"RobV" <robv@nowhere.invalid> wrote in message
news:47c9fa90$0$22824$4c368faf@roadrunner.com...
> bob johnson wrote:
>> OK
>>
>> Got my OC back. Now, how to push my memory carefully and w/
>> stability?
>
> Just a friendly notice here, Bob. alt.comp.hardware.overclocking is a
> text only news group. As such, if you would like those reading to see a
> picture of something, the correct thing to do is post it at a web site,
> then give the link to the web address.
>
> Posting pictures, or attachments, isn't too large a problem for those of
> us using Outlook Express, but many folks use a strict newsreader program
> that shows HTML, pictures, attachments and such as just a jumble of
> characters, letters and numbers.
>
> News groups that allow these things will have binary in the NG name.
>
> Now, carry on...;-)
>
> Today, you can forget this rule.Just to clarify, when
> you post 1:1.2 do you mean a ratio of 1 to 2 (1/2)?
No, I mean 1 : 1.2, i.e. the clock-rate of the DDR2-memory
(400MHz) is 1.2 times higher than the clock-rate of the
FSB (333MHz).
> Motherboards based on the nVidia 680i series chipset DO show
> increased memory through-put when operated at higher Memory
> clock : CPU clock ratios given tight enough memory parameter
> delays.
So it seems Intel did a better Job on that. I clocked my Core
II Extreme at a FSB of 400/1600MHz (core-clock 3.2GHz) so that
the memory-FSB-ratio is 1:1 and the memory-throughput was sligh-
tly lower than at 333MHz (equal memory-clocking of 400MHz aka
800MHz DDR).
And there's another thing to mention: latency is a more
important parameter than throughput in real-world-apps.
'Elcaro Nosille' wrote:
> No, I mean 1 : 1.2, i.e. the clock-rate of the DDR2-memory
> (400MHz) is 1.2 times higher than the clock-rate of the
> FSB (333MHz).
>
>> Motherboards based on the nVidia 680i series chipset DO show
>> increased memory through-put when operated at higher Memory
>> clock : CPU clock ratios given tight enough memory parameter
>> delays.
>
> So it seems Intel did a better Job on that. I clocked my Core
> II Extreme at a FSB of 400/1600MHz (core-clock 3.2GHz) so that
> the memory-FSB-ratio is 1:1 and the memory-throughput was sligh-
> tly lower than at 333MHz (equal memory-clocking of 400MHz aka
> 800MHz DDR).
>
> And there's another thing to mention: latency is a more
> important parameter than throughput in real-world-apps.
_____
Thanks for the clarification. The [,] [.] difference is sometimes
confusing. Especially when combined with differing nomenclature for the
ratio between CPU clock and memory clock (likely aggravated by meddling
sales department meddling in engineering department specifications and
performance data.
As for your
> And there's another thing to mention: latency is a more
> important parameter than throughput in real-world-apps
the larger the L2 cache and the greater the L2 cache associativity the less
important main memory latency AND through-put for 'real world' performance.
Given optimization for L2 cache size, main memory through-put AND latency
become less important for 'real-world' performance. For 'real world'
performance, memory though-put IS dependent on latency AND memory clock
speed. After all, we use computer SYSTEMS (see 'Amdahl's Rule of Thumb',
also known as 'Amdahl's Other Law': "one byte of memory and one byte of I/O
are required for each instruction per second supported by a computer"; part
of
"3. Amdahl's system balance rules
Gene Amdahl is famous for many rules of thumb.
For data engineering, there are four famous ones [6]:
10. Amdahl's parallelism law: If a computation has a
serial component S and a parallel component P, then
the maximum speedup is (S+P)/S.
11. Amdahl's balanced system law: A system needs a bit
of IO per second for each instruction per second:
about 8 MIPS per MBps.
12. Amdahl's memory law: ? ? 1: that is, in a balanced
system the MB/MIPS ratio, called alpha (? ), is 1.
13. Amdahl's IO law: Programs do one IO per 50,000
instructions." http://research.microsoft.com/~gray/...ngineering.pdf .
