Overclocking E6600 on Asus P5W DH Deluxe

"Thomas" <> wrote in message
news:4629cc10$0$29567$.. .
> Aren't you forgetting we're running dual-channel memory here? That means
> that the total data bandwidth of the memory is multiplied by two. So using
> 2 bars of PC533 memory gives you PC1066 performance, enough to keep up
> with the 1066 MHz FSB of the E6 series. I actually can remember reading,
> one day after I finalized my order of 800 MHz DDR2, that for the E4300,
> 800 MHz DDR is nonsense, because to reach the 800 MHz bandwidth, you need
> to have an OC of 100%, which is achieved only rarely and then with exotic
> cooling.
>
> I mean, if 1066 memory would be required to get the full performance out
> of stock clocked CPU's, I think that this would be the standard
> combination sold, but all standard PC's are sold with 533...
>
> Gosh this memory discussion is incredible, how difficult can they make it
> :-)
>
> --
> Met vriendelijke groeten, Thomas vd Horst.
>
Thomas,
Using dual channel memory does not multiply the bandwidth by a factor of 2.
It increases the bandwidth, but for a totally different reason: the
so-called memory latency. This latency comes from the fact that to read one
piece of data, the data goes through several intermediate internal buffers
and also is due to the fact that the address has to be fetched column
address first and then row address. And there must be some idle time in
between. Also, after reading a piece of data from a mem chip, one must wait
a so called bank cycle time before accessing the next piece of data. These
latencies are due to the fact that certain condensator have to be
recharged/discharged between operations. Remember, the mem is Dynamic Random
Acces Memory, which means everything works with condensators inside the mem
chips. Because of this, you cannot access memory at each clock pulse, you
have to wait a certain number of clock pulses between reads/writes on the
same stick.
Reading one piece of data from one mem stick and then the next piece of data
from another stick, you increase the bandwidth but not by a factor of two.
It is a fairly complicated concept, I know.
If you want to undestand why you only need lower speed memory than Phil
claims, read my last answer to him. The dual channel stuff does not explain
why I pretend that he is wrong by a factor of two in his calculations.
Anyway, I am running my E6600 ovelclocked at 3.0 GHz with a pair of 1 GB
DDR2 667 and all goes well at stock voltages at full load (100% load on both
cores). Plus both my BIOS and CPU-Z from CPUID tell me that I am running at
1:1 CPU/Mem ratio.
Now, this being said, the effective bandwidth will increase if you use
faster than necessary mem sticks, because if you underclock them, usually
the latency at lower than sock speed decreases. Example: using DDR2 533 at
stock speed, the timing latencies will be something like 5-5-5-15 (numbers
are clock pulses to wait in between internal mem circuitry operations to
access the proper mem cell). Using DDR2 1066 as if it were DDR2 533, the
latencies will be something like 3-3-3-12 or even 2-2-2-9. So, as you can
see, you gain quite a number of clock pulses to access a mem cell when using
fast memory underclocked. Using faster memory at their stock speed with a
1:1 ratio however would imply overclocking the CPU far above its
capabilities, unless you use some sort of liquid nitrogen cooling system
(with DDR2 1066, you would run the E6600 at close to 5 GHz).

Thomas,
If you want a better explanation about latencies etc., please read this
link:

