The question of the OP was why the number of password candidates is not as he expected.
36^4 = 1,679,616 but should be 36^1 + 36^2 + 36^3 + 36^4 = 1,727,604
The answer to this is simple. If you give a mask you need to tell hashcat that you want to search the range. To do so, you need to add the -i option. It will search through the keyspace as you expect it, but the status screen will not be as intuitive as you may think. This goes back to hashcat session strategy is optimized for maximum performance (to the cost of user comfort). Note both crark and hashcat use the same strategy you can see it iterating through the different password lengths in your own crark screenshot.
What makes 7z cracking so different to all other algos is that the length of the password has a direct impact on the cracking performance.
Just some sample numbers. On the same GPU (GTX980 with 524,288 iterations):
mask length 4: 7376 H/s
mask length 5: 5887 H/s
mask length 6: 5451 H/s
mask length 7: 5093 H/s
But that's not all. Of course also the total number of candidates has a direct impact. That's import for the OP for the understanding why his numbers are lower than the one in crark. If you want to have full speed in hashcat you need to give it enough work to do. Please read this carefully: https://hashcat.net/wiki/doku.php?id=fre...full_speed
As a result you will get a much higher performance in hashcat than with crark. Here's an example taken with my GTX980 cracking the same .7z archive using the same password candidate range:
https://imgur.com/a/WCg0A03
hashcat: 7327 H/s
crark: 6099 H/s
36^4 = 1,679,616 but should be 36^1 + 36^2 + 36^3 + 36^4 = 1,727,604
The answer to this is simple. If you give a mask you need to tell hashcat that you want to search the range. To do so, you need to add the -i option. It will search through the keyspace as you expect it, but the status screen will not be as intuitive as you may think. This goes back to hashcat session strategy is optimized for maximum performance (to the cost of user comfort). Note both crark and hashcat use the same strategy you can see it iterating through the different password lengths in your own crark screenshot.
What makes 7z cracking so different to all other algos is that the length of the password has a direct impact on the cracking performance.
Just some sample numbers. On the same GPU (GTX980 with 524,288 iterations):
mask length 4: 7376 H/s
mask length 5: 5887 H/s
mask length 6: 5451 H/s
mask length 7: 5093 H/s
But that's not all. Of course also the total number of candidates has a direct impact. That's import for the OP for the understanding why his numbers are lower than the one in crark. If you want to have full speed in hashcat you need to give it enough work to do. Please read this carefully: https://hashcat.net/wiki/doku.php?id=fre...full_speed
As a result you will get a much higher performance in hashcat than with crark. Here's an example taken with my GTX980 cracking the same .7z archive using the same password candidate range:
https://imgur.com/a/WCg0A03
hashcat: 7327 H/s
crark: 6099 H/s