TX2 Hashing Performance

TX2 Hashing Performance
Jetson TX2 Module

I recently got 3 TX2(s) donated to me by my friend Sri Krishna Acharya. I did flash it using a comprehensive tutorial provided by him plus slight modifications using an old Macbook Pro running Ubuntu 18.04 (so many NVIDIA quirks here, that this deserves it own blog post).

I was also working on a professional engagement and the discussion about security difference between SHA-512 and SHA-256 came about. The Red-Team guy in me suddenly thought, what if we used HashCat to compare the performance of hashing these, with CUDA support on cheap power envelope, to assess the computation time and thus difficulty of generating enough guesses to match the correct password back.

Here are the results. You can see jetson-stats on the left and ./hashcat --benchmark --backend-ignore-opencl--optimized-kernel-enable on the right.


Here you can clearly see the that SHA-256 is at 223.4 MH/s while SHA-512 at 74.3 MH/s, which makes quite sense as the operations are broken at approx 3X scaling.

Another interesting observation was hashing used in Lastpass, has also been integrated into the test suite, and we can clearly see that this is very costly compared to just 5K iterations present in initial accounts (including mine, originally).

Here almost 900 Hashes / sec are running in approximately 9W of power. Not bad for a such a tiny device. I'll add more posts in future regarding how I achieved this and what all we experiments I did on them.