Athlon 64 queries

[Spiro]

Hi,

I'm currently trying to decide between an Athlon 64 3200+ and a P4, 2.8 -
3.0

My main interest is in digital photography and I want something that will
handle large RAW files quickly. Gaming etc is not really a priority. I'll
also be using 1024 Mb of RAM

I've read extensively about the Athlon 64 and can appreciate that when 64
bit applications are here it will come into it's own - but what about now?

My main worry is that, notwithstanding the benefits of memory control
directly on the chip, etc, the Athlon 64 is still only a 2.0 Ghz processor
(albeit with a 1Mb cache)

Will the P4 (either the 2.8 or the 3.0 version) be faster for the type of
work I want to do - in short, is a 2Ghz Athlon as fast (or faster) than a
3.0 Ghx Pentium in non-3D type applications?

Many thanks for your input.

 

[JK]

An Athlon 64 3200+ should perform better than a Pentium 4 2.8 ghz
running 32 bit Photoshop. When a 64 bit version of Photoshop is
available, an Athlon 64 running it should greatly outperform a P4
running 32 bit Photoshop.

"My main worry is that, notwithstanding the benefits of memory control
directly on the chip, etc, the Athlon 64 is still only a 2.0 Ghz processor
(albeit with a 1Mb cache)"

Clock speed is meaningless for gauging relative performance except when
comparing chips that are otherwise the same. A good analogy would
be comparing a centipede to a race horse. A centipede can take many more steps
per minute than a race horse, but you know that a race horse can travel much
further in a minute than a centipede.

Hi,

I'm currently trying to decide between an Athlon 64 3200+ and a P4, 2.8 -
3.0

My main interest is in digital photography and I want something that will
handle large RAW files quickly.

In that case you might want to use very fast hard drives, and more than a gig
of ram.

 

[Spiro]

An Athlon 64 3200+ should perform better than a Pentium 4 2.8 ghz
running 32 bit Photoshop. When a 64 bit version of Photoshop is
available, an Athlon 64 running it should greatly outperform a P4
running 32 bit Photoshop.

"My main worry is that, notwithstanding the benefits of memory control
directly on the chip, etc, the Athlon 64 is still only a 2.0 Ghz processor
(albeit with a 1Mb cache)"

Clock speed is meaningless for gauging relative performance except when
comparing chips that are otherwise the same. A good analogy would
be comparing a centipede to a race horse. A centipede can take many more
steps
per minute than a race horse, but you know that a race horse can travel
much
further in a minute than a centipede.

Thanks for that - can I now throw problem #2 at you?

I decided to go ahead with the AMD system but have discovered that the 3200+
processor now ships in two versions -

[1] 1024 level 2 cache, with a clock speed of 2.0Ghz

and

[2] 512 level 2 cache, with a clock speed of 2.2Ghz (which is the version
immediately available to me)


Having spoken to AMD they inform me that revisions to the core have
'allowed' them to reduce the cache (not sure why they's wnat to do that) and
by increasing the speed slightly they have maintained the benchmark scores,
and can still call it a 3200 chip.

I'm a bit dubious about this - can a 50% reduction in L2 cache be
compensated by the small increase in speed from 2.0Ghz to 2.2Ghz?

One of my reasons for choosing the AMD was the 1024 l2 cache but I'm now
very hesitant about buying it with the reduced cache - what's your opinion?

Thanks

 

[Alan Walpool]

Spiro> Thanks for that - can I now throw problem #2 at you?

Spiro> I decided to go ahead with the AMD system but have discovered
Spiro> that the 3200+ processor now ships in two versions -

Spiro> [1] 1024 level 2 cache, with a clock speed of 2.0Ghz

Spiro> and

Spiro> [2] 512 level 2 cache, with a clock speed of 2.2Ghz (which is
Spiro> the version immediately available to me)


Spiro> Having spoken to AMD they inform me that revisions to the core
Spiro> have 'allowed' them to reduce the cache (not sure why they's
Spiro> wnat to do that) and by increasing the speed slightly they
Spiro> have maintained the benchmark scores, and can still call it a
Spiro> 3200 chip.

Spiro> I'm a bit dubious about this - can a 50% reduction in L2 cache
Spiro> be compensated by the small increase in speed from 2.0Ghz to
Spiro> 2.2Ghz?

Spiro> One of my reasons for choosing the AMD was the 1024 l2 cache
Spiro> but I'm now very hesitant about buying it with the reduced
Spiro> cache - what's your opinion?

