CPU advantage

[azory]

Hey people,

The 16char password registry process is very funny. Anyway!

I have this massive CPU (Intel i9-7980XE) and wishing to take advantage of it to decrypt some WPA pass’s. On other hand I’m running with a 1070 nvidia GPU.

The-Distribution-Which-Does-Not-Handle-OpenCL-Well (Kali) 2020 ofc. 

GPU is running at 288kH/s.
CPU runs at 120kH/s.

So, I’m wondering if you guys could help me with this data. If it is normal to have these speeds and // or if there’s any chance to improve them.

My question here is mostly related to the CPU as it is powerful, given its specs, however running in poor performance comparing to GPU. 

Thanks,
Azory

[slyexe]

Different type of hashes correspond better with GPU than CPU and vise versa. most SLOW hashes work better with CPU where as fast hashes work much better with GPUs. I am not to knowledgeable in the topic of why but maybe phil can spread some light on it.

[azory]

Right. Trying to find out on my mind what’s a slow and fast hash, maybe concerning its length and complexity?

In a way or another 100kH/s for such a powerful CPU ...well I think it’s extremely questionable (in a first approach). We are talking here about 18 physical processors (32 logical threads). I mean, in some extent - even if only theoretical - output rate should be multiplied at least by 18x...

Even the GPU rate I see it as poor as well even though I updated nvidia drivers and opencl to the latest.
Meanwhile, been checking the web for some values regarding GPU mostly and I see people having better rates with a with a cheaper builds.

I wish some guidance at lest in this subject, to take the best out of my setup.

Thanks.

[philsmd]

I would say in mostly all of the cases even slow hashes are faster on GPU.\

so the statement about slow hashes being faster on CPU is (in general) completey incorrect.

There are only few exceptions to it: algorithms that are designed to be GPU-unfriendly and which use a lot of (random) memory access and a huge amount of memory or other GPU-unfriendly counter-measures (like bcrypt and scrypt with very high cost/parameters). But even with algorithms like bcrypt, it depends a lot... e.g. a system could have 4 GPUs in it, but it's very rare that a PC has more than 1 or 2 CPUs ... the bcrypt cost factor/iterations matter a lot too and therefore there are some thresholds when it is more efficient to use the CPU compared to the GPU speed (you could just test with --benchmark if not sure).

In this particular instance, it's not a good argument to think that 32 threads are a lot... just have a look at how many cores your GPU has (hint: it has several thousands !), but be aware that each core of a GPU is not as performant or "flexible" as a CPU core (also see https://hashcat.net/faq/morework which explains some of these concepts very well)