internal speed is high, external speed is low!!

  • Thread starter Thread starter esara
  • Start date Start date
E

esara

Hi
I do not see in this scenario that why we need high speed memory..

If we have Athlon XP1800 which is using:

11.5 multiplier X 133MHz FSB and if we have DDR266

I understand that the speed of the memory is 266Mhz but the CPU can
communicate with the Memory using FSB speed which is 133. I can
imagine this like a highway connect 2 city one city called CPU which
allow internal traffic at speed 1533 AND another city which is called
"Memory" which allows the internal traffic at speed 266MHz. Now if a
car want to travel from CPU to Memory it can only drive at max speed
which is 133Mhz. Now my question is, what is the point that I have
internal speed of both the CPU and Memory high when the external road
is low 133Mhz.I thought the FSB speed should be high so information
can traval faster between CPU and Memory or any other part of the PC.

Any help would be very much appreciated.
 
esara said:
Hi
I do not see in this scenario that why we need high speed memory..

If we have Athlon XP1800 which is using:

11.5 multiplier X 133MHz FSB and if we have DDR266

I understand that the speed of the memory is 266Mhz but the CPU can
communicate with the Memory using FSB speed which is 133.
Click to expand...

You are forgetting that the DDR is dual data rate. Thus, 266 is clocked
externally at 133, the same as your FSB. But it actually makes no
difference at all, as most (all?) modern chipsets can run your memory at any
speed up to maximum supported by the chipset, whether it matches the FSB or
not. For best performance (read: max benchmark speeds), it's best to have
the RAM match the CPU. Thus, 166 CPU would use DDR333 RAM and so on. But
that 166CPU would be perfectly happy with DDR400 RAM, also. If the chipset
supports it, the DDR400 can run at DDR400, or it will just run at DDR333
(slightly underclocked) if that's all the chipset can do. -Dave
 
What none of can figure out is what gave you idea that the 1800+ XP only
allows a 133MHz FSB.
Every 1800+ data sheet I have from AMD says that the 1800+ allows a 266MHz.
FSB, and thats were you are running it.

JPS
 
jpsga said:
What none of can figure out is what gave you idea that the 1800+ XP only
allows a 133MHz FSB.
Every 1800+ data sheet I have from AMD says that the 1800+ allows a 266MHz.
FSB, and thats were you are running it.

JPS
Click to expand...


That's marketing hype. The XP1800 runs at 133FSB, 266DDR.


bluestringer
 
Howdy!

esara said:
Hi
I do not see in this scenario that why we need high speed memory..

If we have Athlon XP1800 which is using:

11.5 multiplier X 133MHz FSB and if we have DDR266

I understand that the speed of the memory is 266Mhz but the CPU can
communicate with the Memory using FSB speed which is 133. I can
imagine this like a highway connect 2 city one city called CPU which
allow internal traffic at speed 1533 AND another city which is called
"Memory" which allows the internal traffic at speed 266MHz. Now if a
car want to travel from CPU to Memory it can only drive at max speed
which is 133Mhz. Now my question is, what is the point that I have
internal speed of both the CPU and Memory high when the external road
is low 133Mhz.I thought the FSB speed should be high so information
can traval faster between CPU and Memory or any other part of the PC.

Any help would be very much appreciated.
Click to expand...

Dave C. explained part of it pretty good. But there's another
portion - if the CPU is using 100% of the memory bandwidth, exactly WHEN
does the HD / floppy / sound / video / et al get access to it? <B-)

In a perfect world, we could run the RAM at 2x the CPU speed, so
that the CPU got it half the time and the rest of the world the other half.

Matter of fact, that was the primary initial design point behind DDR
on the processor - one half the clock for RAM, one half for I/O. Didn't
stay that way ...

RwP
 
Hi
I do not see in this scenario that why we need high speed memory..

If we have Athlon XP1800 which is using:

11.5 multiplier X 133MHz FSB and if we have DDR266

I understand that the speed of the memory is 266Mhz but the CPU can
communicate with the Memory using FSB speed which is 133.
Click to expand...

- No, no, and no again, basically.
The speed of the memory bus is not 266MHz. The bus frequency is
133MHz, the DDR (double data rate) speed is 266, that's why it's
called DDR266. (Don't ask me how it works. Please feel free to
explain, if anybody knows the details of DDR. I could guess, it's
something similar to the EV6 bus?)