Spiro> Thanks

This issue has been discussed for some time, and I have asked the same
questions. Find a web site that have benchmarked both processors and
benchmarking sites report the reduction in cache is insignificant on
benchmark scores. The question that should be asked will you see the
difference in performance and the answer is no.

I purchased the 512k cache after asking the same question and saved a
few bucks. Anyway it is hard to find the 1024k cache 3200+. I sure
someone has them if you look hard enough.

Good luck, and report how your system build goes regardless if it is
good or bad.

Later,

Alan

 

[Spiro]

Spiro> Thanks for that - can I now throw problem #2 at you?

Spiro> I decided to go ahead with the AMD system but have discovered
Spiro> that the 3200+ processor now ships in two versions -

Spiro> [1] 1024 level 2 cache, with a clock speed of 2.0Ghz

Spiro> and

Spiro> [2] 512 level 2 cache, with a clock speed of 2.2Ghz (which is
Spiro> the version immediately available to me)


Spiro> Having spoken to AMD they inform me that revisions to the core
Spiro> have 'allowed' them to reduce the cache (not sure why they's
Spiro> wnat to do that) and by increasing the speed slightly they
Spiro> have maintained the benchmark scores, and can still call it a
Spiro> 3200 chip.

Spiro> I'm a bit dubious about this - can a 50% reduction in L2 cache
Spiro> be compensated by the small increase in speed from 2.0Ghz to
Spiro> 2.2Ghz?

Spiro> One of my reasons for choosing the AMD was the 1024 l2 cache
Spiro> but I'm now very hesitant about buying it with the reduced
Spiro> cache - what's your opinion?

Spiro> Thanks

This issue has been discussed for some time, and I have asked the same
questions. Find a web site that have benchmarked both processors and
benchmarking sites report the reduction in cache is insignificant on
benchmark scores. The question that should be asked will you see the
difference in performance and the answer is no.

I purchased the 512k cache after asking the same question and saved a
few bucks. Anyway it is hard to find the 1024k cache 3200+. I sure
someone has them if you look hard enough.

Good luck, and report how your system build goes regardless if it is
good or bad.

Later,

Alan

Thanks for your input.

I realise that we're talking miliseconds in speed terms but my worry is that
the reduced cache will adversly impact on manipulating large graphic files.

Did they reduce the cache as a cost saving exercise?

Finally, are you pleased with your own choice (and do you photo edit with
it?)

regards

 

[Alan Walpool]

Spiro> Thanks for your input.

Spiro> I realise that we're talking miliseconds in speed terms but my
Spiro> worry is that the reduced cache will adversly impact on
Spiro> manipulating large graphic files.

I use it for large graphic files (largest so far 12MB), and also edit
movies and then burn DVD's. No problems so far. I guess my computer is
a general use computer for standard computing tasks. I play no games.

Spiro> Did they reduce the cache as a cost saving exercise?

Yes and it is easier to make I have read.

Spiro> Finally, are you pleased with your own choice (and do you
Spiro> photo edit with it?)

No problem so far. Really you made me think about this issue in
another view point. I don't know the answer, but with such large files
it is unlikely that changing the cache from 512 kilobytes to 1024
kilobytes will make a difference. Sounds like random access memory
would be the big issue with such large files. I bet that a bigger
cache would probably help in some circumstances but I'm not
knowledgeable enough to comment on this question. I have read that
large caches are essential for servers, but I don't know the reason
why. Does anyone have a link where one could read about this issue.

I have a duron with 64 kilobyte cache and it was really fast at one
time. ;-).

I have read in this group that the 512k cache version overclocks
better but I don't plan to overclock. I have burned up my share of
processors ;-). That is for folks with cash to burn.

Good luck, and report back success or failure.

Alan

 

[JK]

An Athlon 64 3200+ should perform better than a Pentium 4 2.8 ghz
running 32 bit Photoshop. When a 64 bit version of Photoshop is
available, an Athlon 64 running it should greatly outperform a P4
running 32 bit Photoshop.

"My main worry is that, notwithstanding the benefits of memory control
directly on the chip, etc, the Athlon 64 is still only a 2.0 Ghz processor
(albeit with a 1Mb cache)"

Clock speed is meaningless for gauging relative performance except when
comparing chips that are otherwise the same. A good analogy would
be comparing a centipede to a race horse. A centipede can take many more
steps
per minute than a race horse, but you know that a race horse can travel
much
further in a minute than a centipede.

