'Ivan' wrote:
| any tangible benefit from upgrading to 1066 from 800 Mhz DDR 2?
_____
Depends on what speed you run it at, doesn't it? And the latencies. If you
give specific examples, then an answer could be specific. Otherwise, the
answer is ... 42.
DDR2-1066 has no standard specifications. The latencies are likely to be
higher than for DDR2-800 and the memory voltage is likely to be higher than
standard also. The standard settings for DDR2 memory with an Intel CPU uses
a CPU clock : Memory clock ratio of 2:1, so DDR2-800 memory would run in
spec with a FSB of 1600 MHz. Using a CPU clock : Memory clock ratio of 1:1
give some memory bandwidth increase, but then run in spec, DDR2-1066 memory
would require a 1066 MHz FSB.
For example, I use Patriot DDR2-1066 memory, memory voltage at 2.30 v, on an
EVGA 680i SLI motherboard with an E4300 @ 2.7 GHz. The FSB is 1200 MHz and
the memory clock is set at 1200 MHz so the CPU clock : Memory clock ratio is
1:1. The latencies are set at 5-5-4-9. This gives slightly better memory
bandwidth performance than a 2:1 CPU clock : Memory clock ratio of 2:1
giving a memory clock of 600 MHz and with latencies set at 4-3-3-8.
The large the L2 cache, the less performance increase given by increased
memory bandwidth. The performance difference likewise depends on the
application mix.
Different motherboard manufacturers use different terminology for the memory
speed settings. There have been several threads in this newsgroup since
March 2007 thrashing out what it all means.
If your question is "Should I replace my DDR2-800 memory with DDR2-1066
memory?" then my answer would be no, not unless you need a higher FSB speed
than you can get with the DDR-800 memory OR if a probable 2% performance
increase is worth the money to you.
Phil Weldon
"Ivan" <yeke@yahoo.com> wrote in message news:fien87$smq$1@ss408.t-com.hr...
| any tangible benefit from upgrading to 1066 from 800 Mhz DDR 2?
|
|
TTR has another thread with very helpful tables reducing memory
frequency and CAS latencies to effective latencies. I don't have that
one bookmarked but it's in one of thecableguy's overclocking threads
nb., I'm not up enough on this to say whether the stuff there is
correct or useful for realworld apps
'Howard Goldstein' wrote:
| On Mon, 26 Nov 2007 16:03:29 +0100, Ivan <yeke@yahoo.com> wrote:
| : any tangible benefit from upgrading to 1066 from 800 Mhz DDR 2?
| :
| :
|
| Here's a possibly relevant thread
| http://www.thetechrepository.com/sho...?p=588#post588
|
| TTR has another thread with very helpful tables reducing memory
| frequency and CAS latencies to effective latencies. I don't have that
| one bookmarked but it's in one of thecableguy's overclocking threads
|
| nb., I'm not up enough on this to say whether the stuff there is
| correct or useful for realworld apps
_____
The charts and calculations at the URL you cite miss an important point.
CAS latency is only imposed on the FIRST set of bytes fetched in a series of
contiguous bytes. Dual channel memory will set up two sets of 8 bytes (one
from each channel) with one CAS delay. Data is transferred between the L2
cache and main memory as 64 byte line (the smallest contiguous memory space
in the L2 cache, the CAS latency is only incurred ONCE per 64 bytes of
memory, NOT once per 8 bytes of memory as the information at the URL you
cited implies.
I think I am correct in inferring that the smallest fetch from main memory
is the size of a line in the L1/L2 caches, which is 64 bytes.
As always, when raw performance meets actual application, the outcome is
greatly affected by the instruction mix, programming techniques, and data
set size.
Phil Weldon
"Howard Goldstein" <hgoldste@mpcs.com> wrote in message
news:1196109042.85865@news.queue.to...
| On Mon, 26 Nov 2007 16:03:29 +0100, Ivan <yeke@yahoo.com> wrote:
| : any tangible benefit from upgrading to 1066 from 800 Mhz DDR 2?
| :
| :
|
| Here's a possibly relevant thread
| http://www.thetechrepository.com/sho...?p=588#post588
|
| TTR has another thread with very helpful tables reducing memory
| frequency and CAS latencies to effective latencies. I don't have that
| one bookmarked but it's in one of thecableguy's overclocking threads
|
| nb., I'm not up enough on this to say whether the stuff there is
| correct or useful for realworld apps
"Phil Weldon" <not.disclosed@example.com> wrote in message
news:13kmes67pgqak1e@corp.supernews.com...
> 'Howard Goldstein' wrote:
> | On Mon, 26 Nov 2007 16:03:29 +0100, Ivan <yeke@yahoo.com> wrote:
> | : any tangible benefit from upgrading to 1066 from 800 Mhz DDR 2?