The CPU does not communicate directly with memory. It communicates
with the 'Northbridge' part of the chipset, which contains the memory
controller and connects to ram via the memorybus (133MHz), and also
connects to AGP and Southbridge(PCI, IDE and other).

The FSB speed is 266MHz! The FSB-_CLOCK_ is 133MHz.
....And the (AMD&Athlon) Alpha EV6 protocol FSB effective speed is
266MHz on a 133MHz clock. (The EV6 bus doesn't bother with any
multiplier, like the CPU, it simply syncs transfer enables, on BOTH
rising and falling flank of the clock, so the bus speed is TWICE the
clock).
The FSB-CLOCK is highly visible in BIOS. Often it's also called simply
FSB, in many bios'es, instead of 'fsb-clock' or 'external-clock'. Thus
a lot of people assume this is the bus speed. (and possibly becomes a
bit confused about processor-, chipset-specs)
It's very common that people in this group says things like "set your
FSB to 166MHz". Now that happens to be 'short' form, and what they
really says is: "set your FSB _clock_ to 166MHz, and thus your FSB
_speed_ to 333MHz".
It's also common that people confuses front side bus and memory bus.

[CPU]
|
fsb 133 clock = 266MHz speed
|
[Northbridge]
|
memorybus 133MHz = 266DDR speed
|
[RAM]

So basically we have in both cases some kind of synchronization, that
runs at 133MHz, but both the actual data transfer speeds are 266MHz.

We don't have to have precisely that configuration. The FSB and
memorybus can run on different frequencies. You basically want as fast
FSB as possible. But the normal AMD case nowadays, is that the
available DDR speed is as high or higher than the available fsb speed.
And in this case, the general recommendation is to run FSB and
memorybus in sync. (That is: lowering the memory speed to match FSB. -
Don't lower FSB to match memory. If you can run FSB faster, - do so!)

Examples:
You have an AthlonXP with 333MHz FSB. You also have PC3200 ram, which
can run at DDR400. Run the ram at DDR333, to be more stable and to run
in sync with the FSB.

You have an AthlonXP with 333MHz FSB. You also have PC2100 ram, which
can run at DDR266. Run the ram at DDR266 and the FSB asynchronously
still at 333MHz. (This is also a case where dual channel will serve
you well.)

ancra
 
That's marketing hype. The XP1800 runs at 133FSB, 266DDR.
Click to expand...

No it's not marketing hype. It's confusing issues, confusing memorybus
with cpu fsp, confusing clocks with speed.
The _SPEED_ of the Athlon's FSB has _never_ been lower than 200MHz!
And the FSB speed of the 1800+ is 266MHz, just like the AMD data sheet
says. No marketing hype at all about it.
And DDR or any memory speed at all, have nothing to do with the FSB.

ancra
 
memorybus 133MHz = 266DDR speed
Click to expand...

.... & @ same mem clock (effective), the DDR is practically only 10%
approx. in real life faster (theoretical max bandwith is double in
SiSoft Sandra for example)

so best real life performance bench between different platforms for
compares, is to use incorporated mem subsistem bench in WinRAR;
for example I get 315 points (index) on my Sdram @ 135fsb/mem clock...
 
so best real life performance bench between different platforms for
compares, is to use incorporated mem subsistem bench in WinRAR;
Click to expand...
The LAST (3.3) verion of it .. (sorry I forgot to mention...)
 
... & @ same mem clock (effective), the DDR is practically only 10%
approx. in real life faster (theoretical max bandwith is double in
SiSoft Sandra for example)
Click to expand...

I might be misunderstanding you, but are you thinking of
application/cpu performance here, against ram? As opposed to just
memory bandwidth?
Then you're right. General performance is more affected by memory
latency (I'm not speaking CAS, I mean the whole system, cpu
memorycontroller) than by bandwidth.
And I 'think' (meaning I'm not sure) sdram systems have relatively (to
rate) excellent latency.
But bandwidth do affect performance, and it all depends a lot on the
application and cpu involved. If the application/cpu bottlenecks in
memory bandwidth, then I'm sure you'd see more significant
differences.
so best real life performance bench between different platforms for
compares, is to use incorporated mem subsistem bench in WinRAR;
for example I get 315 points (index) on my Sdram @ 135fsb/mem clock...
Click to expand...

