View Full Version : AI7 and Gat settings
Monge
06-04-2004, 10:41 AM
I'm trying an Abit AI7 board with a 2.4GHz/800 M0 stepping and 2*256 Kingston KHX3200 and I can get the performance mode enabled with the following settings:
A:
2400/200 MHz
By CPU
SPD
Vcore = Default
Vdimm = Default
SPD 2-2-2-5
F1-Auto-Auto-Auto-Auto
Performance mode: Enabled
B:
202-235 MHz
PSB667
1:1
Vcore = Default
Vdimm = 2.6v and up
SPD 2-2-2-5
Gat: Auto-Auto-Auto-Disabled-Disabled
Performance mode: Enabled
C:
250-255-260 MHz
PSB667
5:4
Vcore = Default
Vdimm = 2.6v Vdimm = 2.7v at 260 Mhz
SPD 2-2-2-5
Gat: Auto-Auto-Auto-Disabled-Disabled
Performance Mode = enabled
I'm unable to get performance mode enabled with PSB800 and 5:4 divider with this settings:
D:
260 MHz
PSB800
5:4
Vcore = Default
Vdimm = 2.7v
SPD 2-2-2-5
Gat: Auto-Auto-Auto-Disabled-Disabled
Performance Mode = Disabled
but it works with 1:1 from 200 and up.
With A and B settings the unbuffered Sandra score is 3/5 of buffered score and this is, what I expect.
But with C settings (Divider 5:4) the unbuffered Sandra score is only 1/2 of the buffered score.
Why?
And how do I get the performance mode enabled with NB800 and 5:4?
enzoR
06-04-2004, 12:21 PM
you can never enable it useing a NB strap of 800
macci
06-04-2004, 12:45 PM
And how do I get the performance mode enabled with NB800 and 5:4?
You don't. Its i865PE chipset limitation. 5:4 + PAT Only works when ran in ('unoffical') 667-mode.
But with C settings (Divider 5:4) the unbuffered Sandra score is only 1/2 of the buffered score.
Why?
Because your running the RAM in 5:4 mode. It means lower RAM MHz and it means lower realworld performance (Unbuffered result).
The buffered test only measures 'theoretical' bandwidht which is in most cases not relevant piece of info at all..
Monge
06-04-2004, 02:26 PM
Originally posted by macci
You don't. Its i865PE chipset limitation. 5:4 + PAT Only works when ran in ('unoffical') 667-mode.
Thanks :)
So will 5:4 + PAT work in an I875 board like ASUS P4c800 dlx-e or will this also give relative poor unbuffered scores at 5:4?
Because your running the RAM in 5:4 mode. It means lower RAM MHz and it means lower realworld performance (Unbuffered result).
The buffered test only measures 'theoretical' bandwidht which is in most cases not relevant piece of info at all..
And this will be the same with the I875 chipset running 5:4 mode?
Because maybe I should swap the AI7 for a P4C800 dlx-e.
My 2.4GHz will run at least 275 MHz at 1.675v.
eva2000
06-05-2004, 12:50 AM
try 5:4 800fsb with GAT = F1-A-A-D-D
Monge
06-05-2004, 03:19 AM
try 5:4 800fsb with GAT = F1-A-A-D-D
I just tried 5:4 PSB800 F1-A-A-D-D and performance mode = Disabled
but 1:1 PSB800 F1-A-A-D-D and performance mode = Enabled
with Kingston 2*256MB KHX3200/256.
eva2000
06-05-2004, 07:16 AM
using ctiaw you might find that would = Partial PAT not disabled
Monge
06-05-2004, 08:52 AM
Yes, that's true and this means, if we compare to an I875 chipset?
Would CPIAW tell the same with an I875 with 5:4?
CPU-Z says Performance Mode = Disabled
Like Macci said 5:4 + PAT only works "fully" when ran in 667-mode on an I865PE and the unbuffered Sandra score suffer with the 5:4 setting.
So my humble conclusion must be that it's preferable to run 1:1 with the AI7.
