Searching \ for '[EE]:: Looking for definitive FSB / cpu clock / RA' in subject line. ()
Make payments with PayPal - it's fast, free and secure! Help us get a faster server
FAQ page: www.piclist.com/techref/timers.htm?key=clock
Search entire site for: ': Looking for definitive FSB / cpu clock / RA'.

Exact match. Not showing close matches.
PICList Thread
'[EE]:: Looking for definitive FSB / cpu clock / RA'
2007\03\29@085445 by Russell McMahon

face
flavicon
face
1.    I am looking for a simple (getting old) but complete and correct
explanation of the relationships between Intel processor PCs of -

   CPU clock speed
   FSB speed.
   Necessary RAM speeds / ratings (DDR & DDR2 especially).
   Any other relevant clock speeds and timings (of which there can be
several).

There is any amount of material available via Gargoyle and Wikipedia.
By selecting material to suit you can get any number of incompatible
explanations.
Necessarily most are partially wrong or incomplete.

Can anyone either explain what needs to be known to come to grips with
the subject OR provide a suitably definitive reference?

___________

2.    "Super Pi" as a benchmark  / Slow PC woes.

We've been running the "Super Pi" benchmark, producing pi to 1 million
places, as a means of comparing various PCs. As a consequence my
present desktop PC now runs 30% faster! :-) !!!! It's a P4 3 GHZ 1 MB
L2, 800 MHz FSB and I found that it was substantially slower than
comparable machines. A bit of digging showed that the RAM was being
clocked at 200 MHz and changing this showed why, as the machine got
very unstable. After various playings it appears (jury still out) that
the LGA775 cpu had had too little and inferior thermal grease used and
that apparently the cpu was throttling back to stay alive. No alarms
were being given and this theory seems feasible but a bit unlikely -
but it's now running Super Pi to 1 million places literally 30% faster
and with the RAM clock set to "auto" it's now not crashing. So far
anyway.

My PC now calculates Pi to 1 million places in just over 51 seconds
which is slightly faster per cpu clock speed than some comparable PCs
we've tested. Best figure we have is 43 seconds for a purpose built
super fast gamer PC - suggesting that the less than 20% extra speed he
gets is hardly worth the effort. This is of course only one benchmark
out of zillions but was certainly useful as an indication that there
was something wrong with this PC.

Super Pi program here

           http://others.servebeer.com/misc/superpi.exe
free

Very useful processor and system information extractor.
Program name is not the original one but it makes it easier to
remember :-)

         http://others.servebeer.com/misc/intelinside.exe
free




       Russell



2007\03\29@102425 by M. Adam Davis

face picon face
Now that I've written all the junk below, and had to search for a few
things, check out a better explanation here:
http://www.directron.com/fsbguide.html

You want the relationships between them?  Interesting...  I will spew
what I understand, which might give you some insight so you can search
for what you're really after.

The CPU clock is the internal clock on the CPU itself, which is
created by PLL multiplying the FSB (Front Side Bus) clock.  I believe
the system clock is largely controlled/distributed by the northbridge
chip, which is the chip that connects and routes datta between the
CPU, RAM, high speed busses (AGP, PCIe) and the southbridge which
handles slower busses and peripherals (PCI, ISA, USB, firewire,
motherboard functions, etc).

So with a FSB clock of 1.066GHz, and a CPU clock of 2.6GHz, the PLL
inside the CPU is multiplying the FSB by around 1.5.

The CPU has on-chip cache's (L1 and L2) so that the CPU is not
terribly limited by the slow FSB.  Most programs loop a lot, so the
CPU is often running at top speed a large percentage of its time, and
the slower FSB is fast enough to push data through the program loop
running out of cache.

So the northbridge is running at the FSB clock rate.  It connects
directly to the RAM, the processor, and the high speed busses and
moderates a very intense discussion between the three.

Higher end northbridge chips have dual channel RAM support, meaning
that instead of putting, say, four RAM chips in parallel and sharing
98% of their data and address lines, it has two channels that each
support two sticks of RAM.  This effectively doubles the RAM
throughput.

