lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  PHC 
Open Source and information security mailing list archives
 
Hash Suite for Android: free password hash cracker in your pocket
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Sun, 6 Sep 2015 10:24:16 +0300
From: Solar Designer <solar@...nwall.com>
To: discussions@...sword-hashing.net
Subject: Re: [PHC] Where do authors get these numbers?

On Sun, Sep 06, 2015 at 07:28:44AM +0100, Samuel Neves wrote:
> On 09/06/2015 07:16 AM, Dmitry Khovratovich wrote:
> > What CPU exactly are you testing on? That one has 2.1 GHz. We detected quite big difference in cycle/byte performance on similar CPU with different frequencies.
> 
> Cycles per byte are meant to be independent of the CPU frequency. When external memory gets involved, things do get a
> little more complicated, though.
> 
> Is it possible that such differences are caused by Turbo Boost? Some processors overclock more than others; in
> particular, a 2.1 GHz 4600U chip will overclock to 3.3 GHz, whereas a (say) 3.5 GHz 4770k chip will 'only' overclock to
> 3.9 GHz. Thus the former chip's timings will be significantly more distorted by Turbo Boost than the latter's, even
> though they should both report roughly the same cycle-per-byte value in theory.

A solution is to measure the actual frequency with the intended number
of cores under load (same number as in the target code's benchmark),
verify it against the CPU documentation, and if it matches (it does for
me, on many different CPUs), then just use that and not the CPU's base
clock rate for calculating cycles per byte, etc.

Wikipedia has convenient tables, and so far I haven't found any errors
in them.  For 4600U:

https://en.wikipedia.org/wiki/List_of_Intel_Core_i7_microprocessors#.22Haswell-ULT.22_.28SiP.2C_dual-core.2C_22_nm.29

2.1 GHz + either 9 or 12 turbo bins (100 MHz each), so 3.0 GHz with both
cores in use, 3.3 GHz with 1 core in use.

CPUs usually do run at the max turbo rate for the given number of cores
in use.  In my testing so far, this is always the case for 1 core in
use, and is usually the case for all cores in use.

Alexander

Powered by blists - more mailing lists