Suggestions for Core 2 Duo systems that use PCI - not PCI express?

[JR Weiss]

http://en.wikipedia.org/wiki/Intel_Celeron
Subsequent Celeron branded CPUs were based on the Pentium III, Pentium 4,
Pentium M, and Core 2 Duo branded processors.

I am unable to find any reference on the Intel web site to a Core2 Duo
Celeron. If you can find such an animal, please post the Intel reference.

 

[Conor]

Isn't the cheapest Intel CPU that's called a "Core 2" about
$120? That seems a a bit higher than bottom of barrel when
you can get a dual core Athlon x2 for $40.

But there's one hell of a difference in performance.

 

[geoff]

Here is a low-end machine

Intel Celeron E1400 Allendale 2.0GHz LGA 775 65W Dual-Core Processor
http://www.newegg.com/Product/Product.aspx?Item=N82E16819116069
$50

Foxconn G31MXP-K LGA 775 Intel G31 Micro ATX Intel Motherboard - Retail
http://www.newegg.com/Product/Product.aspx?Item=N82E16813186167
$45

Rendition by Crucial 1GB 240-Pin DDR2 SDRAM DDR2 800 (PC2 6400) Desktop
Memory - Retail
http://www.newegg.com/Product/Product.aspx?Item=N82E16820148239
$10 x 2 = $20 (2 gigs)

Seagate Barracuda 7200.11 ST3160813AS 160GB 7200 RPM 8MB Cache SATA 3.0Gb/s
Hard Drive
http://www.newegg.com/Product/Product.aspx?Item=N82E16822148397
$40

Celeron Dual Core
Foxconn Motherboard
2 gigs memory
160 gb HD
$145

I do not think pci express is going to affect that price much, in case he
feels the above is high end.

--g

 

[geoff]

Keep in mind the OP was hoping for 3X the speed of a 2.4GHz

P4, an E1400 isn't likely to achieve this.

This change should:

http://www.newegg.com/Product/Product.aspx?Item=N82E16819116063
Intel Pentium E2200 Allendale 2.2GHz 1MB L2 Cache LGA 775 65W Dual-Core
Processor - Retail
$70

Intel Pentium E2200 Dual-Core
New Price $165

--g

 

[geoff]

Maybe if it's overclocked...

With SATA 3 gb, DDR2 memory, and dual core, it would be significantly
faster. However, one can bump up the processor another level to the dual
core E5400 and the CPU would still be under $100.

The new total price would be $185.

If that is still not enough of a bump compared to his P4 system then there
is no answer for him.

--g

 

[JR Weiss]

This change should:

Intel Pentium E2200 Dual-Core

How do you figure 3X speed by going from a 2.4 GHz P4 to a 2.2 GHz C2D? What
app will work with 3X the speed?

 

[Rarius]

How do you figure 3X speed by going from a 2.4 GHz P4 to a 2.2 GHz C2D?
What app will work with 3X the speed?

Point 1:
The P4 is a single core, the C2D is a dual core (the clue is in the
name!)
Point 2:
Each C2D core does about 30-40% more per MHz than the P4.

This means a 2.2GHz C2D is about 2.5x as fast as a 2.4GHz P4... then you
overclock it! Almost all the C2Ds are easily overclockable with standard air
cooling (I am not talking expensive coolers either!). It would EASILY be
able to overclock a C2D from 2.2GHz up to 3GHz. I took my C2Q from 2.4 to
3.2!

As for "What app will work with 3x the speed?"... Most of them! Try doing
some heavy Photoshop or video encoding, or play a game or two!

Rarius

 

[geoff]

You seem focused more on general system building than what

was asked for. If that's not enough of a bump it's same
story as always, spend a little more.

That is how marketing works, with cars, for example, if you want more HP,
you pay more.

At $185, that is pretty bottom of the barrel but would still be a lot faster
than a P4.

I dont need to be on the cutting edge, what Core 2 chips should I look
for that would net me about 3x the speed of that 2.4gig P4?

The above system is so far away from cutting edge that one could drive an
airplane through a gap as wide as that.

--g

 

[~misfit~]

See below. I'd say an E7300 or better, depending on what you are doing with
it.


VirtualDub is multi-core aware no? If that's right then even the entry-level
core2 Duo should do what you want.

A magazine article, or a web site now, will tend to use
benchmarks that emphasize processor performance this way.

(clock_speed * instructions_per_clock) * number_of_cores

What they do, is test multithreaded software. Multithreading works
best in multimedia applications, because a number of problems there
(processing large data sets) benefit from a divide and conquer
algorithm.

For example, Photoshop could split a picture in two pieces, and
a processor core could work on each half of the picture.

But the truth is, activities on a computer consist of a mix
of single threaded ones and multithreaded ones. So a typical
user doesn't see the huge speedup the above equation might
suggest.