Also see
Petascale Computational Systems:
Balanced CyberInfrastructure in a Data-Centric World
Gordon Bell1, Jim Gray1 and Alex Szalay2
1. Microsoft Research
2. The Johns Hopkins University
"Elcaro Nosille" Elcaro.Nosille@arcor.de wrote in message
news:47cae5d9$0$25508$9b4e6d93@newsspool1.arcor-online.net...
> Phil Weldon schrieb:
>
>> Today, you can forget this rule.Just to clarify, when
>> you post 1:1.2 do you mean a ratio of 1 to 2 (1/2)?
>
> No, I mean 1 : 1.2, i.e. the clock-rate of the DDR2-memory
> (400MHz) is 1.2 times higher than the clock-rate of the
> FSB (333MHz).
>
>> Motherboards based on the nVidia 680i series chipset DO show
>> increased memory through-put when operated at higher Memory
>> clock : CPU clock ratios given tight enough memory parameter
>> delays.
>
> So it seems Intel did a better Job on that. I clocked my Core
> II Extreme at a FSB of 400/1600MHz (core-clock 3.2GHz) so that
> the memory-FSB-ratio is 1:1 and the memory-throughput was sligh-
> tly lower than at 333MHz (equal memory-clocking of 400MHz aka
> 800MHz DDR).
>
> And there's another thing to mention: latency is a more
> important parameter than throughput in real-world-apps.
bob johnson wrote:
> sorry
>
> they were only 30k
>
>
> bob
No need to apologize. Just a FYI regarding text groups. Now, get busy
overclocking! ;-)
> "RobV" <robv@nowhere.invalid> wrote in message
> news:47c9fa90$0$22824$4c368faf@roadrunner.com...
>> bob johnson wrote:
>>> OK
>>>
>>> Got my OC back. Now, how to push my memory carefully and w/
>>> stability?
>>
>> Just a friendly notice here, Bob. alt.comp.hardware.overclocking is
>> a text only news group. As such, if you would like those reading to
>> see a picture of something, the correct thing to do is post it at a
>> web site, then give the link to the web address.
>>
>> Posting pictures, or attachments, isn't too large a problem for
>> those of us using Outlook Express, but many folks use a strict
>> newsreader program that shows HTML, pictures, attachments and such
>> as just a jumble of characters, letters and numbers.
>>
>> News groups that allow these things will have binary in the NG name.
>>
>> Now, carry on...;-)
So what is the difference between SPD and Memory in cpu-z? SPD just reports
the spec not the actual speed?
Also, is it safe to run as I am (scaled down timing)? I'm still a little
confused because I though the Gig MB was suppposed to compensate everything
but cpu speed unless you choose to do so!
thx
bob
"Paul" <nospam@needed.com> wrote in message news:fqcnng$8s0$1@aioe.org...
> bob johnson wrote:
>> Thanks, Paul
>>
>> My memory is rated at 5/5/5/15, if I'm not mistaken?
>>
>> So, if cpu-z is reporting 5/7/7/21, does that mean it's not running to
>> spec (slower)?
>>
>> thx
>>
>> bob
>>
>
> Your memory is DDR2-800 5-5-5-15, but it is running
> at DDR2-1088. The BIOS appears to have scaled the timing
> numbers, except CAS. So CAS is tighter than the spec
> for the memory.
>
> Paul
bob johnson wrote:
> I see,
>
> So what is the difference between SPD and Memory in cpu-z? SPD just reports
> the spec not the actual speed?
>
>
> Also, is it safe to run as I am (scaled down timing)? I'm still a little
> confused because I though the Gig MB was suppposed to compensate everything
> but cpu speed unless you choose to do so!
>
> thx
>
> bob
>
There are three sets of timing:
1) SPD - intended to be read by the BIOS and used to set
timing when the BIOS memory settings are at "Auto". So
the SPD is what makes Auto possible. SPD conditions should
be non-aggressive, so for motherboards that don't deliver
enough Vdimm at startup, the motherboard will still run.
