Need to know more about the different lines of processors

[Yugo]

Tom's Hardware, I suppose -- that AMD was reshaping its processors with less
cache to meet Intel's prices. Isn't this the cause of the price drop? Still,
according to Bazzy Smith's link -- Thanks Bazzy! I've bookmarked this link --
AMD scores best for the first 11 processors.run a Celeron 850 with 256 MB ram. On Slackware Linux, it's a real mighty!
OTOH, until then, I might find something interesting in the garbage...which I'm not? It's so good not feeling that urge to upgrade like in the bad
old days. Man, that was costly!

Dont go by that chart for a future upgrade unless by future you dont
mean too far off.

Was: Price differentials in AMd processors are out of control

Tom's Hardware, I suppose -- that AMD was reshaping its processors with less
cache to meet Intel's prices. Isn't this the cause of the price drop? Still,
according to Bazzy Smith's link -- Thanks Bazzy! I've bookmarked this link --
AMD scores best for the first 11 processors.run a Celeron 850 with 256 MB ram. On Slackware Linux, it's a real mighty!
OTOH, until then, I might find something interesting in the garbage...which I'm not? It's so good not feeling that urge to upgrade like in the bad
old days. Man, that was costly!

Dont go by that chart for a future upgrade unless by future you dont
mean too far off.

I thought «that chart» was updated regularly. So, my final choice might not be
an AMD 64 3500+. Just before I buy a new computer, I read reviews, listen to
what "some" salesmen say, and make a decison on a 100$ processor mainly for
word processing and cropping a few pictures to use as backgrounds. If I was
using databases, editing film... or playing games, it would be a whole
different story.

Ten years ago, I was more aware of the hardware scene. It was easier because
there weren't that many different lines of processors and I had to keep an
interest in the matter since a new processor + more faster memory was always a
blessing. But today... I couldn't care less if Vista required 1 G of RAM, I'll
never use that crap! I wouldn't use it if they paid me!

Still, sometimes, I'd like to understand better what's going on on the
processor scene. I know 64 bit processors have arrived on the popular market a
few years ago and that, nowadays, it's better to buy a 64 bit processor. I
know some basics: what a register is, what cache is, what a dual core
processor is, what 65 nanometer means vs 90 nm. Still, because I don't read
much on the subject, I'm quite confused.

For instance, AMD used to offer only Athlons, then Semprons, CPUs. Now they
have an FX, an AM2 and what, an X series? Trying to sort out the processors of
only one manufacturer is really a pain. They seem much more interested in
serving you a big publicity hoopla than the bare basics.

So, before following your exposé on 754s, 939s, Conroes E6300, D805s, X2s,
AM2s, I would need to know a little bit about the different lines of
processors. You know, what's the conceptual scheme behind them. Is there any
place I can find this information or can you give a *short* description of the
different lines of processors. Don't insist too much on OC capabilities, I'm
really not much into this.

 

[Paul]

So, before following your exposé on 754s, 939s, Conroes E6300, D805s, X2s,
AM2s, I would need to know a little bit about the different lines of
processors. You know, what's the conceptual scheme behind them. Is there any
place I can find this information or can you give a *short* description of the
different lines of processors. Don't insist too much on OC capabilities, I'm
really not much into this.

Intel:

Memory controller is on motherboard Northbridge chip. Motherboard
chipset type determines memory type. Some motherboards, with 915 chipset,
even offered the usage of both DDR and DDR2 memory sticks. Most
modern chipsets for Intel are dual channel, with only the cheapest
of motherboards supporting single channel memory operation. Memory
performance somewhat more important than for AMD. Pentium-M, Core Duo,
Core Duo2 (Conroe) behave more like Athlon64 processors in terms of
clock rate and memory dependence. The Intel processors with "blowout
pricing", are the older hotter running Netburst (P4) stock.

S478 Northwood 512KB cache 0.13u - moderately warm
S478 Prescott 1MB cache 90nm - hot
S775 Various netburst processors, some single, some dual core,
90nm are hotter than 65nm. Max cache was 2x2MB on some dual cores.
D805 is one of the hotter dual core offerings. The processor
price is low, but the cooling cost can be higher, plus need
good mobo and power supply if overclocking.
S775 Conroe family. Higher IPC (instructions per clock cycle) than
Netburst, so not as high a clock rate is needed. High end Conroe
much faster than current AMD high end. Cooler running than previous
offerings.

Celeron, Celeron D - Intel economy version, least cache included.
- Less performance than other members of Pentium 4.
- Compare benchmarks to be sure.
Pentium 4 & schlock - More cache than Celeron. Hyperthreading feature
makes for smoother desktop performance than
Athlons that are single core.

http://processorfinder.intel.com (for clock rate and cache details)

AMD:

Memory controller is inside the processor. Processor type determines
what kind of memory is used.

