"md5 -t" as indicator of CPU speed

[Peter Hessler]

It depends on a few things.  Do you have any flags for apmd on start
up?  Did you run apm with any flags?  Its also entirely possible that
your system cannot handle speed changes (Support for *many* cpus was
added recently to -current).  A full dmesg would tell us that.


I boot my (-current) system with `apm -C`, man page excerpt below:

     -C      Start apmd in cool running performance adjustment mode.  In this
             mode, when CPU idle time falls below 10%, apmd raises hw.setperf
             as much as necessary.  Otherwise, when idle time is above 30%
             apmd reduces hw.setperf as much as possible to reduce heat,
             noise, and power consumption.

If you have that enabled, that will cause the results you are seeing.

I ran `md5 -t` with automatic mode disabled, and manually triggered
each CPU speed jump.  Here are my results.

# apm -L (hw.setperf=0)
Time   = 1.095103 seconds
Time   = 1.096470 seconds
Time   = 1.097863 seconds
Time   = 1.094756 seconds
Time   = 1.101693 seconds
# apm -H (hw.setperf=100)
Time   = 0.430246 seconds
Time   = 0.430267 seconds
Time   = 0.430384 seconds
Time   = 0.433962 seconds
Time   = 0.432229 seconds


That indicates that `md5 -t` is a decent way to compare cpu speed.  And
by 'decent' I mean "just as bad as the others".  Some systems have
built-in crypto accelerators, which would speed up md5 calculations by
an incredible amount.  ISTR a mail to misc@ that showed a VIA C3 system
beating the pants off a top of the line Pentium 4.  The VIA chip had a
a crypto accelerator included.

$ apm
Battery state: unknown, 0% remaining, unknown life estimate
A/C adapter state: not known
Performance adjustment mode: cool running (800 MHz)
$ dmesg | grep cpu
...
cpu0: AMD Turion(tm) 64 Mobile ML-37 , 1995.31 MHz
...
cpu0: PowerNow! K8 1995 Mhz: speeds: 2000 1800 1600 800 Mhz



On Thu, 15 Jun 2006 23:15:53 -0400
marius <anarcap at gmail.com> wrote:

: Hi all,
: 
: A really simple question, to which I probably already know the answer
: from googling and testing[1], but I'd like someone to confirm it (or
: beat me with a clue stick)...
: 
: How reliable is "md5 -t" as an indicator of relative CPU speed?
: "sysctl hw.cpuspeed" returns either 1499 or 650 depending on my apm
: settings, however "md5 -t | grep Time >> md5time.txt" always gives the
: same time range: 0.60s +/- 0.015s (I've run this around 50 times with
: various power settings).
: 
: So if "md5 -t" is an appropriate test of relative CPU speed (and from
: what I've seen, it is) then the hw.cpuspeed value is incorrect, and my
: CPU does not slow down. And if that's the case, then I'm one step
: closer to figuring out why SpeedStep does not work on my Acer
: TravelMate2300 laptop.
: 
: Thanks,
: 
: //mts
: 
: [1] I tried "md5 -t" on two additional computers (a P-100 with 32 megs
: of RAM and a P-III 500 with 128 megs of RAM) and their results were
: around 18 and 3 times what I saw on the Acer's 1500 Mhz Celeron-M with
: 768 megs of RAM.
: 
: -- 
: "Goddammit Jim, I'm an anthropologist, not a computer scientist!"
:      me, to a client, 2001
:      (surprisingly, I didn't get fired)
: _______________________________________________
: Openbsd-newbies mailing list
: Openbsd-newbies at sfobug.org
: http://mailman.theapt.org/listinfo/openbsd-newbies
: 


-- 
Philogeny recapitulates erogeny; erogeny recapitulates philogeny.