http://www.lostcircuits.com/memory/ddrii/

"Michel R. Carleer" <> wrote in message
news:462a2e38$0$13855$...
> Thomas,
> Using dual channel memory does not multiply the bandwidth by a factor of
> 2.
> It increases the bandwidth, but for a totally different reason: the
> so-called memory latency. This latency comes from the fact that to read
> one piece of data, the data goes through several intermediate internal
> buffers and also is due to the fact that the address has to be fetched
> column address first and then row address. And there must be some idle
> time in between. Also, after reading a piece of data from a mem chip, one
> must wait a so called bank cycle time before accessing the next piece of
> data. These latencies are due to the fact that certain condensator have to
> be recharged/discharged between operations. Remember, the mem is Dynamic
> Random Acces Memory, which means everything works with condensators inside
> the mem chips. Because of this, you cannot access memory at each clock
> pulse, you have to wait a certain number of clock pulses between
> reads/writes on the same stick.
> Reading one piece of data from one mem stick and then the next piece of
> data from another stick, you increase the bandwidth but not by a factor of
> two.
> It is a fairly complicated concept, I know.
> If you want to undestand why you only need lower speed memory than Phil
> claims, read my last answer to him. The dual channel stuff does not
> explain why I pretend that he is wrong by a factor of two in his
> calculations.
> Anyway, I am running my E6600 ovelclocked at 3.0 GHz with a pair of 1 GB
> DDR2 667 and all goes well at stock voltages at full load (100% load on
> both cores). Plus both my BIOS and CPU-Z from CPUID tell me that I am
> running at 1:1 CPU/Mem ratio.
> Now, this being said, the effective bandwidth will increase if you use
> faster than necessary mem sticks, because if you underclock them, usually
> the latency at lower than sock speed decreases. Example: using DDR2 533 at
> stock speed, the timing latencies will be something like 5-5-5-15 (numbers
> are clock pulses to wait in between internal mem circuitry operations to
> access the proper mem cell). Using DDR2 1066 as if it were DDR2 533, the
> latencies will be something like 3-3-3-12 or even 2-2-2-9. So, as you can
> see, you gain quite a number of clock pulses to access a mem cell when
> using fast memory underclocked. Using faster memory at their stock speed
> with a 1:1 ratio however would imply overclocking the CPU far above its
> capabilities, unless you use some sort of liquid nitrogen cooling system
> (with DDR2 1066, you would run the E6600 at close to 5 GHz).
>

"Phil Weldon" <> wrote in message
newsPhWh.3640$ ink.net...
> Beg to differ. The CPU stock clock speed for the E6600 is 266 MHz giving
> a
> an FSB of 1066 MHz. To use a 1:1 CPU clock : memory clock ratio REQUIRES
> PC1066 (PC8500) DDR2 memory. The P1066 rating is based on a 4 X 266 MHz =
> 1066 MHz FSB. The advantage of DDR2 is NOT that it runs at a double clock
> rate compared to DDR memory; THAT advantage is nullified by the INTERNAL
> to
> the memory chip division of the clock rate by two. The advantage is that
> DDR2 can run at higher EXTERNAL clock speeds (and at lower power.) Thus,
> to
> operate with the EXTERNAL memory clock at 266 MHz requires DDR2 memory
> that
> has DDR2 PC1066 performance. When overclocking an E6600, the FSB must be
> higher than 1066 MHz (266 MHz clock rate). In this case, to use a 1:1 CPU
> clock : memory clock ratio requires DDR2 memory that has greater than
> PC1066
> performance. DDR2 PC533 performance memory requires a 2:1 CPU clock :
> memory clock ratio.
>
> Phil Weldon

Another link to convince you, and this one is from Crucial:

http://www.crucial.com/library/memory_speeds.asp

Apparently your are confusing the FSB clock frequency (266 MHz in your
example) with the FSB Data Transfer rate (1066 MT/s, but it is not MHz here
as you can see)

'Michel R. Carleer' wrote:
| Another link to convince you, and this one is from Crucial:
|
| http://www.crucial.com/library/memory_speeds.asp
|
| Apparently your are confusing the FSB clock frequency (266 MHz in your
| example) with the FSB Data Transfer rate (1066 MT/s, but it is not MHz
here
| as you can see)
_____

Good discussion! And my counter cite is
http://www.tomshardware.com/2005/07/...max/index.html
(see
"To attain clock rates that support DDR2-1066 without overclocking, we also
used the 1:1 memory to FSB clock ratio on a system with the FSB clock set to
266 MHz. One of the few CPUs designed to handle this kind of abuse is the
3.73 GHz Pentium 4 Extreme Edition." on page 2.)

