TurboSlugs: Adventures in overclocking
- THIS WILL (oh so) VOID THE WARRANTY ON YOUR NSLU2.
From the early days of the nslu2-linux project, the developers and
many users have noticed certain signs that the slug was not running
at 266MHz core speed.
Be it comparing benchmarks with dev boards, looking at bogomips,
or running Dhrystone tests, the slug appeared to be running at 133MHz
In the last 24 hours, we have discovered this suspicion to be true,
as well as a way to modify the slug for true 266MHz core speed.
Remember, this will void the warranty on your NSLU2. We take no
responsibility for your actions.
Intel Application Note 25406701 talks of ways to downclock IXP42x
using the Expansion Bus Configuration Register 0
(see http://www.intel.com/design/network/applnots/254067.htm ).
After much testing with scopes, writing of ARM assembly, and more
scope testing, it was discovered that removing a resistor on the board
will cause it to run at 266MHz core speed.
Bear in mind that only the processor core speed changes - the RAM bus
and PCI bus speeds are unchanged (133MHz and 33MHz respectively).
While this is not a true "doubling" of whole device speed, it has led
to some significant speedups:
Perl compile went from 90 minutes to 60 minutes
Python compile went from 52 minutes to 36 minutes
For instructions on how to perform the mod, see
You can find a new "TurboSlugs" database on the Yahoo site listing
slugs which have been modified in this manner.
Pretty much the entire dev team contributed to this discovery.
Did we mention that this will (like totally) void the warranty on your
NSLU2, and that we take no responsibility for your actions? Sure we
did. Don't bother sending your legal folk our way ...
The NSLU2-Linux Development Team
- --- In firstname.lastname@example.org, Rod Whitby <list.yahoo@r...>
> On 7/12/05, Rob Lockhart <rlockhar@g...> wrote:friend or
> > I would suggest if you can't do this yourself, ask either a
> > someone at a nearby EE hardware lab. You can't use a bluntsoldering
> > iron tip to do this without risking possible damage to the padsand
> > corresponding traces.in
> Note that we had a choice between two resistors for this mod, and we
> chose the one which was safest to remove with a blunt instrument.
> There are no traces underneath or running parallel to the resistor
> question, so it would be pretty hard to damage something.Thanks Rod, that's good to know. I may end up doing something like
Jason did with a snipper of some sort.
The detailed tips from Jeff and Rob are inspiring though and much
appreciated. I'm going to open up my slug, take a look at the task,
and see if I'm up for a little desoldering.