A8N-E Overclocking

[name]

Hi, hope someone can help. I'm in the process of learning how to overclock
my A8N-E with an AMD 64 3200. The research I've done indicates that it's
best to use a low multiplier like 9.5 and in increments of 5 boost the FSB.
I've read reports that with a 9.5 multiplier there were successfull
overclocks to 245-255 FSB with no increase in voltage. All this I
understand, but I have three questions.

1. How does the Hypertransport enter into this? Is it best to leave it on
Auto or should I lower it?

2. Should I enable or disable the AI overclocking feature? Obviously I'll
be in manual settings mode.

3. I'm leaving the PCI settings at their default level, I assume this is
ok.


TIA,
john

 

[Paul]

Hi, hope someone can help. I'm in the process of learning how to overclock
my A8N-E with an AMD 64 3200. The research I've done indicates that it's
best to use a low multiplier like 9.5 and in increments of 5 boost the FSB.
I've read reports that with a 9.5 multiplier there were successfull
overclocks to 245-255 FSB with no increase in voltage. All this I
understand, but I have three questions.

1. How does the Hypertransport enter into this? Is it best to leave it on
Auto or should I lower it?

2. Should I enable or disable the AI overclocking feature? Obviously I'll
be in manual settings mode.

3. I'm leaving the PCI settings at their default level, I assume this is
ok.


TIA,
john

I don't have this board, so if some Athlon64 board owner
reads this, they can correct it.

FSB is the same thing as HT. AFAIK, FSB time HT_multipler
must be less than HT_Rating. Say your HT is good for 1000.
Then 245 x 3 = 735 and 245 x 4 = 980. Both an HT multiplier
of x3 or x4 would work, and the resulting HT frequency would
be less than the 1000 I used in this example. Some people
push their HT well above the limit, for whatever that is
worth (more 3DMarks ?).

There is also a ratio between the FSB and the memory bus.
Some people swear by using a 1:1 divider, while others
feel using non 1:1 dividers is acceptable too. There is
some complicated relationship between the multiplier and
what RAM speed will result, and at least one private
forum web page had a spread sheet to show the possible
values that result.

If you used a 1:1 FSB:Mem divider, then the RAM would run
at 245 x 2 = DDR490, which could be handled by a dual pak
of DDR500 (PC4000) memory. As I understand it, some of the
overclockers are happy when they hit 9 x 300, and run some
TCCD memory at DDR600.

Manual settings is always the way to go. Assuming you can
find the right web page on one of the private forums, you
can likely find a recipe for overclocking the board. Even
if you find instructions for a similar board, the advice is
likely to still be 95% applicable.

An overclocker should have a repertoire of Windows utilities
for examining what is happening while overclocking.
CPUZ is good for reviewing the basic parameters of the system.
http://www.cpuid.com/CPUZ

You can also look for the CG-NVNF4 version of "clockgen" here:
http://www.cpuid.org/clockgen.php

Clockgen will tell you if the PCI bus is really locked at
33.33 MHz or not.

Getting a copy of A64tweaker by CodeRed is another handy
tool for RAM settings. Here is the download, and the
development link. You would hope that CPUZ and A64tweaker
agree on the settings they see.

http://www.akiba-pc.com/download.php?view.40
http://www.xtremesystems.org/forums/showthread.php?s=&threadid=37345

HTH,
Paul

 

[name]

I don't have this board, so if some Athlon64 board owner
reads this, they can correct it.

FSB is the same thing as HT. AFAIK, FSB time HT_multipler
must be less than HT_Rating. Say your HT is good for 1000.
Then 245 x 3 = 735 and 245 x 4 = 980. Both an HT multiplier
of x3 or x4 would work, and the resulting HT frequency would
be less than the 1000 I used in this example. Some people
push their HT well above the limit, for whatever that is
worth (more 3DMarks ?).

There is also a ratio between the FSB and the memory bus.
Some people swear by using a 1:1 divider, while others
feel using non 1:1 dividers is acceptable too. There is
some complicated relationship between the multiplier and
what RAM speed will result, and at least one private
forum web page had a spread sheet to show the possible
values that result.

If you used a 1:1 FSB:Mem divider, then the RAM would run
at 245 x 2 = DDR490, which could be handled by a dual pak
of DDR500 (PC4000) memory. As I understand it, some of the
overclockers are happy when they hit 9 x 300, and run some
TCCD memory at DDR600.

Manual settings is always the way to go. Assuming you can
find the right web page on one of the private forums, you
can likely find a recipe for overclocking the board. Even
if you find instructions for a similar board, the advice is
likely to still be 95% applicable.