At present I run my E4300 / EVGA 680i system with a 1200 MHz FSB (300 MHz
clock) and my PC1066 memory also with a 1200 MHz bus and a 300 MHz clock
giving a CPU clock : memory clock ratio of 1:1. It may be that the
differing nomenclature among the various BIOS adds to the confusion.

I plan to study the question further and post again. I would really like to
have a definitive understanding.

Phil Weldon

"Michel R. Carleer" <> wrote in message
news:462a6cde$0$13865$...
|
| "Phil Weldon" <> wrote in message
| newsPhWh.3640$ ink.net...
| > Beg to differ. The CPU stock clock speed for the E6600 is 266 MHz
giving
| > a
| > an FSB of 1066 MHz. To use a 1:1 CPU clock : memory clock ratio
REQUIRES
| > PC1066 (PC8500) DDR2 memory. The P1066 rating is based on a 4 X 266 MHz
=
| > 1066 MHz FSB. The advantage of DDR2 is NOT that it runs at a double
clock
| > rate compared to DDR memory; THAT advantage is nullified by the INTERNAL
| > to
| > the memory chip division of the clock rate by two. The advantage is
that
| > DDR2 can run at higher EXTERNAL clock speeds (and at lower power.)
Thus,
| > to
| > operate with the EXTERNAL memory clock at 266 MHz requires DDR2 memory
| > that
| > has DDR2 PC1066 performance. When overclocking an E6600, the FSB must
be
| > higher than 1066 MHz (266 MHz clock rate). In this case, to use a 1:1
CPU
| > clock : memory clock ratio requires DDR2 memory that has greater than
| > PC1066
| > performance. DDR2 PC533 performance memory requires a 2:1 CPU clock :
| > memory clock ratio.
| >
| > Phil Weldon
|
| Another link to convince you, and this one is from Crucial:
|
| http://www.crucial.com/library/memory_speeds.asp
|
| Apparently your are confusing the FSB clock frequency (266 MHz in your
| example) with the FSB Data Transfer rate (1066 MT/s, but it is not MHz
here
| as you can see)
|
|

"Phil Weldon" <> wrote in message
news:O2wWh.7182$ ink.net...
> Good discussion! And my counter cite is
> http://www.tomshardware.com/2005/07/...max/index.html
> (see
> "To attain clock rates that support DDR2-1066 without overclocking, we
> also
> used the 1:1 memory to FSB clock ratio on a system with the FSB clock set
> to
> 266 MHz. One of the few CPUs designed to handle this kind of abuse is the
> 3.73 GHz Pentium 4 Extreme Edition." on page 2.)
>
> At present I run my E4300 / EVGA 680i system with a 1200 MHz FSB (300 MHz
> clock) and my PC1066 memory also with a 1200 MHz bus and a 300 MHz clock
> giving a CPU clock : memory clock ratio of 1:1. It may be that the
> differing nomenclature among the various BIOS adds to the confusion.
>
> I plan to study the question further and post again. I would really like
> to
> have a definitive understanding.
>
> Phil Weldon

Yes indeed this is a good discussion, as many confusing infos appear on the
net.
For example, a FSB clocked at 266 MHz is often wrongly mentionned as FSB
1066 MHz. In fact it should be written FSB 1066 MT/s, to reflect the fact
that the FSB transmits four data chunks per clock period.
As for your setup, in fact you agree with me: 300 MHz CPU clock (or FSB
clock) with a 300 MHz memory I/O bus clock. In other words, you are
undeclocking your memory: a 300 MHz I/O bus clock means DDR2 600 or PC4800.
What you call PC1066 is in fact a DDR2 1066 (or PC8500), and to run it full
speed, you should use a 533 MHz I/O bus clock.
Or leave it as is, but then I am almost sure you can lower the latency
timings of your sticks in the BIOS, because you underclock it by almost a
factor of 1.8.
Anyway, I tend to be more confident in the info given on the Crucial web
site than in the info from Tom's Hardware. After all, Crucial makes and
sells memory modules for quite a long time and they surely know better than
Tom, or myself by the way. And they would never tell the people to buy less
expensive modules if it would not work, I guess. Once again, I run my E6600
at 333 MHz clock speed, coupled to DDR2 667 memory (333 MHz I/O bus clock)
and it works without a problem, and all this at stock voltage. If my DDR2
667 (low cost) was a factor of 2 too slow, I am pretty sure it would not
work at all.

