The L12 Mod

Does anyone know, specifically, what the L12 mod actually does? I have a
theory that it steps up the voltage to the FSB side of the CPU's L2 cache,
to stabilize a default 133MHz FSB for higher frequencies. It seems to
suggest that Athlon XPs are running at dual voltages but the second one is
automatically set once it detects the CPU type and it's native FSB. Does
this sound correct? Any thoughts are appreciated.

"Tim" wrote in message
...
Does anyone know, specifically, what the L12 mod actually does? I have a
theory that it steps up the voltage to the FSB side of the CPU's L2 cache,
to stabilize a default 133MHz FSB for higher frequencies. It seems to
suggest that Athlon XPs are running at dual voltages but the second one is
automatically set once it detects the CPU type and it's native FSB. Does
this sound correct? Any thoughts are appreciated.

The bios selects a system timing table matched to the CPU's default FSB.
Therefore if you drop in a 133FSB CPU the timing table will be that best
suited to 133MHz. This does not change even if you manually over-ride the
FSB to a higher value in the BIOS. The L12 mod changes the default FSB and
therefore gets the system to selects a timing table more suited to high FSB.
It's nothing to do with voltage.

As least this is what the guys who do the BIOS modding are reporting, they
have even been able to do the L12 mod in the BIOS itself by substituting the
timing tables.

--
*****Replace 'NOSPAM' with 'btinternet' in the reply address*****

Tim wrote:
Does anyone know, specifically, what the L12 mod actually does? I
have a theory that it steps up the voltage to the FSB side of the
CPU's L2 cache, to stabilize a default 133MHz FSB for higher
frequencies. It seems to suggest that Athlon XPs are running at dual
voltages but the second one is automatically set once it detects the
CPU type and it's native FSB. Does this sound correct?

Nope

Given the lack of any significant components on the packaging, it's not
possible to construct a reasonable voltage regulator on the die itself.
Furthermore, there is only one (significant) voltage input to the CPU, so
the entire CPU runs at this single voltage. So the CPU cannot change itself
to cope with higher frequencies. And in any case, why would it? There would
be an almost unmeasurable change in power use/heat output by running the BIU
at a reduced voltage, so they might as well run it at the 200MHz voltage in
all cases and not have to worry about it.

Also, the AC and DC characteristics are specified (rather tightly) in terms
of vcore, so the voltages used on the FSB should not change depending on FSB
frequency (apart from the usual changes resulting from driving lines at a
higher speed).

That said, AMD CPUs almost certainly do internal configuration based on the
bridges, and also based on what they were designated as being at the factory
(regardless of any changes made to the bridges). For the first case,
PowerNow capability is a good example. If the 3rd L5 is cut, PowerNow
capability is disabled. If it's joined, then it's enabled. A good (though
somewhat less conclusive) example for the second point is the recent mobile
Bartons failing to work in SMP mode on some boards (though working on some
others), even when the 3rd L5 has been cut and the chip made identical to a
Barton MP. The location of these changes is unknown. I have my suspicions on
the SIP stream that goes to and from the CPU during initialisation, though
details of what is communicated in this exchange is closely guarded by AMD
and only available to motherboard and BIOS manufacturers under NDA.

Also, there are many undocumented and even actively hidden configurations
registers (called MSRs) in the Athlon, and although motherboard/BIOS
manufacturers have more details available to them than I can get (as
evidenced by them using undocumented MSRs in the BIOS), I beleive the
purpose of a large number of them is only know to AMD engineers. There's
programs to dump the undocumented and the known "actively hidden" MSRs, but
they tend to vary a lot between CPUs so any sort of analysis is very hard
work.

These MSRs control and report back on a vast number of tiny details of the
chip. Optimisations in the execution units and cache controller can be
enabled or disabled for example: in one revision of the Opteron there was a
bug in the L2 cache controller, but that can be fixed through a BIOS update
that disables the troublesome optimisation. It's quite possible that these
MSRs may hold the clues as to why the mobile Bartons are not stable in SMP
mode, and also why the L12 mod has an effect on the degree to which the FSB
can be cranked up.

An equally likely possibility, IMO, is the chipset itself. There's a lot
less known about the northbridges than about the Athlon. NVidia in
particular is very tight lipped about any sort of details on its
Northbridge, an unfortunate hangover from it's GPU-centric history. Heck,
you can't even get PCI configuration register maps for it (AFAIK), let alone
pinouts or any sort of low-level details. This means that there is a very
good chance that something on the chipset is changing depending on the
default FSB reported by the processor.

A final reason is that the Athlon's PLL (the thing in the die that generates
the core frequency from the FSB and the multiplier) simply cannot stabilise
properly when asked to ramp from 133MHz to 200MHz+. Going from 200MHz to
230MHz is not so much of an issue, but if the motherboard ramps the FSB up
too quickly, it stands a chance of making the PLL lose it's lock. I don't
know what happens when the PLL gets lost. There may be some form of recovery
on the CPU, but since the CPU is not designed to have the FSB moving around
there's a chance that it stuffs up the internal state of the CPU. Again, the
low-level technical details on the limitations of the PLL are unavailable to
mere mortals (though available to manufacturers).

--
Michael Brown
www.emboss.co.nz : OOS/RSI software and more
Add michael@ to emboss.co.nz - My inbox is always open

Thank you both for your informative, enlightening replies. I am both
relieved and disturbed to learn how much we *don't* know regarding the
goings-on inside these little slabs of power. After much experimentation I
decided to clock the FSB of my mobile Barton to166 MHz (166x13.5 at 1.55v).
200 MHz just won't run stably, even with chipset and vcore voltage boosts,
and I can't see any real-world performance difference anyway. I know I
could try the L12 mod, on my Abit NF7-S mobo, but I prefer to run the CPU
without it.

Also, I'm cooling the processor with a Speeze (there's a reason it rhymes
with "wheeze") Whisperrock HSF, which has it idling at 48 degrees Celsius.
So until I get better cooling I don't think it's worth trying for higher
speeds. Thanks again for the great info.

"Tim" wrote in message
...
Thank you both for your informative, enlightening replies. I am both
relieved and disturbed to learn how much we *don't* know regarding the
goings-on inside these little slabs of power. After much experimentation I
decided to clock the FSB of my mobile Barton to166 MHz (166x13.5 at
1.55v).
200 MHz just won't run stably, even with chipset and vcore voltage boosts,
and I can't see any real-world performance difference anyway. I know I
could try the L12 mod, on my Abit NF7-S mobo, but I prefer to run the CPU
without it.

Also, I'm cooling the processor with a Speeze (there's a reason it rhymes
with "wheeze") Whisperrock HSF, which has it idling at 48 degrees Celsius.
So until I get better cooling I don't think it's worth trying for higher
speeds. Thanks again for the great info.


You can now get modified bios versions with the L12 mod built in. This is a
link to a thread where you can download such a bios
http://forum.abit-usa.com/showthread...5&pagenumber=2 I'm
using that very bios myself running 230FSB x 11.5 (2645MHz) so I know it
works.

--
*****Replace 'NOSPAM' with 'btinternet' in the reply address*****

"BigBadger" wrote in message
...


You can now get modified bios versions with the L12 mod built in. This is
a
link to a thread where you can download such a bios
http://forum.abit-usa.com/showthread...5&pagenumber=2
I'm
using that very bios myself running 230FSB x 11.5 (2645MHz) so I know it
works.


Thanks for the link. They mention a lot of versions in that thread, which
particular BIOS mod are you using?