An overclocker should have a repertoire of Windows utilities
for examining what is happening while overclocking.
CPUZ is good for reviewing the basic parameters of the system.
http://www.cpuid.com/CPUZ

You can also look for the CG-NVNF4 version of "clockgen" here:
http://www.cpuid.org/clockgen.php

Clockgen will tell you if the PCI bus is really locked at
33.33 MHz or not.

Getting a copy of A64tweaker by CodeRed is another handy
tool for RAM settings. Here is the download, and the
development link. You would hope that CPUZ and A64tweaker
agree on the settings they see.

http://www.akiba-pc.com/download.php?view.40
http://www.xtremesystems.org/forums/showthread.php?s=&threadid=37345

HTH,
Paul

Thanks for the reply.
I'm still a little confused. The "CPU frequeny x CPU multiplier" would
result in the CPU speed, for example, 220 x 9.5 = 2090 CPU speed, but "CPU
frequency x HT frequency" results in what?
In other words, my chip is rated at 2000MHz (AMD 64 3200) is that the same
as the HT because if it is then a CPU frequency of 220 x HT frequency of 5 =
1010. The HT frequency settings on my board go from 1x - 5x or Auto. Does
HT frequency indicate the CPU speed or the speed between RAM and the onboard
memory controller? Pleae clarify.

TIA,
john

 

[Paul]

Thanks for the reply.
I'm still a little confused. The "CPU frequeny x CPU multiplier" would
result in the CPU speed, for example, 220 x 9.5 = 2090 CPU speed, but "CPU
frequency x HT frequency" results in what?
In other words, my chip is rated at 2000MHz (AMD 64 3200) is that the same
as the HT because if it is then a CPU frequency of 220 x HT frequency of 5 =
1010. The HT frequency settings on my board go from 1x - 5x or Auto. Does
HT frequency indicate the CPU speed or the speed between RAM and the onboard
memory controller? Pleae clarify.

TIA,
john

There is an architecture drawing at the bottom of this page.
I'd have liked to use an Nvidia page for this, but I don't
think they like to do this sort of thing:

http://www.via.com.tw/en/products/chipsets/k8-series/k8t890/

FSB:Mem
Channel 0 DDR ------- Processor
Channel 1 DDR ------- FSB x Core_Mult

FSB x HT_Mult = HT_Bus
|
| HT Bus (bidirectional)
|
PCI Express Video ------- Northbridge
|
| Proprietary Hub Bus
|
Southbridge
|
Legacy PCI bus.
Any other slow busses.

Notes:

0) "FSB" and "HTT" terms are used interchangable. "LDT" is
the "HT_Mult". If I owned one of these boards, I would
be more precise in my terminology...

1) In Via architectures, they like to use separate Northbridge
and Southbridge chips. Nvidia is using a combined chip, which
contains both the Northbridge and the Southbridge. The combined
chip will run a bit hotter than two separate chips would, but
might mean a significant cost saving for Asus (chip package
cost is a signicant portion of manufacturing cost).

2) The Hypertransport bus in this case, is the connection from
processor to chipset. Your graphics performance depends on
that 245x3 or 245x4 setting. Also, any I/O going to disks,
LAN, USB, Firewire, is going through the HT bus.

3) FSB x Core_Mult gives the processor core clock.
4) Through some complicated logic, the FSB:Mem divider is
set. I used to have a very nice web page that discusses the
arithmetic, but can no longer find the link I posted in
Google.

This link explains how the memory divider is approximated.
The core clock is divided down, to get the memory clock.

http://web.archive.org/web/20040423180605/www.dugu9tweaks.net/guides/a64oc/index2.html

There is more info here:
http://web.archive.org/web/20041112170147/http://www.bleedinedge.com/forum/showthread.php?t=3697

I had to use the slow web.archive.org server, because the
original links are gone. I don't know why the private forums
are aging out their old posts, when disk drives are so
cheap, and one drive can store a lot of posts. web.archive.org
can be a PITA to use, because hyperlinks on a rendered page
are frequently truncated, and you have to hand craft a corrected
link to jump to another page.

search engine: altavista.com
search terms: athlon64 spreadsheet multiplier fsb

HTH,
Paul

 

[name]

There is an architecture drawing at the bottom of this page.
I'd have liked to use an Nvidia page for this, but I don't
think they like to do this sort of thing:

http://www.via.com.tw/en/products/chipsets/k8-series/k8t890/

FSB:Mem
Channel 0 DDR ------- Processor
Channel 1 DDR ------- FSB x Core_Mult

FSB x HT_Mult = HT_Bus
|
| HT Bus (bidirectional)
|
PCI Express Video ------- Northbridge
|
| Proprietary Hub Bus
|
Southbridge
|
Legacy PCI bus.
Any other slow busses.

