"Skid" wrote in message
news:vdbTb.156614$sv6.860918@attbi_s52...
It's not a simple topic. I can tell you that PAT is enabled on my IC7 at
either 1:1 or 5:4. I can still tweak memory settings and overclock to the
limits of my ram with PAT enabled.
When I went from two to four sticks of ram, PAT was disabled at 5:4 but
worked at 1:1 or when the NB strap was set from 800 to 667 -- which
limited
fsb overclocking and lowered performance. The Abit released a new bios
update that got all the performance back -- even improved it at 5:4 --
though PAT still reads disabled. The 5850+ Sandra memory scores I'm
getting
at 5:4 now are right up there with the PC4000 1:1 scores I see with PAT
enabled. That leads me to believe it's actually working, but just not
registering in the apps that normally pick up on it -- like CPU-Z or
Sandra.
And if you think that's confusing, try to figure out what GAT does, or how
to get it working at anything other than the default AUTO settings.
According to an FAQ at Mushkin, setting GAT at anything other than AUTO
overrides both the SPD and manual timings and runs the memory at 1:1 at
settings that equate to 2,2,2,5 in F1 -- faster than most memory will run
comfortably. I've never had any luck at all with it, any setting I select
either makes the system less stable or shows no measureable improvement.
The hassles with getting GAT configured are one of the main reasons I
plunked down an extra $20 for a Canterwood board. After reading some of
the
laments from those trying to emulate the results from the review sites on
Springdale boards using GAT, I'm glad I did.
In general, the Canterwoods clock higher and don't have to resort to
tricks
to get good memory performance. Why settle for a simulation when the real
deal doesn't cost much more?
I've an Asus P4P800 with Hyperpath (the name Asus gives to PAT, see in the
article hereafter why yhey did) enabled. The fsb is overclocked at 240 MHz.
DDR are Corsair Twin 1 GB (2x512 MB) PC3700 at 1:1 and cas settings of
3-4-4-8. Sandra benchmark gives 5457 - 5334. Compared to the normal values
of 5007 - 5005 with the 875P chipset, I found my values high enough and
stopped further overclocking.
If you want detailed information about the PAT story, there is an excellent
article on the link below. There is also an extensive memory benchmark with
the P4P800 and some personal considerations of the testers. A must read !
http://www.xbitlabs.com/articles/mai...us-p4p800.html