Then we can start talking about DDR and DDR, and this is where it gets
interesting because RAM is not the only thing in the system running at
DDR or DDR2.

DDR is Double Data Rate.  For each clock _cycle_ data is pumped down
the wire twice.  This means that at both the positive and negative
edges of the clock, data is placed on the buss in parallel.

This doubles the data rate relative to the clock.

DDR2 doubles this - each clock cycle data is placed on the bus four
times - it's simple to quadruple a clock of a relatively known
frequency without involving a PLL.  So each clock edge and inbetween
each edge data is placed on the buss.

Technically I've just described QDR - Quad data rate.  DDR2 is not
QDR, and the ram is not pumped four times per clock.  In the end the
throughput is the same, but the technique and technology are
different.

The CPU itself is running at QDR.  The FSB clock is actually only
266MHz.  The CPU PLL is actually multiplying that FSB clock by 9 to
get your 2.4GHz cpu clock.

Then we finally get to the memory.  You can research online for exact
details (including specific memory timing info) but the numbers are
using (ie, DDR2-5300) give relative performance information.  5300 in
this case means that the memory is theoretically capable of a
throughput of 5,300MB/s.  The FSB is running at 266, and since it's
DDR2 ram then it's effectively transferring data at a clock of 1066MHz
(Some fiddling here, as the memory clock actually is 667).

So... Go read the article I linked above.  It corrects some of the
info I've given, and expounds in more detail what it means to have
_matching_ component speeds, rather than just the fastest available.
For instance, depending on your processor and FSB, it may be better to
have slower ram than the fastest available so the data is there when
you need it and not before.

-Adam

On 3/27/07, Russell McMahon <spam_OUTapptechTakeThisOuTspamparadise.net.nz> wrote:
{Quote hidden}

> -

2007\03\30@065059 by Russell McMahon

face
flavicon
face
1.    I am looking for a simple (getting old) but complete and correct
explanation of the relationships between Intel processor PCs of -

   CPU clock speed
   FSB speed.
   Necessary RAM speeds / ratings (DDR & DDR2 especially).
   Any other relevant clock speeds and timings (of which there can be
several).

There is any amount of material available via Gargoyle and Wikipedia.
By selecting material to suit you can get any number of incompatible
explanations.
Necessarily most are partially wrong or incomplete.

Can anyone either explain what needs to be known to come to grips with
the subject OR provide a suitably definitive reference?

___________

2.    "Super Pi" as a benchmark  / Slow PC woes.

We've been running the "Super Pi" benchmark, producing pi to 1 million
places, as a means of comparing various PCs. As a consequence my
present desktop PC now runs 30% faster! :-) !!!! It's a P4 3 GHZ 1 MB
L2, 800 MHz FSB and I found that it was substantially slower than
comparable machines. A bit of digging showed that the RAM was being
clocked at 200 MHz and changing this showed why, as the machine got
very unstable. After various playings it appears (jury still out) that
the LGA775 cpu had had too little and inferior thermal grease used and
that apparently the cpu was throttling back to stay alive. No alarms
were being given and this theory seems feasible but a bit unlikely -
but it's now running Super Pi to 1 million places literally 30% faster
and with the RAM clock set to "auto" it's now not crashing. So far
anyway.

My PC now calculates Pi to 1 million places in just over 51 seconds
which is slightly faster per cpu clock speed than some comparable PCs
we've tested. Best figure we have is 43 seconds for a purpose built
super fast gamer PC - suggesting that the less than 20% extra speed he
gets is hardly worth the effort. This is of course only one benchmark
out of zillions but was certainly useful as an indication that there
was something wrong with this PC.

Super Pi program here

           http://others.servebeer.com/misc/superpi.exe
free

Very useful processor and system information extractor.
Program name is not the original one but it makes it easier to
remember :-)

         http://others.servebeer.com/misc/intelinside.exe
free




       Russell



More... (looser matching)
- Last day of these posts
- In 2007 , 2008 only
- Today
- New search...