_set

FPU:
Julia    - OK
Mandel    - must be refined for better multithreading support https://forums.aida64.com/topic/13433-threadripper-7975wx-bench-results/#comment-63473
SinJulia    - OK
FP32 RayTrace    - OK
FP64 RayTrace    - OK

+It would also be interesting to see the LINPACK benchmark (or at least DGEMM/SGEMM)

NPU:
some benchmark needs to be added

[MAA]

I also discovered some strange SHA3 results. Why doesn't the Zen 5 architecture gain an advantage over Zen 4? Why doesn't the 512-bit datapath gain an advantage over the 256-bit datapath? Perhaps additional AVX-512 optimizations are required?

[MAA]

If SHA3 can't be optimized, I suggest adding BLAKE3, which supports practically unlimited degree of parallelism (both SIMD and multithreading) https://shattered.io/blake3-vs-sha-512-2026/