P4A,B and C and memory bandwidth

[Guest]

Archived from groups: alt.comp.hardware.overclocking (More info?)

General question:

A P4 400mhz w. a 845 chipset at FSB100 is optimized at 3,2mhz mem.bandwith ,
right ?
A P4 533mhz at FSB133 with a 865 chipset is maxed out at 4,2 ghz
mem.bandwith, isn´t it?
A P4 800mhz at FSB200 - 6,4ghz mem.bandwith ?

I´m wondering if there´s a point, where more memory bandwith is simply just
a "waste of money" due to CPU and chipset limitations?

Here´s my second question:

My system:

Motherboard:
P4P800 w. Intel i865P/PE/G/i848P rev. A2
CPU:
2,8ghz P4B 533mhz
Ram:
Manufacturer (ID) GEIL (7F7F7F1300000000)
Size 256 MBytes
Max bandwidth PC3200 (200 MHz)
Part number GL 128M NNTTG-7
Serial number 31323334
Manufacturing date Week 03/Year 01

I´m sure I bought this ram as PC2700. no idea why CPUZ claims it is PC3200!

Right now : 2,8@3,36 P4B w. org.Vcore at FSB 160 now.As i had to made a
choice betrween a single stick of 512 PC3200 noname og two sticks of 256mb
PC2700 Geil I picked the last one due to dual channel feature.

Memory Frequency 200.0 MHz (4:5)
CAS# 2.5
RAS# to CAS# 4
RAS# Precharge 4
Cycle Time (tRAS) 8

It´s all stable.

I´would run my memory synch at 2x FSB160 (DDR320) and the change in mem.
bandwith is under 200mhz... :-/


But
I tried synch, asynch, large timings, best timings, turbo-best timings so I
even lost my dual channel, but still Sandra and AIDA32 benches my memory
bandwith between 3700mhz and 4200mhz with these IMO totally different diff.
ram settings.There´s not much difference between single channel mode or dual
channel mode eighter. Shouldn´t dual channel double-up mem. bandwith ?

All confused right now..pls. help out here...

G.

 

[Guest]

Archived from groups: alt.comp.hardware.overclocking (More info?)

"N´far" wrote in message...
> A P4 400mhz w. a 845 chipset at FSB100 is optimized at 3,2mhz mem.
> bandwith , right ?

Yes and no. A 400MHz FSB P4 has 3.2GB sec theoretical bandwidth between the
northbridge and the CPU (the "bus bandwidth"). However, the original 845
chipset only supported PC133 single data rate SDRAM memory, so this solution
was a bit of a car crash.

Intel's single channel SDRAM solutions were always severely compromised,
just to get an SDRAM (and then DDR SDRAM)/P4 combination out to market.
Claiming the 845 chipset was "optimised" is probably a misuse of the term!
The numbers only started to add up when the 800FSB/dual channel DDR400
solutions started to come out.

> A P4 533mhz at FSB133 with a 865 chipset is maxed out at 4,2 ghz
> mem.bandwith, isn´t it?

Again the bus bandwidth is 4.2GB/sec, but with 865 and 875 in dual channel
configurations there's the ability for the memory bus to be theoretically a
lot faster than this figure. In practice the memory bus is never 100%
efficient, but with a 533MHz P4 on an i865 dual channel platform, there's
little point running memory any quicker than PC2700, as the processor just
can't use it all.

> A P4 800mhz at FSB200 - 6,4ghz mem.bandwith ?

6.4GB/sec bus bandwidth, and 6.4GB/sec *theoretical* memory bandwidth. Real
world memory throughput is lower than this due to latencies and other
controller inefficiencies.

> I´m wondering if there´s a point, where more memory bandwith is
> simply just a "waste of money" due to CPU and chipset limitations?

There certainly is with the 400 and 533MHz FSB CPU's on the dual channel
chipsets. With the 800FSB parts though, the bus bandwidth is almost always
higher than the available memory bandwidth, and this also scales if you
overclock, so the law of diminishing returns never sets in to the same
extent - you just liberate even greater levels of performance until the rig
gets unstable.

> Here´s my second question:
>
> My system:
> Motherboard:
> P4P800 w. Intel i865P/PE/G/i848P rev. A2

Eh? Is it 865P or 848P? ;-)

> CPU: 2,8ghz P4B 533mhz
> Ram: GEIL PC3200 (200 MHz)
>
> I´m sure I bought this ram as PC2700. no idea why CPUZ claims it
> is PC3200!

Don't complain! ;-) Maybe they had a surplus of PC3200 in their stock bins
when your memory was packed, so you got the free upgrade! Could also be that
the SPD chips don't match the capability of the memory, so might not be all
roses.

> Right now : 2,8@3,36 P4B w. org.Vcore at FSB 160 now.As i had to
> made a choice betrween a single stick of 512 PC3200 noname og two
> sticks of 256mb PC2700 Geil I picked the last one due to dual channel
> feature.

Good move. Got you a lot more memory performance.

> But I tried synch, asynch, large timings, best timings, turbo-best
> timings so I even lost my dual channel, but still Sandra and AIDA32
> benches my memory bandwith between 3700mhz and 4200mhz with
> these IMO totally different diff. ram settings.

Could be that you're running into the processor bus bottleneck. Have you
checked with CPU-Z to make sure that the RAM is actually following the
settings you make in the BIOS? Some motherboards unhelpfully override manual
settings if they're too far outside SPD values for example.

> There´s not much difference between single channel mode or
> dual channel mode eighter.

Not that I've done too much testing with this combination but you shouldn't
expect huge differences. You're only running a 533MHz CPU (albeit
overclocked) so the bus bandwidth is likely to be the limiter.

> Shouldn´t dual channel double-up mem. bandwith ?

No, it doesn't multiply by 200% (just as two disk RAID 0 arrays aren't twice
as fast as a singleton). Depending on RAM timings, bus speeds, memory
controller design and so-on, you could probably expect a 170-180%
improvement in bandwidth between running one channel and two of the same
memory, provided all other things were equal and that there were no other
factors bottlenecking the performance.

> All confused right now..pls. help out here...

Not sure I've given you the unambiguous answer you're looking for, but hope
the above helps!
--


Richard Hopkins
Cardiff, Wales, United Kingdom
(replace .nospam with .com in reply address)

The UK's leading technology reseller www.dabs.com

 

[Spajky]

Archived from groups: alt.comp.hardware.overclocking (More info?)

On Fri, 7 Jan 2005 17:38:21 +0100, "N´far" <rudiNOSPAM@mail.dk> wrote:

>I tried synch, asynch, large timings, best timings, turbo-best timings so I
>even lost my dual channel, but still Sandra and AIDA32 benches my memory
>bandwith between 3700mhz and 4200mhz with these IMO totally different diff.
>ram settings.There? not much difference between single channel mode or dual
>channel mode eighter. Shouldn? dual channel double-up mem. bandwith ?
>
>All confused right now..pls. help out here...

read stuff @ my site under comp/benches .. 1st "new" link;
more points you get with WinRAR, better is; forget for practical life
Sandra mem bandwith bench ...
--
Regards & Happy New Year Everyone, SPAJKY ®
& visit my site @ http://www.spajky.vze.com
"Tualatin OC-ed / BX-Slot1 / inaudible setup!"
E-mail AntiSpam: remove ##

 

[Guest]

Archived from groups: alt.comp.hardware.overclocking (More info?)

I tried the RAR stuff.
68 ns and 370. I think I´m all fine.
Thanks a lot, both of you.