Thanks for that - can I now throw problem #2 at you?

I decided to go ahead with the AMD system but have discovered that the 3200+
processor now ships in two versions -

[1] 1024 level 2 cache, with a clock speed of 2.0Ghz

and

[2] 512 level 2 cache, with a clock speed of 2.2Ghz (which is the version
immediately available to me)

Having spoken to AMD they inform me that revisions to the core have
'allowed' them to reduce the cache (not sure why they's wnat to do that) and
by increasing the speed slightly they have maintained the benchmark scores,
and can still call it a 3200 chip.

I'm a bit dubious about this - can a 50% reduction in L2 cache be
compensated by the small increase in speed from 2.0Ghz to 2.2Ghz?

It can on average, since the Athlon 64 is an efficiently designed chip with
a relatively short pipeline.

One of my reasons for choosing the AMD was the 1024 l2 cache but I'm now
very hesitant about buying it with the reduced cache - what's your opinion?

http://forums.pimprig.com/archive/topic/20627.html

There is much more discussion about it.

http://www.google.com/search?hl=en&ie=UTF-8&q=Clawhammer+vs+Newcastle+benchmarks&btnG=Google+Search

It seems like the newer Newcastle chips at the same speed
rating are on average slightly better performers than the older
Clawhammer chips. The differences aren't that great though,
and for the software where the Clawhammer is faster, it usually
is only faster by a small amount.

 

[Elaine]

They were able to reduce the cache size without impacting performance.
The reason is that the newer Socket 939 CPUs use DDR (double data rate)
memory
at its full dual channel double data rate speed. The older Socket 940 chips
could use the same DDR memory but they only used it in single channel mode
or half speed. The upshot of this is that they could cut the cache in half
and not hurt performance. All the benchmarks for memory tests indicate that
the newer chip is much faster on memory intensive applications.
The difference in systems built with the new Socket 939 CPU is that they
require pairs of DIMMs since they are making use of the dual channel
capabilities. The older Socket 940 could take the same DDR memory DIMMs but
without pairing them. So the older systems could use one, two or three
DIMMs instead of pairs of DIMMs.

HTH

"Spiro" == Spiro <[email protected]> writes:

An Athlon 64 3200+ should perform better than a Pentium 4 2.8 ghz
running 32 bit Photoshop. When a 64 bit version of Photoshop is
available, an Athlon 64 running it should greatly outperform a P4
running 32 bit Photoshop.

"My main worry is that, notwithstanding the benefits of memory
control directly on the chip, etc, the Athlon 64 is still only a
2.0 Ghz processor (albeit with a 1Mb cache)"

Clock speed is meaningless for gauging relative performance except
when comparing chips that are otherwise the same. A good analogy
would be comparing a centipede to a race horse. A centipede can
take many more steps per minute than a race horse, but you know
that a race horse can travel much further in a minute than a
centipede.

Spiro> Thanks for that - can I now throw problem #2 at you?

Spiro> I decided to go ahead with the AMD system but have discovered
Spiro> that the 3200+ processor now ships in two versions -

Spiro> [1] 1024 level 2 cache, with a clock speed of 2.0Ghz

Spiro> and

Spiro> [2] 512 level 2 cache, with a clock speed of 2.2Ghz (which is
Spiro> the version immediately available to me)


Spiro> Having spoken to AMD they inform me that revisions to the core
Spiro> have 'allowed' them to reduce the cache (not sure why they's
Spiro> wnat to do that) and by increasing the speed slightly they
Spiro> have maintained the benchmark scores, and can still call it a
Spiro> 3200 chip.

Spiro> I'm a bit dubious about this - can a 50% reduction in L2 cache
Spiro> be compensated by the small increase in speed from 2.0Ghz to
Spiro> 2.2Ghz?

Spiro> One of my reasons for choosing the AMD was the 1024 l2 cache
Spiro> but I'm now very hesitant about buying it with the reduced
Spiro> cache - what's your opinion?

Spiro> Thanks

This issue has been discussed for some time, and I have asked the same
questions. Find a web site that have benchmarked both processors and
benchmarking sites report the reduction in cache is insignificant on
benchmark scores. The question that should be asked will you see the
difference in performance and the answer is no.