eva2000
06-08-2004, 05:52 AM
AI7 has always been a 1:1 performer :)
875P will show ctiaw report PAT enabled at 5:4 too
Monge
06-08-2004, 02:21 PM
Also memtest86+ 1.15 shows "Pat: Disabled" with 800PSB 5:4 with GAT = F1-A-A-D-D on AI7
eva2000: Thanks for the help anyway :)
Monge
06-16-2004, 08:57 AM
eva2000: Nice investigation :)
Abit AI7 17 bios - different levels of PAT compared [comparison table]
http://forum.abit-usa.com/showthread.php?s=&threadid=53614
I tried these settings with 5:4 and NB Strap PSB667 today with 2*256MB Kingston KHX3200:
AI7 17 Bios CPU:Memory = 5:4 and according to CPU-Z and CTIAW
NB Strap PSB667 GAT = A-A-A-D-D = Fully Enabled PAT
NB Strap PSB667 GAT = T-A-A-D-D = Fully Enabled PAT
NB Strap PSB667 GAT = Str-A-A-D-D = Fully Enabled PAT
NB Strap PSB667 GAT = F1-A-A-D-D = Not possible to get into Windows
eva2000
06-16-2004, 05:11 PM
i see ya found my comparison thread :)
Yup 667fsb strap always enables FULL PAT :)
Clevor
06-23-2004, 02:42 PM
On the Abit 865P boards, just run SR-Enh-A-D-D with the 800 strap and you will get partial PAT at 5:4 or 3:2. CPUZ and AIDA will say PAT is disabled but no big deal as the unbuffered bandwidth is excellent, and that's what counts. Most BH-5/6 will run SR-Enh-A-D-D on the 800 strap OK.
On the AI7 board, I got 3600+ unbuffered at 289, 5:4, 2-2-2-5 with 2x512 MB XMS3500 (see below). What me worry CPUZ/AIDA say (full) PAT is disabled? :p: Remember, this is 3600+ unbuffered at 5:4, not 1:1. :)
Even at 302 fsb, 3:2, 2-2-2-5 I got around 3200. Running the same settings on an IC7 875P board netted me 3150 unbuffered even with PAT on, and not unexpectedly, I got a lower 3DMark2001 score. So an 875P board doesn't always beat an 865P board.
eva2000
06-23-2004, 02:48 PM
hey Clevor, read that you said newer AI7 seem to overvolt the 3.2v vdimm option ??? my was first rev1.0 AI7 when it came out 3.2v gives me 3.16-3.19 averaging 3.18v
Clevor
06-23-2004, 03:10 PM
Originally posted by eva2000
hey Clevor, read that you said newer AI7 seem to overvolt the 3.2v vdimm option ??? my was first rev1.0 AI7 when it came out 3.2v gives me 3.16-3.19 averaging 3.18v
Yeah, 3.2 shows as 3.278 with a multi :D. I have another AI7 I got in Feb. 2004 and that one is like yours exactly. That board is a C3-000693 and the latest one is a C4-004793. I scrutinized both boards and did not see any hardware differences, except a different batch of chips were used on some components. This is in contrast to the NF7-S, where there is a theory Abit has made secret revisions of the board and mosfets seem to come and go on the board at random. The screenie above is with the old AI7 board.
I still need to check if I can reproduce the same bandwidth with the new board.
Monge
06-25-2004, 05:38 AM
Originally posted by Clevor
On the Abit 865P boards, just run SR-Enh-A-D-D with the 800 strap and you will get partial PAT at 5:4 or 3:2. CPUZ and AIDA will say PAT is disabled but no big deal as the unbuffered bandwidth is excellent, and that's what counts. Most BH-5/6 will run SR-Enh-A-D-D on the 800 strap OK.
On the AI7 board, I got 3600+ unbuffered at 289, 5:4, 2-2-2-5 with 2x512 MB XMS3500 (see below). What me worry CPUZ/AIDA say (full) PAT is disabled? :p: Remember, this is 3600+ unbuffered at 5:4, not 1:1. :)
Even at 302 fsb, 3:2, 2-2-2-5 I got around 3200. Running the same settings on an IC7 875P board netted me 3150 unbuffered even with PAT on, and not unexpectedly, I got a lower 3DMark2001 score. So an 875P board doesn't always beat an 865P board.
Great :)
My AI7 17 BIOS, 2.4 MO stepping and 2*KHX3200/256 at:
260 and 270MHz
PSB800
5:4
Vcore = 1.625v
Vdimm = 2.7-2.9v
Memory: Manual 2-2-2-5
Gat: SR-Enh-Auto-Disabled-Disabled
ctiaw: Pat Partially Enabled
I get Sandra unbuffered score around 2800 with the Gat: SR-Enh-Auto-Disabled-Disabled PSB800 and 5:4 at 260Mhz and 2900 unbuff. at 270 MHz.
My AI7 is a new one from May04.
My AI7 overvolts 0.05-0.07V on Vdimm.
echow87
06-25-2004, 11:48 AM
...So how do you enable 5:4 on the AI7?
Because everytime I try 5:4 to o/c it doesnt work, it crashes, but with 1:1, it works fine.
eva2000
06-25-2004, 11:58 AM
i see Clevor let us know how the 2 boards go bandwidth wise :)
echow87, you need NB Strap 667 with 5:4
zillaoc
06-25-2004, 01:22 PM
yea ime also very interested in teh new rev 1.5 and the performance differences...
vBulletin® v3.7.0, Copyright ©2000-2009, Jelsoft Enterprises Ltd.