Notes:

0) "FSB" and "HTT" terms are used interchangable. "LDT" is
the "HT_Mult". If I owned one of these boards, I would
be more precise in my terminology...

1) In Via architectures, they like to use separate Northbridge
and Southbridge chips. Nvidia is using a combined chip, which
contains both the Northbridge and the Southbridge. The combined
chip will run a bit hotter than two separate chips would, but
might mean a significant cost saving for Asus (chip package
cost is a signicant portion of manufacturing cost).

2) The Hypertransport bus in this case, is the connection from
processor to chipset. Your graphics performance depends on
that 245x3 or 245x4 setting. Also, any I/O going to disks,
LAN, USB, Firewire, is going through the HT bus.

3) FSB x Core_Mult gives the processor core clock.
4) Through some complicated logic, the FSB:Mem divider is
set. I used to have a very nice web page that discusses the
arithmetic, but can no longer find the link I posted in
Google.

This link explains how the memory divider is approximated.
The core clock is divided down, to get the memory clock.

http://web.archive.org/web/20040423180605/www.dugu9tweaks.net/guides/a64oc/index2.html

There is more info here:
http://web.archive.org/web/20041112170147/http://www.bleedinedge.com/forum/showthread.php?t=3697

I had to use the slow web.archive.org server, because the
original links are gone. I don't know why the private forums
are aging out their old posts, when disk drives are so
cheap, and one drive can store a lot of posts. web.archive.org
can be a PITA to use, because hyperlinks on a rendered page
are frequently truncated, and you have to hand craft a corrected
link to jump to another page.

search engine: altavista.com
search terms: athlon64 spreadsheet multiplier fsb

HTH,
Paul

Thanks again, the above has been very helpful, the fog is lifting. You've
given me a good starting point to work from. I've been doing some research
and it seems it's all in the memory timings. I fiddled around with the CPU
frequencies and CPU and HTT multipliers and while the CPU speed increased
the memory bandwidth went through the floor. AIDA showed memory read
dropping from 57xxMBs to 37xxMBs. I beginning to understand with A64s
already low memory latency there some adjustments is memory timings to be
made, otherwise just boosting CPU speed can, actually make your system
slower. I found one good explaination at
www.insaneteck.com/index.php?page=overclocka64 .
Once, I figure all this out I'll post my results, bearing in mind it'll take
awhile.

john

 

[name]

There is an architecture drawing at the bottom of this page.
I'd have liked to use an Nvidia page for this, but I don't
think they like to do this sort of thing:

http://www.via.com.tw/en/products/chipsets/k8-series/k8t890/

FSB:Mem
Channel 0 DDR ------- Processor
Channel 1 DDR ------- FSB x Core_Mult

FSB x HT_Mult = HT_Bus
|
| HT Bus (bidirectional)
|
PCI Express Video ------- Northbridge
|
| Proprietary Hub Bus
|
Southbridge
|
Legacy PCI bus.
Any other slow busses.

Notes:

0) "FSB" and "HTT" terms are used interchangable. "LDT" is
the "HT_Mult". If I owned one of these boards, I would
be more precise in my terminology...

1) In Via architectures, they like to use separate Northbridge
and Southbridge chips. Nvidia is using a combined chip, which
contains both the Northbridge and the Southbridge. The combined
chip will run a bit hotter than two separate chips would, but
might mean a significant cost saving for Asus (chip package
cost is a signicant portion of manufacturing cost).

2) The Hypertransport bus in this case, is the connection from
processor to chipset. Your graphics performance depends on
that 245x3 or 245x4 setting. Also, any I/O going to disks,
LAN, USB, Firewire, is going through the HT bus.

3) FSB x Core_Mult gives the processor core clock.
4) Through some complicated logic, the FSB:Mem divider is
set. I used to have a very nice web page that discusses the
arithmetic, but can no longer find the link I posted in
Google.