2) Advert or datasheet timing. A stated value, as to the max
speed at some voltage value, that the manufacturer has tested
to. This could be more aggressive or higher than the value
stored in the SPD. Generally, this difference exists on
"enthusiast" memory, while a more generic RAM, the advert
value is the same as the SPD. An enthusiast dials up the
datasheet value manually, as part of testing the limits of
the RAM. For example some people return memory products,
when they don't meet the advertised speed rating.
3) BIOS manual setting. This is the value of timing and
voltage, that the user arrives at by thorough testing.
For example, if your memory passes DDR2-1088 CAS5 and is
error free in memtest86+, Prime95, and a run of Crysis,
then you might conclude the setting is stable. Some people
"crank until unstable" and then back off a notch or two,
and run tests for extended periods of time. That is how my
system is set right now, tested for instability, and then
backed off a notch or two.
So right now, you're at (3), but by accident. You raised the
CPU speed, and the memory went up too. Right now, you don't
know if your memory would be stable at CAS4 instead of CAS5 for
example. Or perhaps, you might not know what Vdimm is being used,
and perhaps a little less could be used, without compromising
memory speed.
If you wanted to drop the memory speed, selecting something lower
than DDR2-800 in the BIOS, should cause a lower ratio to be selected
with respect to the CPU input clock. That takes some of the stress
off the memory, perhaps allowing a test to run longer without
reporting an error.
My 2 settings in bios for timing and speed are both set to "auto", so I'm
still confused
thx
bob
"Paul" <nospam@needed.com> wrote in message news:fqfbs2$kap$1@aioe.org...
> bob johnson wrote:
>> I see,
>>
>> So what is the difference between SPD and Memory in cpu-z? SPD just
>> reports
>> the spec not the actual speed?
>>
>>
>> Also, is it safe to run as I am (scaled down timing)? I'm still a little
>> confused because I though the Gig MB was suppposed to compensate
>> everything
>> but cpu speed unless you choose to do so!
>>
>> thx
>>
>> bob
>>
>
> There are three sets of timing:
>
> 1) SPD - intended to be read by the BIOS and used to set
> timing when the BIOS memory settings are at "Auto". So
> the SPD is what makes Auto possible. SPD conditions should
> be non-aggressive, so for motherboards that don't deliver
> enough Vdimm at startup, the motherboard will still run.
>
> 2) Advert or datasheet timing. A stated value, as to the max
> speed at some voltage value, that the manufacturer has tested
> to. This could be more aggressive or higher than the value
> stored in the SPD. Generally, this difference exists on
> "enthusiast" memory, while a more generic RAM, the advert
> value is the same as the SPD. An enthusiast dials up the
> datasheet value manually, as part of testing the limits of
> the RAM. For example some people return memory products,
> when they don't meet the advertised speed rating.
>
> 3) BIOS manual setting. This is the value of timing and
> voltage, that the user arrives at by thorough testing.
> For example, if your memory passes DDR2-1088 CAS5 and is
> error free in memtest86+, Prime95, and a run of Crysis,
> then you might conclude the setting is stable. Some people
> "crank until unstable" and then back off a notch or two,
> and run tests for extended periods of time. That is how my
> system is set right now, tested for instability, and then
> backed off a notch or two.
>
> So right now, you're at (3), but by accident. You raised the
> CPU speed, and the memory went up too. Right now, you don't
> know if your memory would be stable at CAS4 instead of CAS5 for
> example. Or perhaps, you might not know what Vdimm is being used,
> and perhaps a little less could be used, without compromising
> memory speed.
>
> If you wanted to drop the memory speed, selecting something lower
> than DDR2-800 in the BIOS, should cause a lower ratio to be selected
> with respect to the CPU input clock. That takes some of the stress
> off the memory, perhaps allowing a test to run longer without
> reporting an error.
>
> HTH,
> Paul
bob johnson wrote:
> Paul,
>
>
> My 2 settings in bios for timing and speed are both set to "auto", so
> I'm still confused
>
>
> thx
>
> bob
>
OK, so how did it happen ? Here is my guess...
You bought DDR2-800 RAM. "Auto" for memory selects DDR2-800
divider, with respect to the nominal FSB of the processor.