[snip]

I use a benchmark designed by Australian PC User magazine (they made it
available to their users on a cover disk) called 'UserBench Encode 2009'
http://darrenyates.com.au/?p=573 that uses a mix of single threaded and
multi threaded work, to simulate real-world use. It only benches
CPU/FSB/RAM. They set the score of 10 to represent a 2GHz Pentium 4 with a
533 FSB running 1GB RAM running XP SP2.

Examples of machines I'm using here, all running XP SP3:

[IBM R51 ThinkPad] Dothan 1.7GHz/400MHz/2GB RAM. Score: 13.82
Flatmate's desktop. E4500 2.2GHz/800/2GB RAM. Score: 29.77
My desktop. E7300 2.66GHz/1066/4GB RAM. Score: 39.45
The above with the FSB raised to 1333. Score 48.81


As it's the same benchmark that my favourite magazine uses to test it's
review machines I find it very useful.

Cheers,

 

[geoff]

Hello,

It seems more complicated than I thought. I have a 2 ghz pentium M laptop,
with 2 gigs of RAM and XP.

It can run single task apps ok but it also has some programming environments
that multi-task.

If one were to build a desktop with:
1. vista 64-bit
2. case
3. PS
4. MB
5. CPU - at least dual core

.. . . (no monitor) that was 4x faster, total price under $500, which CPU and
MB would you use?

--g

 

[Paul]

Hello,

It seems more complicated than I thought. I have a 2 ghz pentium M laptop,
with 2 gigs of RAM and XP.

It can run single task apps ok but it also has some programming environments
that multi-task.

If one were to build a desktop with:
1. vista 64-bit
2. case
3. PS
4. MB
5. CPU - at least dual core

. . . (no monitor) that was 4x faster, total price under $500, which CPU and
MB would you use?

--g

So you're comparing a laptop to a desktop ?

Pentium M at 2GHz, comes in the 755 and 760 models, with different FSB.
It is probably a single core, but you know that better than I do.

http://processorfinder.intel.com/details.aspx?sSpec=SL869

SuperPI 1M is 38.83sec at 2GHz. SuperPI 32M is 38:11 minutes.

http://www.hwbot.org/ResultBrowseByProcessor.do?cpuModelId=365

Compare that to a single core of an E8400 (from my previous post)
SuperPI 1M in 15-16 seconds
SuperPI 32M in 14:10 to 15:59 minutes (say 900 seconds in round numbers)

39/16 = 2.4x
2291/900 = 2.5x

So core to core, a single core of the E8400 is not 4x faster than
the Pentium M 2GHz. You would need a situation and an application
which could use both cores, for the combined effect to be 5x. But
on single threaded applications, you'll come up short. You'd need
to raise the frequency significantly, for the processor to always
be at least 4x faster. The E8400 runs at 3GHz.

If you bought a quad, that will help make the multithreaded situations
even faster, without improving the single threaded ones. On Intel,
the scaling on a quad isn't perfect, due to choking of the FSB.
The caches need to maintain coherence between the two die, so
it is possible there is snoop traffic on the bus. On one of the
multimedia benchmarks known for perfect scaling, the Intel Quad
scales to 3.5x the performance, over a single core. While an
AMD Quad scales to 4x (doesn't choke). The Intel processor
is sufficiently faster than the AMD, that this effect isn't
too important. I haven't seen a Core i7 run on the same
benchmark, but my guess would be the four cores (without
considering HT), would be 4x faster than a single core.

core core core core Q9550, Q9650
| | | | Block Diagram
-+----+- -+----+- Two silicon die, joined inside.
| 6MB L2 | | 6MB L2 |
----+--- ---+----
| |
+-----+------+
|
LGA775 FSB (used for memory access and I/O)

Picking the motherboard isn't an arbitrary exercise. The buyer
has to pick the slot configuration they want, what built-in
peripherals are absolutely essential (firewire?) and so on.
For example, I can find really cheap motherboards with only
two DIMM slots, but who wants that ?

Something in a P45 based board, for around $150, might be
middle of the road. An E8400 at $165, allocate $150 for
a motherboard, say $40 for some DDR2 RAM, and you should be
able to stay under $500. The Q9550 is $270, and is the
cheapest quad with 12MB total cache. You'd still be in
the $500 ballpark.

I used a Core2 Duo 2.6GHz/FSB800 for my upgrade, but I didn't
really save a bundle of money doing it that way. I used a $70
motherboard, for that ghetto touch. I think the hardware was
in the $300 range for the upgrade, and since the motherboard
had an AGP slot, I got to reuse a five year old video card
If I'd wanted to save more, I'd have to go AMD. The AMD
6000+ might be a bit slower, but it's priced at $99.
A 5600+ is $80. So that is one way to shave off a
few bucks. But with $500 to spend, you can do a
bit better than that.

