Problem with speed 23800 RAR3-p (Compressed)

[Crew]

Good afternoon. I ask for help with my question.

I have a clear suspicion that after some update (it is very possible that after fixing the bug with "false positives" on the 23800 RAR3-p (Compressed) hash) the speed of working with this type of hashes became extremely unstable and much slower.
Previously, the average speed was 170-190 kH/s on one GPU 4090, below is an example for 8 GPU

Code:

------------------------------------------------------------
* Hash-Mode 23800 (RAR3-p (Compressed)) [Iterations: 262144]
------------------------------------------------------------

Speed.#1.........:  197.6 kH/s (16.20ms) @ Accel:32 Loops:16384 Thr:512 Vec:1
Speed.#2.........:  196.0 kH/s (16.21ms) @ Accel:32 Loops:16384 Thr:512 Vec:1
Speed.#3.........:  197.4 kH/s (16.17ms) @ Accel:32 Loops:16384 Thr:512 Vec:1
Speed.#4.........:  196.2 kH/s (16.16ms) @ Accel:32 Loops:16384 Thr:512 Vec:1
Speed.#5.........:  196.6 kH/s (16.17ms) @ Accel:32 Loops:16384 Thr:512 Vec:1
Speed.#6.........:  196.9 kH/s (16.25ms) @ Accel:32 Loops:16384 Thr:512 Vec:1
Speed.#7.........:  201.9 kH/s (15.37ms) @ Accel:32 Loops:16384 Thr:512 Vec:1
Speed.#8.........:        0 H/s (0.00ms) @ Accel:32 Loops:16384 Thr:512 Vec:1
Speed.#*.........:  1382.7 kH/s

Source https://www.onlinehashcrack.com/tools-be...x-4090.php

I pulled up my notes from February, that was real speed.

At the moment, when measuring the speed on different PCs, I get data that is significantly lower in speed and with a larger range (on average from 28034 H/s to 8753 H/s on a 4090 GPU).

I'm not talking about the 5090 results now, given its relatively recent release, but the data there is also obviously not up-to-date. The best I could get was 114.7 kH/s on the 5090, which is less than the 4090 had previously.

In my opinion, there is obviously some kind of bug here.

I ask for help and advice, what can be done in this situation? Or should I create an issue on GitHub?

[penguinkeeper]

Where are you seeing this decrease in speed? In your own attacks or in -m 23800 -b? If it's in your own attacks, just know that there's a setting called "iterations" that, when increased, will make the hash significantly slower so the only apples-to-apples comparison is using -b because that's what onlinehashcrack used

[Crew]

Exclusively from my attacks. If in February, when using 14*4090 (on vast.ai) I got a total speed of around 1400-1600 kH/s, now I barely get a total speed of 110 - 120 kH/s. This is a fact that I observe.