> | :
> | :
> |
> | Here's a possibly relevant thread
> | http://www.thetechrepository.com/sho...?p=588#post588
> |
> | TTR has another thread with very helpful tables reducing memory
> | frequency and CAS latencies to effective latencies. I don't have that
> | one bookmarked but it's in one of thecableguy's overclocking threads
> |
> | nb., I'm not up enough on this to say whether the stuff there is
> | correct or useful for realworld apps
> _____
>
> The charts and calculations at the URL you cite miss an important point.
> CAS latency is only imposed on the FIRST set of bytes fetched in a series
> of
> contiguous bytes. Dual channel memory will set up two sets of 8 bytes
> (one
> from each channel) with one CAS delay. Data is transferred between the L2
> cache and main memory as 64 byte line (the smallest contiguous memory
> space
> in the L2 cache, the CAS latency is only incurred ONCE per 64 bytes of
> memory, NOT once per 8 bytes of memory as the information at the URL you
> cited implies.
>
> See
> http://www.lenovo.hu/kszf/adatlap/Pr...re2_Mobile.pdf
> for information on the Core 2 Duo L2 cache and
> http://etd.lsu.edu/docs/available/et...ash_thesis.pdf
> for an analysis of performance for the Core 2 Duo.
>
> I think I am correct in inferring that the smallest fetch from main memory
> is the size of a line in the L1/L2 caches, which is 64 bytes.
>
> As always, when raw performance meets actual application, the outcome is
> greatly affected by the instruction mix, programming techniques, and data
> set size.
>
> Phil Weldon
>
> "Howard Goldstein" <hgoldste@mpcs.com> wrote in message
> news:1196109042.85865@news.queue.to...
> | On Mon, 26 Nov 2007 16:03:29 +0100, Ivan <yeke@yahoo.com> wrote:
> | : any tangible benefit from upgrading to 1066 from 800 Mhz DDR 2?
> | :
> | :
> |
> | Here's a possibly relevant thread
> | http://www.thetechrepository.com/sho...?p=588#post588
> |
> | TTR has another thread with very helpful tables reducing memory
> | frequency and CAS latencies to effective latencies. I don't have that
> | one bookmarked but it's in one of thecableguy's overclocking threads
> |
> | nb., I'm not up enough on this to say whether the stuff there is
> | correct or useful for realworld apps
>
>
Actually I'm considering upgrading from GEIL Black Dragon 800 Mhz to the
same make and model running on 1066 Mhz, but the price difference is about
200 $, which is nothing to sneeze at.
But the performance gain (if any) doesn't seem to justify the expense.
"According to The Hitchhiker's Guide to the Galaxy, a race of vast
pan-dimensional hyper-intelligent beings constructed the second greatest
computer in all of time and space, Deep Thought, to calculate The Ultimate
Answer to The Great Question of Life, the Universe, and Everything.
Distracted by a demarcation dispute with two philosophers, a "simple
answer"[1] is requested. After seven and a half million years of computing
cycles, Deep Thought's answer is: forty two.
“"I think the problem is that the question was too broadly based..."[1]
"Forty two?!" yelled Loonquawl. "Is that all you've got to show for seven
and a half million years' work?"
"I checked it very thoroughly," said the computer, "and that quite
definitely is the answer. I think the problem, to be quite honest with you,
is that you've never actually known what the question is."[2]"
Quoted from the above URL.
Phil Weldon
"Fishface" <invalid@ddress.ok?> wrote in message
news:XFH2j.7046$Mr.6242@trnddc04...
| Phil Weldon wrote:
|
| >Otherwise, the answer is ... 42.
|
| http://en.wikipedia.org/wiki/42_(number)
|
'Ivan' wrote:
| Actually I'm considering upgrading from GEIL Black Dragon 800 Mhz to the
| same make and model running on 1066 Mhz, but the price difference is about
| 200 $, which is nothing to sneeze at.
|
| But the performance gain (if any) doesn't seem to justify the expense.
_____
My vote would be to save the money for more meaningful upgrades later.
I think I posted earlier this year the results of SiSoft Sandra XI Lite SP1a
benchmarks with my Memory clock set at various speeds from 600 MHz to 1200
MHz.
Yes, this data and the meaning of DDR2-1066 and CPU clock : Memory clock
ratios is discussed in two earlier threads
#2. Subject line: What does PC1066 mean, and what advantage does a 1:1
ratio confer?
From: Phil Weldon
Sent: 4/23/2007 4:22 PM
I got confused all over again just rereading the posts B^) What isn't
discussed in the two threads is the effect of latency and effect of the L1
and L2 caches. They were mainly an attempt to find a common understanding
of the murky nomenclature that existed early in 2007.