Paul

 

[geoff]

I don't know

what the locality of reference is like in SuperPI, but I would be
a bit suspicious that the benchmark is overestimating the speedup.

I guess I'm not following, if a large cache is part of the CPU design then
it should be considered in whatever speedup it provides.

--g

 

[geoff]

2.00 gigahertz Intel Mobile Pentium 4 - M
8 kilobyte primary memory cache
512 kilobyte secondary memory cache
bus clock: 400 mhz

What kind of non-gaming video card would you recommend for a P45 MB?

--g

 

[Paul]

I guess I'm not following, if a large cache is part of the CPU design then
it should be considered in whatever speedup it provides.

--g

It is true, if you're looking at the processor in isolation,
it is taken into account.

I'm mainly interested in doing it that way (eliminating
the cache influence), so that processors of different cache sizes
can be shown to have some minimum level of performance
available. By "cache busting" the L2, it means I get
to see more effect from the FSB and main memory choices.

In the real world, there are things that benefit from
the cache, and things that don't. There are games that
fall into both camps. When you buy a large cache processor,
what that does is help with some portion of all possible
processing tasks. But it doesn't help with all of them.
If the benchmark only picks out the "big cache", then
it may not represent pathological situations that arise.

I want to be able to tell someone, "your new processor is
*at least* 2.5x faster than your old one". So they
won't be disappointed.

Your clarification on the processor type makes a bit of a
difference. This processor is a different generation
and architecture (Northwood).

2.00 gigahertz Intel Mobile Pentium 4 - M
82-91 seconds for SuperPI 1M 82sec/16sec = a factor of five wrt E8400
78 minutes for SuperPI 32M 4680sec/900sec = a factor of five wrt E8400

http://www.hwbot.org/ResultBrowseByProcessor.do?cpuModelId=1396

So an E8400 would be 5x faster than Mobile P4-M 2GHz when
comparing a single core. For multithreaded applications,
it will be faster than that. A leveling effect, is the
disk speed will be the same in the new system, as in the
old. So doing a search could still be sluggishly slow,
if it is dominated by disk seek time. But things that
are compute bound, should be much better.

Also note, that my simple minded choice above, is to give
ballpark figures for what to expect. There are always
some pathological applications out there, that aren't
going to benefit as much. For example, say an app is
compiled by a particularly bad compiler. Or an application
does nothing but floating point. The speedup
might not be very good at all. In cases like that,
you can try taking the ratio of clock rates, as
a "pessimistic as possible" estimate. So if I was
really clever at sabotaging the E8400, I might be
able to make it only 1.5x faster than your old processor.
But most of the time, it'll be 5x.

HTH,
Paul

 

[DevilsPGD]

In message <[email protected]> kony

I don't know about the app itself, I don't use it for
special effects but as described the task the OP seems to
want to do is decompress various video formats, possibly
resize, and re-encode to MPEG2 for the DVD. I'd suspect
then that the majority of the processing is encoding.

Whether multiple cores are used effectively for that will
have to do with whether the decompressing and/or compressing
codecs are multi-core capable. Virtualdub may be able to
use multiple cores for multiple threads but the video
compression thread tends to be the bottleneck if it's not
using a codec with multi-core capability itself.

To put it another way, with a semi-modern version of Divx
claiming 2 core support, I have been able to keep both CPUs
in a dual core system pegged at 100% using Virtualdub, but I
would not be confident this will happen with all Virtualdub
jobs depending on exactly what they are.

Two-ish cores is also the upper limit, or seems to be with my
configuration anyway. I'm on a quad core system and usually see 55%-60%
CPU utilization that I can attribute to VirtualDub. I did a bit of
digging and the running theory seemed to be that the codec I'm using for
compression was only dual-core capable, but that a third CPU was useful
to decode as well as assemble the resulting file.

I've always been just reencoding the video and doing a direct stream
copy of the audio, it wouldn't surprise me if recompressing audio would
step up the utilization of the third core even further.

 

[JR Weiss]

VirtualDub is multi-core aware no? If that's right then even the entry-level
core2 Duo should do what you want.

The author's web site indicates it is NOT.

I use a benchmark designed by Australian PC User magazine (they made it
available to their users on a cover disk) called 'UserBench Encode 2009'
http://darrenyates.com.au/?p=573 that uses a mix of single threaded and multi
threaded work, to simulate real-world use. It only benches CPU/FSB/RAM. They
set the score of 10 to represent a 2GHz Pentium 4 with a 533 FSB running 1GB
RAM running XP SP2.