Michka

'Michka' wrote:
| Yes indeed this is a good discussion, as many confusing infos appear on
the
| net.
| For example, a FSB clocked at 266 MHz is often wrongly mentionned as FSB
| 1066 MHz. In fact it should be written FSB 1066 MT/s, to reflect the fact
| that the FSB transmits four data chunks per clock period.
| As for your setup, in fact you agree with me: 300 MHz CPU clock (or FSB
| clock) with a 300 MHz memory I/O bus clock. In other words, you are
| undeclocking your memory: a 300 MHz I/O bus clock means DDR2 600 or
PC4800.
| What you call PC1066 is in fact a DDR2 1066 (or PC8500), and to run it
full
| speed, you should use a 533 MHz I/O bus clock.
| Or leave it as is, but then I am almost sure you can lower the latency
| timings of your sticks in the BIOS, because you underclock it by almost a
| factor of 1.8.
| Anyway, I tend to be more confident in the info given on the Crucial web
| site than in the info from Tom's Hardware. After all, Crucial makes and
| sells memory modules for quite a long time and they surely know better
than
| Tom, or myself by the way. And they would never tell the people to buy
less
| expensive modules if it would not work, I guess. Once again, I run my
E6600
| at 333 MHz clock speed, coupled to DDR2 667 memory (333 MHz I/O bus clock)
| and it works without a problem, and all this at stock voltage. If my DDR2
| 667 (low cost) was a factor of 2 too slow, I am pretty sure it would not
| work at all.
_____

The DDR2 rating of PC1066 indicates a clock of 266 MHz. With 1:1 CPU clock
: memory clock ratio, a FSB speed of 1066 MHz requires PC1066 performance
memory. The DDR2 rating is FOUR X the memory clock. The nomenclature is
confusing, but the fact remains that PC1066 performance memory is necessary
for FSB 1066 MHz operation at a 1:1 CPU clock : memory clock ratio.

Phil Weldon

"Michel R. Carleer" <> wrote in message
news:462a9838$0$14248$...
|
| "Phil Weldon" <> wrote in message
| news:O2wWh.7182$ ink.net...
| > Good discussion! And my counter cite is
| >
http://www.tomshardware.com/2005/07/...max/index.html
| > (see
| > "To attain clock rates that support DDR2-1066 without overclocking, we
| > also
| > used the 1:1 memory to FSB clock ratio on a system with the FSB clock
set
| > to
| > 266 MHz. One of the few CPUs designed to handle this kind of abuse is
the
| > 3.73 GHz Pentium 4 Extreme Edition." on page 2.)
| >
| > At present I run my E4300 / EVGA 680i system with a 1200 MHz FSB (300
MHz
| > clock) and my PC1066 memory also with a 1200 MHz bus and a 300 MHz clock
| > giving a CPU clock : memory clock ratio of 1:1. It may be that the
| > differing nomenclature among the various BIOS adds to the confusion.
| >
| > I plan to study the question further and post again. I would really
like
| > to
| > have a definitive understanding.
| >
| > Phil Weldon
|
| Yes indeed this is a good discussion, as many confusing infos appear on
the
| net.
| For example, a FSB clocked at 266 MHz is often wrongly mentionned as FSB
| 1066 MHz. In fact it should be written FSB 1066 MT/s, to reflect the fact
| that the FSB transmits four data chunks per clock period.
| As for your setup, in fact you agree with me: 300 MHz CPU clock (or FSB
| clock) with a 300 MHz memory I/O bus clock. In other words, you are
| undeclocking your memory: a 300 MHz I/O bus clock means DDR2 600 or
PC4800.
| What you call PC1066 is in fact a DDR2 1066 (or PC8500), and to run it
full
| speed, you should use a 533 MHz I/O bus clock.
| Or leave it as is, but then I am almost sure you can lower the latency
| timings of your sticks in the BIOS, because you underclock it by almost a
| factor of 1.8.
| Anyway, I tend to be more confident in the info given on the Crucial web
| site than in the info from Tom's Hardware. After all, Crucial makes and
| sells memory modules for quite a long time and they surely know better
than
| Tom, or myself by the way. And they would never tell the people to buy
less
| expensive modules if it would not work, I guess. Once again, I run my
E6600
| at 333 MHz clock speed, coupled to DDR2 667 memory (333 MHz I/O bus clock)
| and it works without a problem, and all this at stock voltage. If my DDR2
| 667 (low cost) was a factor of 2 too slow, I am pretty sure it would not
| work at all.
|
| Michka
|
|