Next, you increased the CPU clock. Now, not only the
CPU core frequency rises as a result, the memory input
clock is also tied to the CPU clock. Now, your memory
*actually* runs at DDR2-1088.
So that was with the memory set at "Auto".
Now, instead, imagine you set the memory to "Manual".
And set the memory to DDR2-800 (i.e. its rated value).
Well, you'd get the same results. The resulting memory
speed is DDR2-1088, because your CPU is still overclocked.
Finally, imagine you set the memory to "Manual", and
selected DDR2-667. Now, you and I know, your memory
is rated for DDR2-800. "Wait", you say. "I'm going
to lose performance!". Well, the overclock on the CPU
would mean the DDR2-667 setting, is still going to
get overclocked by the same proportion, so the actual
running rate will be DDR2-907. That is still faster
than the actual RAM rating of 800.
Similarly, if you did "Manual" for the memory, and
selected DDR2-533 in the BIOS menu, the actual
running speed of the memory will be DDR2-725. Now
the memory is running slightly below its rated
speed.
The CPU clock change from 200 to 272, speeded up
everything. By going into the BIOS, setting the
memory clock to "Manual", and selecting a lower
value to compensate, you can get some lower
clock settings for the memory. "DDR2-533" setting
will yield DDR2-725, with your overclock of the CPU
to 272.
The memory timings have to be set in proportion to the
stated memory specs, and the actual running speed.
For example.
Memory rating DDR2-800 5-5-5-15
(1088 / 800) * 5 = 6.8, round to next highest integer, gives 7.
(1088 / 800) * 15 = 20.4, round to next highest integer, gives 21.
Since your memory is actually at 1088, the timing should be 7-7-7-21.
The BIOS used 5-7-7-21, which means only the CAS is still too tight.
The BIOS made a valiant attempt at fixing it, but doesn't seem to have
set the CAS correctly. It means the memory is doing **** good,
delivering uncorrupted data in only 5 cycles, when the arithmetic
predicts it needs 7 cycles before the data is ready on a read
cycle.
To be an overclocker, ratios and proportions have to be second
nature to you. While there are some motherboard BIOS that do
more of the thinking for you, for whatever reasons, this overclocking
thing seems to receive less attention to details. If you want
your computer to work well, it means verifying all the settings
the BIOS uses. Some guys, for example, cannot manage to think
this stuff through, and they'd conclude "this hardware sucks",
when in fact a little extra arithmetic, experimenting, and
verification with CPUZ, would have given them a working
system.
> The best fine-grained, free performance testing suite I know of is
> SiSoft Sandra Lite XIIc available at
> http://www.sisoftware.co.uk/ .
> SiSoft Sandra reports a large number of performance parameters for
> memory, CPU, and storage. AND it provides comparisons with hundreds
> of CPUs, chipsets, memory arrangements, and storage configurations. SiSoft
> Sandra XIIc is the opposite of a 'real world' benchmark; it is
> most useful in determining the effect of system changes and less
> useful for determing the real gains in computing power.
Hmmm. I gave up on Sandra a long time ago as unreliable. However, on your
recommendation I've just downloaded and installed the latest version (and
DotNet Framework 3.5 as it seems to use it fairly extensively).
Perhaps it's good for performance parameters as you suggest. However, I find
that it's just as useless as I found it several years ago (2005?) at
identifying devices/speeds accurately.
It reports my CPU speed as being 4.54GHz (as does Windows, it's actually
3.3, 413 x 8). It tells me that all my HDDs are "SATA150" and that the two
that are actually SATA II/300 have NCQ enabled while the third, older
SATA150 HDD doesn't have NCQ enabled.
It's telling me that my memory bus speed is 2x 568 (1.14GHz) when in fact I
have it set at 1:1 in BIOS as my RAM is only DDR2-800. It's actually also
telling me that my FSB is 568MHz
Under 'Performance Tips' it gives me warnings that my CPU FSB speed is
higher than rated ("!") and to reduce it. Same for CPU speed ("!").
Heh! I could go on. I guess that, as you say, it's trengths are in it's
testing suite rather than in it's hardware IDing.
Re: Gigabyte GA-P35 DS3P - RAM? 
Linear Mode
