hashcat Forum

When I say "manycore CPU" I'm talking about a Xeon Phi-like accelerator, such as if Adapteva ever made the 4096-core variant of their Epiphany chip that they claim they can make.

If you saw Katja Malvoni's talk at Passwords14, you know the 64-core Epiphany pulls 4812 c/s on bcrypt while drawing only 2W of power. Adapteva claims the 4096-core version only draws 80W, and if the performance scales linearly (it should based on the 16-core vs 64-core speeds) then the 4096-core version would pull 308 KH/s on bcrypt. That's 44x-56x faster than today's GPUs & high-end CPUs on a fraction of the power.

I imagine Intel will also release a version of the Phi that will be worthwhile for password cracking at some point in the future, even if it does take them 6 years like it did Nvidia

Yep, that's two different stories leading to the same thing: the CPU power cannot be raised much mure so you add massive parallel RISC cores (a single shaders layout is very similar to a classic RISC ALU) to boost performance. The "graphic card" that's onboard now will become the CPU like the FPU became the CPU when they first nailed it on the 486 die. It'll be fully integrated and optimized and not specified as a graphic card. It might also handle graphic card issues, but that will be only a side thing. I guess we both see the same story here

I didn't see it and just put Malvonis presentation on my watch list, so a big thx for that.