This link explains how the memory divider is approximated.
The core clock is divided down, to get the memory clock.

http://web.archive.org/web/20040423180605/www.dugu9tweaks.net/guides/a64oc/index2.html

There is more info here:
http://web.archive.org/web/20041112170147/http://www.bleedinedge.com/forum/showthread.php?t=3697

I had to use the slow web.archive.org server, because the
original links are gone. I don't know why the private forums
are aging out their old posts, when disk drives are so
cheap, and one drive can store a lot of posts. web.archive.org
can be a PITA to use, because hyperlinks on a rendered page
are frequently truncated, and you have to hand craft a corrected
link to jump to another page.

search engine: altavista.com
search terms: athlon64 spreadsheet multiplier fsb

HTH,
Paul

Thanks again, the above has been very helpful, the fog is lifting. You've
given me a good starting point to work from. I've been doing some research
and it seems it's all in the memory timings. I fiddled around with the CPU
frequencies and CPU and HTT multipliers and while the CPU speed increased
the memory bandwidth went through the floor. AIDA showed memory read
dropping from 57xxMBs to 37xxMBs. I beginning to understand with A64s
already low memory latency there some adjustments is memory timings to be
made, otherwise just boosting CPU speed can, actually make your system
slower. I found one good explaination at
www.insaneteck.com/index.php?page=overclocka64 .
Once, I figure all this out I'll post my results, bearing in mind it'll take
awhile.

john

 

[Paul]

Thanks again, the above has been very helpful, the fog is lifting. You've
given me a good starting point to work from. I've been doing some research
and it seems it's all in the memory timings. I fiddled around with the CPU
frequencies and CPU and HTT multipliers and while the CPU speed increased
the memory bandwidth went through the floor. AIDA showed memory read
dropping from 57xxMBs to 37xxMBs. I beginning to understand with A64s
already low memory latency there some adjustments is memory timings to be
made, otherwise just boosting CPU speed can, actually make your system
slower. I found one good explaination at
www.insaneteck.com/index.php?page=overclocka64 .
Once, I figure all this out I'll post my results, bearing in mind it'll take
awhile.

john

And that article leads to the spreadsheet from Oskar

http://xtremesystems.org/forums/showthread.php?t=41595&highlight=divider

FSB * HT_Mult / Mem_Divider = memory_clock
(I think that is how it works.)

It could be that the BIOS is botching the value for the
Mem_Divider. The HT_Mult is fractional integer (9.0, 9.5, 10.0)
and the Mem_Divider might be whole integer (9, 10). The
BIOS is supposed to pick the closest Mem_Divider to get
the memory clock speed the user desires.

I think it is pretty important to use your Windows utilities,
to see what is going on. For example, I've read that the BIOS
does not necessarily listen to the user's LDT multiplier setting,
so depending on BIOS version, you may not have too much control
over that.

To get that drastic a drop in memory bandwidth, you would almost
need command rate to change from 1T to 2T, plus have a bad
Mem_Divider choice. The 37xx number is not low enough to suggest
you've switched to single channel mode (64 bit mode) - in that
case you would also note that the amount of memory available
in Windows would be cut in half.

Paul

 

[name]

And that article leads to the spreadsheet from Oskar

http://xtremesystems.org/forums/showthread.php?t=41595&highlight=divider

FSB * HT_Mult / Mem_Divider = memory_clock
(I think that is how it works.)

According to,
http://forums.extremeoverclocking.com/showthread.php?t=151373, it's
(CPU speed) = (HTT/FSB) x (CPU multiplier),
(HTT) x (HTT multiplier) = (HT/FSB),
and (HTT frequency) x (CPU multiplier) / (memory divider) = memory speed.

I'm using HT and FSB interchageably and I assume you're using LDT and HT
multiplier interchangeable.
Please confirm.

It could be that the BIOS is botching the value for the
Mem_Divider. The HT_Mult is fractional integer (9.0, 9.5, 10.0)
and the Mem_Divider might be whole integer (9, 10). The
BIOS is supposed to pick the closest Mem_Divider to get
the memory clock speed the user desires.

I think it is pretty important to use your Windows utilities,
to see what is going on. For example, I've read that the BIOS
does not necessarily listen to the user's LDT multiplier setting,
so depending on BIOS version, you may not have too much control
over that.

To get that drastic a drop in memory bandwidth, you would almost
need command rate to change from 1T to 2T, plus have a bad
Mem_Divider choice. The 37xx number is not low enough to suggest
you've switched to single channel mode (64 bit mode) - in that
case you would also note that the amount of memory available
in Windows would be cut in half.

I'll get back to you on this. I've set everything to "Auto" in the BIOS,
taken readings with Aida32, Sandra 2002 and Everest, plus I'm using A64
memory helper, A64 Tweaker, CG-NVNF4, and cpuz.
Now, I'll start tweaking and see what happens.