Phil Weldon wrote:
> 'Michka' wrote:
> > Once again, I run my E6600 at 333 MHz
> > clock speed, coupled to DDR2 667 memory (333 MHz I/O bus clock) and
> > it works without a problem, and all this at stock voltage. If my
> > DDR2 667 (low cost) was a factor of 2 too slow, I am pretty sure it
> > would not work at all.
> _____
>
> The DDR2 rating of PC1066 indicates a clock of 266 MHz. With 1:1 CPU
> clock
> > memory clock ratio, a FSB speed of 1066 MHz requires PC1066
> > performance
> memory. The DDR2 rating is FOUR X the memory clock. The
> nomenclature is confusing, but the fact remains that PC1066
> performance memory is necessary for FSB 1066 MHz operation at a 1:1
> CPU clock : memory clock ratio.

Phil, there's a very simple way of finding out which thesis is valid; See
what the impact on memory performance is by running some benchmarks with yr
memory at 1200 and at 600 'MHz'. This could clarify alot. I'd like to do it
myself, but my stuff won't arive for some time... Videocard isnt deliverable
until May 1st :'(

--
Met vriendelijke groeten, Thomas vd Horst.

"Phil Weldon" <> wrote in message
news:1yEWh.24145$ link.net...
> The DDR2 rating of PC1066 indicates a clock of 266 MHz. With 1:1 CPU
> clock
> : memory clock ratio, a FSB speed of 1066 MHz requires PC1066 performance
> memory. The DDR2 rating is FOUR X the memory clock. The nomenclature is
> confusing, but the fact remains that PC1066 performance memory is
> necessary
> for FSB 1066 MHz operation at a 1:1 CPU clock : memory clock ratio.
>
> Phil Weldon

"Phil Weldon" <> wrote in message
news:1yEWh.24145$ link.net...
> The DDR2 rating of PC1066 indicates a clock of 266 MHz.

Yes but that is the internal mem clock. The 1:1 ratio takes the ratio
between the CPU input clock (which is at the same time the CPU I/O clock)
and the mem I/O clock. At least that's the definition from the mem
manufacturers and many others. And they recommend the best match between CPU
and mem, based on that definition. I have a feeling that the mem
manufacturers would be stupid to tell on their web sites that a 2 times
slower mem is enough for a given CPU clock speed, given the price difference
you have to pay for the higher speed memory.

> With 1:1 CPU clock
> : memory clock ratio, a FSB speed of 1066 MHz requires PC1066 performance
> memory. The DDR2 rating is FOUR X the memory clock. The nomenclature is
> confusing, but the fact remains that PC1066 performance memory is
> necessary
> for FSB 1066 MHz operation at a 1:1 CPU clock : memory clock ratio.