Examples of machines I'm using here, all running XP SP3:

[IBM R51 ThinkPad] Dothan 1.7GHz/400MHz/2GB RAM. Score: 13.82
Flatmate's desktop. E4500 2.2GHz/800/2GB RAM. Score: 29.77
My desktop. E7300 2.66GHz/1066/4GB RAM. Score: 39.45
The above with the FSB raised to 1333. Score 48.81

Unfortunately, taking HD I/O out of the equation will tend to grossly overstate
relative performance gains. VirtualDub's author even warns that on a fast CPU
the app can become "disk-bound," and the faster CPU will then work at much less
than full capacity.

 

[JR Weiss]

What kind of non-gaming video card would you recommend for a P45 MB?

Something like an ATI 4670 is relatively inexpensive, cool-running, and not an
energy hog: http://www.newegg.com/Product/Product.aspx?Item=N82E16814161252

I got one to replace an older X1950. it IS cool and quiet!

There are certainly cheaper ones, but I can't vouch for their performance.

 

[geoff]

Hello,

I actually do not see why a newer core would be faster. 45 nm technology,
for example, is the size of the gate or transistor, and more can be placed
on the chip but that, by itself, would not mean faster.

The only two thing I can think of, with regard to the core only, not the
caches, etc. is the fact that they have instructions for moving data 64 bits
at a time and a higher clock speed.

The clock speed limit has been reached and most/many programs are 32 bit,
so, they would not be doing 64-bit register moves.

The only other thing is possibly there are scenarios they can optimize for,
at the cpu level, but that would not make a cpu 4 or 5 times faster.

IMHO.

--g

 

[Rarius]

Hello,

I actually do not see why a newer core would be faster. 45 nm technology,
for example, is the size of the gate or transistor, and more can be placed
on the chip but that, by itself, would not mean faster.

45nm architecture offers more than just more transistors per sq mm.

Firstly the transistors are closer together and the chip is smaller... this
means that the time taken to get signals around the chip is reduced.

Secondly, the architectural techniques improve with each new generation.
This particularly effects things like the efficiency of the cache. Each C2D
core is generally 20-30% faster than a P4 core at the same clock speed.

The only other thing is possibly there are scenarios they can optimize
for, at the cpu level, but that would not make a cpu 4 or 5 times faster.

I agree that 4-5 times as fast is a little excessive. An improvement of
2.5x-3x is more like it. Remember that a C2D has at least 2 cores, the P2
only has one. Given that each core is faster and an improvement in RAM
speed, you have a 3 times improvement over a P4. A C2Q chip with 4 cores
could easily manage 4-5 times the performance of a P4.

Remember that there is ALWAYS more than just one thing going on in a modern
computer. Even if the main application is singlethreaded, the OS, device
drivers, other programs running in the background will use the other cores,
leaving the first core to run 100% on the main app. This in itself will
often result in a 20-30% performance increase. On the other hand, many apps
are limited by things other than the CPU, such as RAM and HDD.

Rarius

 

[DevilsPGD]

In message <[email protected]> "geoff"

I actually do not see why a newer core would be faster. 45 nm technology,
for example, is the size of the gate or transistor, and more can be placed
on the chip but that, by itself, would not mean faster.

The only two thing I can think of, with regard to the core only, not the
caches, etc. is the fact that they have instructions for moving data 64 bits
at a time and a higher clock speed.

The clock speed limit has been reached and most/many programs are 32 bit,
so, they would not be doing 64-bit register moves.

The only other thing is possibly there are scenarios they can optimize for,
at the cpu level, but that would not make a cpu 4 or 5 times faster.

There are a ton of other changes that can be made to improve efficiency.

The change from a 386 to 486 was one of the most staggering, at equal
clock speeds the 486 was twice as fast for many instructions.

The picture is far more complex with modern CPUs, but many operations a
CPU can perform entirely on chip still take more then one clock cycle to
complete, anything that hits the various caches or system RAM will take
many more cycles, this leaves room for substantial improvement.

Some CPUs use predictive branching to move forward while waiting for
data to return, essentially making a guess, if they guessed right on the
results of the slow operation, the delay from the slow operation is
negated. If they guessed wrong, no harm done aside from a bit of energy
consumed.

Compare the performance of a Core 2 core vs an Atom core, for example.
Atom cores perform slower at equal clock speeds vs a Core 2 CPU, but
also have a more favourable watterformance ratio, so they're ideal for
installations where power savings is more important then performance.
One of the big things Atom processors don't do is predictive processing,
instead they offer Hyperthreading, which at least potentially allows a
single core to work on something else while waiting on a slow operation,
but if there is no other real work done, no power is wasted on incorrect
branches.

There is a lot more at play here then raw clock speed, especially when
comparing CPUs optimized for different purposes, the P4 --> Core/Core 2
architecture signaled a shift from raw MHz rates to looking at overall
performance, and to some extent performance-per-watt.