I purchased the 512k cache after asking the same question and saved a
few bucks. Anyway it is hard to find the 1024k cache 3200+. I sure
someone has them if you look hard enough.

Good luck, and report how your system build goes regardless if it is
good or bad.

Later,

Alan

Thanks for your input.

I realise that we're talking miliseconds in speed terms but my worry is
that the reduced cache will adversly impact on manipulating large graphic
files.

Did they reduce the cache as a cost saving exercise?

Finally, are you pleased with your own choice (and do you photo edit with
it?)

regards

 

[Alan Walpool]

Elaine> They were able to reduce the cache size without impacting
Elaine> performance. The reason is that the newer Socket 939 CPUs use
Elaine> DDR (double data rate) memory at its full dual channel double
Elaine> data rate speed. The older Socket 940 chips could use the
Elaine> same DDR memory but they only used it in single channel mode
Elaine> or half speed. The upshot of this is that they could cut the
Elaine> cache in half and not hurt performance. All the benchmarks
Elaine> for memory tests indicate that the newer chip is much faster
Elaine> on memory intensive applications. The difference in systems
Elaine> built with the new Socket 939 CPU is that they require pairs
Elaine> of DIMMs since they are making use of the dual channel
Elaine> capabilities. The older Socket 940 could take the same DDR
Elaine> memory DIMMs but without pairing them. So the older systems
Elaine> could use one, two or three DIMMs instead of pairs of DIMMs.

This does not explain why there is very little difference in
performance an 754 which is a single channel design.

Later,

Alan

 

[Wes Newell]

They were able to reduce the cache size without impacting performance.
The reason is that the newer Socket 939 CPUs use DDR (double data rate)
memory
at its full dual channel double data rate speed. The older Socket 940 chips
could use the same DDR memory but they only used it in single channel mode
or half speed. The upshot of this is that they could cut the cache in half
and not hurt performance. All the benchmarks for memory tests indicate that
the newer chip is much faster on memory intensive applications.
The difference in systems built with the new Socket 939 CPU is that they
require pairs of DIMMs since they are making use of the dual channel
capabilities. The older Socket 940 could take the same DDR memory DIMMs but
without pairing them. So the older systems could use one, two or three
DIMMs instead of pairs of DIMMs.

I don't know how you came up with this, but it's totally wrong. Both use
DDR. Both support dual channel, and both will also work single channel..
Th main difference is that 939 uses unbuffered standard DDR ram while 940
requires registered DDR ram, which is a little slower, but also a lot
safer in terms of data integrity. Now socket 754 is single channel, and
contrary to popular belief is just about as fast as 939 using dual
channel and the same speed cpu. There's been way too much hype put on dual
channel. While in memory bandwidth benchmark test there's a big
difference, in real applications there's little to none. You can see this
in benchmark test on the web in many places. And that probably why AMD
chooses not to produce socket 754 cpu's that run at the faster speeds of
the 939/940 cpu's.

 

[Elaine]

I stand corrected. My comparison on the single channel versus dual channel
chips should have been to Socket 754 not 940. You are correct, the 940 used
ECC registered memory which is slightly slower and more expensive than
non-registered memory. After having an old 2.0GHz P4 system with RDRAM
memory, I did not want to get stuck with another system that required
expensive, hard to find memory. I considered all three Socket types before
purchasing Socket 939.

I do not have a Socket 754 or a 940 machine to do an actual side by side
speed comparison on. The quote on speed was from widely published memory
benchmark results available on many reputable web pages. Synthetic memory
benchmarks are focused on measuring just the speed of memory not actual
application performance. But in point of fact dual channel DDR is
physically twice as fast as single channel DDR.
I refer to Corsair Memory's web page which has an excellent tutorial on the
subject of double channel versus single channel memory at:
http://www.corsairmemory.com/corsair/products/tech/memory_basics/153707/index.html

Apparently, you are partial to a 754. If that works for you, great. There
are some differences in the number of DIMMs that can be accessed on a 754.
According to the Athlon64 Socket 754 and 939 tech specs on AMD's web page
(http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/24659.PDF)
a 754 chip is limited to accessing memory from 3 slots. The 939 chip is
limited to 4 DIMM slots. Socket 939 does require pairs of DIMMS while
Socket 754 does not require pairs of DIMMs due to the single channel
architecture. Given the same capacity memory DIMMs obviously, you can put
more memory on a system with 4 DIMM slots than you can on a system with just
3 DIMM slots. With the advent of 64-bit OS's bringing the increased
potential for memory address space to 16TB, the extra memory slot is useful.