And the mem I/O clock is twice the internal mem clock. So, what you are
doing is connect a CPU I/O 266 MHz clock to a mem I/O clock of 533 MHz. It
is not the mem internal clock that you have to consider, it is the mem I/O
clock, which is twice the internal mem clock. You see, from what I have
read, the 1:1 ratio is NOT between the CPU I/O clock and the internal mem
clock, but between the CPU I/O clock and the mem I/O clock.

>
> Phil Weldon
>

So, we simply do not have the same definition of what a CPU/mem clock 1:1
ratio is. And it would indeed be nice to see if there is any improvement in
overall system performance using your definition instead of mine. Of course,
I do not take into account that when using a higher speed mem, but
underclocked, the latency timings can be lowered by some significant value.
Unfortunately I cannot test it, as I do not have DDR2 1066 mem at my
disposal. Maybe you can?

By the way, from what I understand: it is not PC1066, it is either DDR2 1066
(the chip speed in megabytes/sec) or PC2-8500 (the mem DIMM module speed in
megabytes/sec, as the module bus width is 8 bytes large).
Confusion, confusion... Arising from people on the net using different names
for the same thing.

'Thomas' wrote:
| Phil, there's a very simple way of finding out which thesis is valid; See
| what the impact on memory performance is by running some benchmarks with
yr
| memory at 1200 and at 600 'MHz'. This could clarify alot. I'd like to do
it
| myself, but my stuff won't arive for some time... Videocard isnt
deliverable
| until May 1st :'(
_____


Exactly.

For my system; E4300 / PC1066 (PC8500) memory / nVidia 690i motherboard

Stock; E4300 @ 1.8 GHz / FSB 800 MHz / Memory 800 MHz / 200 MHz Clock / 1:1
CPU clock : memory clock ratio

SiSoft Sandra 2007 Memory Bandwidth: 4252 MBytes/sec.

Overclocked; E4300 @ 2.7 GHz / FSB 1200 MHz / Memory 1200 MHz / 300 Clock /
1:1 CPU clock : memory clock ratio

SiSoft Sandra 2007 Memory Bandwidth: 6307 MBytes/sec.

Most telling is that the system WILL NOT BOOT at the settings
FSB 1200 MHz
Memory 1400 MHz
while it WILL boot at the settings
FSB 1200 MHz
Memory 1200 MHz
and
FSB 1200 MHz
Memory 1300 MHz.

I am fairly certain I can a higher CPU overclock with improved cooling.
With a CPU clock of 350 (1400 MHz FSB) and an increased core voltage, the
system can load Windows XP, but the CPU temperature is in the high 50s C
even at very low CPU usage.

Phil Weldon

"Thomas" <> wrote in message
news:<462b18cb$0$11743$> ...
| Phil Weldon wrote:
| > 'Michka' wrote:
| > > Once again, I run my E6600 at 333 MHz
| > > clock speed, coupled to DDR2 667 memory (333 MHz I/O bus clock) and
| > > it works without a problem, and all this at stock voltage. If my
| > > DDR2 667 (low cost) was a factor of 2 too slow, I am pretty sure it
| > > would not work at all.
| > _____
| >
| > The DDR2 rating of PC1066 indicates a clock of 266 MHz. With 1:1 CPU
| > clock
| > > memory clock ratio, a FSB speed of 1066 MHz requires PC1066
| > > performance
| > memory. The DDR2 rating is FOUR X the memory clock. The
| > nomenclature is confusing, but the fact remains that PC1066
| > performance memory is necessary for FSB 1066 MHz operation at a 1:1
| > CPU clock : memory clock ratio.
|
| Phil, there's a very simple way of finding out which thesis is valid; See
| what the impact on memory performance is by running some benchmarks with
yr
| memory at 1200 and at 600 'MHz'. This could clarify alot. I'd like to do
it
| myself, but my stuff won't arive for some time... Videocard isnt
deliverable
| until May 1st :'(
|
| --
| Met vriendelijke groeten, Thomas vd Horst.