Yes, I mean I don't think there is such a thing as 'best R.L.P.
bench', but sure, performance doesn't scale linear with bandwidth.

Have you actually measured the performance on similar systems with
both SDRAM and DDR?
You should have posted something about this in the SDRAM vs DDR
thread. ;)

ancra
 
I have often read, as here, that it is best not to have memory running
asychronously to the CPU. But with whatever benchmark I try I get better
results with my Athlon TB (133Mhz) and my memory running at 166Mhz (PC2700).
I.E. 3Dmark 2001 score drops by about 500 if I set the memory to the same
133Mhz as the CPU...So what is the argument based on?
 
I might be misunderstanding you, but are you thinking of
application/cpu performance here, against ram? As opposed to just
memory bandwidth?
Click to expand...

yes ...
Then you're right. General performance is more affected by memory
latency (I'm not speaking CAS, I mean the whole system, cpu
memorycontroller) than by bandwidth.
And I 'think' (meaning I'm not sure) sdram systems have relatively (to
rate) excellent latency.
But bandwidth do affect performance, and it all depends a lot on the
application and cpu involved. If the application/cpu bottlenecks in
memory bandwidth, then I'm sure you'd see more significant
differences.
Click to expand...

yes, differences in "real" life applications go approx. 5 - 20%
better, in average around 10% (depends how particular program uses
memory & its bandwith....)
Yes, I mean I don't think there is such a thing as 'best R.L.P.
bench', but sure, performance doesn't scale linear with bandwidth.
Click to expand...

sure ...
Have you actually measured the performance on similar systems with
both SDRAM and DDR?
Click to expand...

Not by myself in person but:
Quite time ago, when DDR have been introduced, there were a lot of
different Benches made & published on web sites; (compares on similar
MoBos & same CPUs but with both Ram technologies; in average there
showed DDR around 10% boost in performance of a sistem using AMDs, a
bit better with Intel ...); with going to dual channel DDR, you get
another few % boost (but bandwith measured almost doubles!)

While OC-ing I also time ago noticed, that higher you go with FSB &
mem speed & more fast Cas timing /on same CPU speed!/, you gain less
than you should in real life (not benches using theoretical scores)
; interesting that... but ... its more an evolution than a revolution
in technology ...
on the other side few years ago OC-ing made a huge impact on
performance, but these days much less IMHO ... you get much more with
filling the machine with as much ram it can handle & fastest HD you
can afford, since their performance is the slowest stuff you have
(meccanical stuff!) inside the case ... & OSes progs tend to get
bigger & bigger (more bloated!) so you need that ...

the other what matters: past months the DDR was much cheaper than
Sdram; I filled my machine with sdram when the situation was oposite a
year or two ago ...

Try that WinRAR3.3 bench, it shows clearly the practical speed of a
mem subsistem & depends only on mem real MHz clock (not other like Cpu
speed etc...); in some local Usenet thread here we got interesting
results ... http://www.rarlab.com/ ...
 
I have often read, as here, that it is best not to have memory running
asychronously to the CPU. But with whatever benchmark I try I get better
results with my Athlon TB (133Mhz) and my memory running at 166Mhz (PC2700).
I.E. 3Dmark 2001 score drops by about 500 if I set the memory to the same
133Mhz as the CPU...So what is the argument based on?
Click to expand...

Cool. Now try your PC2700 at DDR266 cl2.0.

Well, it's basically based upon the recommendation mostly being given
in this group.
That in turn is based upon a number of benchmarks, published on
various hardware websites, that show at the best, an "insignificant"
gain from running memory speed higher. And also even lower
performance. But also, I suspect, it's based upon "conclusions",
formulated in those articles.

However, it may be those results can't be generalized across all
chipsets and speeds the way they generally are.
My own experiements tend to hint that you can reduce CL ( PC3200 cl2.5
can usually be run at DDR333 cl2.0) to achieve slightly better
performance at lower - in sync - speeds.
That's also why I used the expression "the general recommendation is"
rather than just flatly stating "run in sync" - the way a conceited,
arrogant B. like me, would normally put it.
Wildly speculating, it may also be that memory bandwidth benchmarks
are overemphasized in that argument. There could be a latency factor
that is more significant. I don't know, just speculating a bit.

ancra
 
Back
Top