Whether or not the increase in memory speed helps performance is of course
totally application and user dependent. If you see no difference in
performance between socket 754 and socket 939, and you don't need the
potentially larger memory capacity available on a Socket 939 based system,
then by all means go for the more cost effective 754. I run CPU and memory
intensive applications and prefer the fastest memory performance and the
largest memory capacity available.

The original poster's question was could they reduce the cache size and
still get good performance. The answer is yes. Did they do it as cost
reduction? Absolutely.

I have a Socket 939, Athlon64 3800. The overall performance of this machine
with 10K RPM WD Raptors in a dual SATA RAID configuration is approximately
2.4 times that of a P4 2.8GHz with a single 7200 RPM SATA drive. The
performance number is based on compiling approximately one million lines of
C++ code. Your mileage and performance on any machine will vary with the
application in use.

 

[Elaine]

Correction. Substitute Socket 754 for Socket 940 in the message below for
single channel versus dual channel memory. Socket 940 is the one with the
slower, more expensive registered ECC memory.

Sorry for the confusion.

They were able to reduce the cache size without impacting performance.
The reason is that the newer Socket 939 CPUs use DDR (double data rate)
memory
at its full dual channel double data rate speed. The older Socket 940
chips could use the same DDR memory but they only used it in single
channel mode or half speed. The upshot of this is that they could cut the
cache in half and not hurt performance. All the benchmarks for memory
tests indicate that the newer chip is much faster on memory intensive
applications.
The difference in systems built with the new Socket 939 CPU is that they
require pairs of DIMMs since they are making use of the dual channel
capabilities. The older Socket 940 could take the same DDR memory DIMMs
but without pairing them. So the older systems could use one, two or
three DIMMs instead of pairs of DIMMs.

HTH

Spiro> An Athlon 64 3200+ should perform better than a Pentium 4 2.8 ghz
running 32 bit Photoshop. When a 64 bit version of Photoshop is
available, an Athlon 64 running it should greatly outperform a P4
running 32 bit Photoshop.

"My main worry is that, notwithstanding the benefits of memory
control directly on the chip, etc, the Athlon 64 is still only a
2.0 Ghz processor (albeit with a 1Mb cache)"

Clock speed is meaningless for gauging relative performance except
when comparing chips that are otherwise the same. A good analogy
would be comparing a centipede to a race horse. A centipede can
take many more steps per minute than a race horse, but you know
that a race horse can travel much further in a minute than a
centipede.



Spiro> Thanks for that - can I now throw problem #2 at you?

Spiro> I decided to go ahead with the AMD system but have discovered
Spiro> that the 3200+ processor now ships in two versions -

Spiro> [1] 1024 level 2 cache, with a clock speed of 2.0Ghz

Spiro> and

Spiro> [2] 512 level 2 cache, with a clock speed of 2.2Ghz (which is
Spiro> the version immediately available to me)


Spiro> Having spoken to AMD they inform me that revisions to the core
Spiro> have 'allowed' them to reduce the cache (not sure why they's
Spiro> wnat to do that) and by increasing the speed slightly they
Spiro> have maintained the benchmark scores, and can still call it a
Spiro> 3200 chip.

Spiro> I'm a bit dubious about this - can a 50% reduction in L2 cache
Spiro> be compensated by the small increase in speed from 2.0Ghz to
Spiro> 2.2Ghz?

Spiro> One of my reasons for choosing the AMD was the 1024 l2 cache
Spiro> but I'm now very hesitant about buying it with the reduced
Spiro> cache - what's your opinion?

Spiro> Thanks

This issue has been discussed for some time, and I have asked the same
questions. Find a web site that have benchmarked both processors and
benchmarking sites report the reduction in cache is insignificant on
benchmark scores. The question that should be asked will you see the
difference in performance and the answer is no.

I purchased the 512k cache after asking the same question and saved a
few bucks. Anyway it is hard to find the 1024k cache 3200+. I sure
someone has them if you look hard enough.

Good luck, and report how your system build goes regardless if it is
good or bad.

Later,

Alan

Thanks for your input.

I realise that we're talking miliseconds in speed terms but my worry is
that the reduced cache will adversly impact on manipulating large graphic
files.

Did they reduce the cache as a cost saving exercise?

Finally, are you pleased with your own choice (and do you photo edit with
it?)

regards