Phil Weldon
"Ivan" <yeke@yahoo.com> wrote in message news:fifh2a$75j$1@ss408.t-com.hr...
|
| "Phil Weldon" <not.disclosed@example.com> wrote in message
| news:13kmes67pgqak1e@corp.supernews.com...
| > 'Howard Goldstein' wrote:
| > | On Mon, 26 Nov 2007 16:03:29 +0100, Ivan <yeke@yahoo.com> wrote:
| > | : any tangible benefit from upgrading to 1066 from 800 Mhz DDR 2?
| > | :
| > | :
| > |
| > | Here's a possibly relevant thread
| > | http://www.thetechrepository.com/sho...?p=588#post588
| > |
| > | TTR has another thread with very helpful tables reducing memory
| > | frequency and CAS latencies to effective latencies. I don't have that
| > | one bookmarked but it's in one of thecableguy's overclocking threads
| > |
| > | nb., I'm not up enough on this to say whether the stuff there is
| > | correct or useful for realworld apps
| > _____
| >
| > The charts and calculations at the URL you cite miss an important point.
| > CAS latency is only imposed on the FIRST set of bytes fetched in a
series
| > of
| > contiguous bytes. Dual channel memory will set up two sets of 8 bytes
| > (one
| > from each channel) with one CAS delay. Data is transferred between the
L2
| > cache and main memory as 64 byte line (the smallest contiguous memory
| > space
| > in the L2 cache, the CAS latency is only incurred ONCE per 64 bytes of
| > memory, NOT once per 8 bytes of memory as the information at the URL you
| > cited implies.
| >
| > See
| > http://www.lenovo.hu/kszf/adatlap/Pr...re2_Mobile.pdf
| > for information on the Core 2 Duo L2 cache and
| > http://etd.lsu.edu/docs/available/et...ash_thesis.pdf
| > for an analysis of performance for the Core 2 Duo.
| >
| > I think I am correct in inferring that the smallest fetch from main
memory
| > is the size of a line in the L1/L2 caches, which is 64 bytes.
| >
| > As always, when raw performance meets actual application, the outcome is
| > greatly affected by the instruction mix, programming techniques, and
data
| > set size.
| >
| > Phil Weldon
| >
| > "Howard Goldstein" <hgoldste@mpcs.com> wrote in message
| > news:1196109042.85865@news.queue.to...
| > | On Mon, 26 Nov 2007 16:03:29 +0100, Ivan <yeke@yahoo.com> wrote:
| > | : any tangible benefit from upgrading to 1066 from 800 Mhz DDR 2?
| > | :
| > | :
| > |
| > | Here's a possibly relevant thread
| > | http://www.thetechrepository.com/sho...?p=588#post588
| > |
| > | TTR has another thread with very helpful tables reducing memory
| > | frequency and CAS latencies to effective latencies. I don't have that
| > | one bookmarked but it's in one of thecableguy's overclocking threads
| > |
| > | nb., I'm not up enough on this to say whether the stuff there is
| > | correct or useful for realworld apps
| >
| >
|
| Actually I'm considering upgrading from GEIL Black Dragon 800 Mhz to the
| same make and model running on 1066 Mhz, but the price difference is about
| 200 $, which is nothing to sneeze at.
|
| But the performance gain (if any) doesn't seem to justify the expense.
|
|
|
On Mon, 26 Nov 2007 23:25:28 -0500, Phil Weldon <not.disclosed@example.com> wrote:
: 'Ivan' wrote:
: | Actually I'm considering upgrading from GEIL Black Dragon 800 Mhz to the
: | same make and model running on 1066 Mhz, but the price difference is about
: | 200 $, which is nothing to sneeze at.
: |
: | But the performance gain (if any) doesn't seem to justify the expense.
: _____
:
: My vote would be to save the money for more meaningful upgrades later.
:
: I think I posted earlier this year the results of SiSoft Sandra XI Lite SP1a
: benchmarks with my Memory clock set at various speeds from 600 MHz to 1200
: MHz.
:
: Yes, this data and the meaning of DDR2-1066 and CPU clock : Memory clock
: ratios is discussed in two earlier threads
:
: #1. Subject line: E4300 / 680i overclocking
: From: Phil Weldon
: Sent: 4/16/2007 9:38 PM
(teh google was my friend) I now need to spend some time with these,
but the salt mine will soon be calling so it'll be this afternoon if I
can't sneak a peak at work.
Like Ivan, I've been pondering whether to drop $80 on 1066 sticks for
the Q6600 but just skimming over those 2 posts and your earlier
remarks it looks like a big waste of money, particularly on this
motherboard with its its endlessly programmable dividers and no real
memory holdbacks for FSB play.
DDR 2 800 vs 1066 
Linear Mode
