I would like to know the extent of the gap between advertised
processor speed and the amount of that processing speed that can be
used before it reaches the barrier that is created by the naturally
slower ram. I am fluent in technology and realize that newer chipsets
promise faster FSB's but these speeds (in Mhz) are less than half of
the high end processors available. I realize that the CPU does not
need to transfer memory from cache to ram after every computation, but
there has to be some ratio at which no more gain is achevied in
increasing processor speed given a set ram speed. How much faster does
a P4 3.2E computer run compared to a celeron 2.0 with an infinite
supply of PC3200 ram running dual channel and negating HD and other
device latency? Huge difference, or not so huge?

Hello Dabryan,

The amount of "huge difference" will depend greatly on the type of
application you are running. If I had to pick one page to illustrate
this, see
  http://www.techreport.com/reviews/2002q3/celeron-1.7/index.x?pg=3
which compares a 1.7 Ghz Celeron with a 1.7 Ghz Pentium 4. Scroll down
to the Linpack results.

The performance is almost identical on the two processors until the
larger cache on the Pentium 4 comes into play. At that point, the
performance of the Pentium 4 continues to grow until it peaks at six
times as high as the Celeron. The performance difference drops after
that peak but still remains about 20% better than the Celeron. I
generally don't consider 20% improvement is "huge", but 6x is
certainly pretty good.

By increasing the clock speed (and related FSB speeds) you generally
get a proportional increase in performance. You are asking about
speeds roughly 2x of that in the charts described above - so let's
review
  http://techreport.com/reviews/2004q1/pentium4-3.4ghz/index.x?pg=3
which for the same Linpack test shows about 2x the performance for the
3.4 Ghz Pentium 4 extreme.

So in conclusion, the Pentium 4 will outperform a Celeron due to two factors:
 - more cache memory (effect varies by type of application)
 - raw clock speed increases (roughly proportional to HZ difference)

Search phrases used include:
  benchmark pentium 4
  benchmark celeron site:techreport.com
If you hae any further questions related to thic topic or if the
answer is unclear, please make a clarification request.


  --Maniac

---

Request for Answer Clarification by dabryan-ga on 09 Jan 2005 16:05 PST

thank you for your answer, jus a little clarification

"I realize that the CPU does not need to transfer memory from cache to
ram after every computation, but there has to be some ratio at which
no more gain is achevied in increasing processor speed given a set ram
speed."

I guess I could have said it easier by asking if there is a larger
gain of performance (in general) by increasing RAM speed or processor
speed.
I do not need sample benchmarks just a short explanation will suffice.

Thank you for your time
dabryan

---

Clarification of Answer by maniac-ga on 09 Jan 2005 19:17 PST

Hello Dabryan,

Again, the relative performance of two processes will vary with the
type of application. Let me use a relatively old example for
comparison. In the mid 1970's there were three major types of
supercomputers:
 - Cray 1 (vector processor)
 - Illiac (single instruction, multiple data)
 - CDC (fast, more general system)
and some applications were faster on one system than either of the
other two. There were other applications that would be better on a
second machine and worse on the first and third (and so on). There was
no clear "better" system between the three - if your application could
fit within the capacity of one system (and not in another), the
performance on that system would be MUCH better than another.

Going back to the original question, lets say you have an application
that fits within a 1 Mbyte CPU cache. For that application, you can
basically increase the CPU speed at will without the need to increase
the speed of main memory and get a performance boost. The CPU
basically at that point makes no accesses to memory.

For a second application that needs 4 Mbyte memory, you need to
increase both the CPU and memory speed to get a comparable
improvement. Alternatively, you can purchse a CPU with a 4 Mbyte CPU
cache and get the same effect as the previous one.

So again, the performance difference will depend on the specific
application you are trying to run. In general, an increase of CPU
speed will result in a proportional reduction in application run time
(as I mentioned before). This will be limited in large applications -
those much larger than the CPU caches. To compare, an increase in
memory speed will help only if your application makes frequent
references to more memory than in the CPU caches.

  --Maniac

Almost too thorough, if that is possible. Thank you for your speedy followup.