S754 - single channel memory controller
- performance not as memory dependent as Intel
- every bit as good as S939 processors, but limited
number of usable DIMM slots
- memory type is DDR
S939 - dual channel memory controller
- two or four sticks of RAM are usable, with two preferred
for overclocking or performance configurations
- memory type is DDR
AM2 - core reasonably similar to others, so buying these is not
currently for performance
- dual channel memory controller, four slots, can use two
or four sticks as useful config
- memory type is DDR2

Sempron - AMD economy version, least cache included
- P.R. rating, like 3500+, compares to Intel Celeron D 3.5GHz
Athlon64 - main stream
- P.R. rating, like 3500+, compares to Pentium 4 3.5GHz
and since Pentium 4 is faster than Celeron at same clock
rate, you should not compare Sempron P.R. numbers to
the rest of the Athlon64 family.
Athlon64FX - enthusiast, single core, unlocked multiplier
- also some dual core versions with FX moniker
Athlon64 X2 - dual core, like two Athlon64, with various amounts
of cache, depending on whims of AMD and marketing
pressures.

The Performance Rating system attempts to take into account the
clock rate and amount of cache. The P.R. system has a much tougher
job when evaluating dual core processors, since you cannot compare
the usage pattern of a single core to a dual core. Some people will
get more use from a dual core than others. When comparing X2, take
the clock rate and amount of cache, and compare to the single core
Athlon64, to determine roughly how each core would perform.

http://www.amdcompare.com (for clock rate and cache details)

In terms of clock rates (just very rough numbers, not flame war material):

Pentium4 at 4GHz = Athlon64 at 2.5GHz = Conroe at 2.2GHz

the general idea being, that the latest stuff with the lower clock
rates, is quite effective when compared to the Pentium4 family.

Generally, for finer resolution about performance, you are
best advised to find a web site with benchmarks. Since this
is an overview, I haven't wasted hours digging up benchmarks,
as that is too time consuming.

Right now, price is a reasonably good indicator of relative
performance, at least if reviewing prices on the largest of
retailers. Some of the smaller retailers are still trying to
get the big bucks for old stock.

Paul

 

[Bazzer Smith]

Tom's Hardware, I suppose -- that AMD was reshaping its processors with
less cache to meet Intel's prices. Isn't this the cause of the price drop?
Still, according to Bazzy Smith's link -- Thanks Bazzy! I've bookmarked
this link --
AMD scores best for the first 11 processors.
run a Celeron 850 with 256 MB ram. On Slackware Linux, it's a real mighty!
OTOH, until then, I might find something interesting in the garbage...
which I'm not? It's so good not feeling that urge to upgrade like in the
bad old days. Man, that was costly!


Was: Price differentials in AMd processors are out of control


Tom's Hardware, I suppose -- that AMD was reshaping its processors with
less cache to meet Intel's prices. Isn't this the cause of the price drop?
Still, according to Bazzy Smith's link -- Thanks Bazzy! I've bookmarked
this link --
AMD scores best for the first 11 processors.
run a Celeron 850 with 256 MB ram. On Slackware Linux, it's a real mighty!
OTOH, until then, I might find something interesting in the garbage...
which I'm not? It's so good not feeling that urge to upgrade like in the
bad old days. Man, that was costly!

I thought «that chart» was updated regularly. So, my final choice might
not be an AMD 64 3500+. Just before I buy a new computer, I read reviews,
listen to what "some" salesmen say, and make a decison on a 100$ processor
mainly for word processing and cropping a few pictures to use as
backgrounds. If I was using databases, editing film... or playing games,
it would be a whole different story.

Ten years ago, I was more aware of the hardware scene. It was easier
because there weren't that many different lines of processors and I had to
keep an interest in the matter since a new processor + more faster memory
was always a blessing. But today... I couldn't care less if Vista required
1 G of RAM, I'll never use that crap! I wouldn't use it if they paid me!

Still, sometimes, I'd like to understand better what's going on on the
processor scene. I know 64 bit processors have arrived on the popular
market a few years ago and that, nowadays, it's better to buy a 64 bit
processor. I know some basics: what a register is, what cache is, what a
dual core processor is, what 65 nanometer means vs 90 nm. Still, because I
don't read much on the subject, I'm quite confused.

For instance, AMD used to offer only Athlons, then Semprons, CPUs. Now
they have an FX, an AM2 and what, an X series? Trying to sort out the
processors of only one manufacturer is really a pain. They seem much more
interested in serving you a big publicity hoopla than the bare basics.

So, before following your exposé on 754s, 939s, Conroes E6300, D805s, X2s,
AM2s, I would need to know a little bit about the different lines of
processors. You know, what's the conceptual scheme behind them. Is there
any place I can find this information or can you give a *short*
description of the different lines of processors. Don't insist too much on
OC capabilities, I'm really not much into this.

It certaintly can be hard work trying to keep up with modern processors
especially if you have not been following the market.
Anyway using the amdcheck link
http://www.amdcompare.com/us-en/desktop/Default.aspx
It seems you should look for one which fits in an AM2 socket as that
is the most upgradeable, so it the actual mother boaard in the computer
which is the most important, you don't want to end up with a PC you can't
upgrade, for example my recent purchase had a 939 socket which is more
upgradable then socket 754 machines which sold for the same price (often
more).
It 'only' has a Sempron 3000 in it, but it supports AMD Athlon 64 X2 (up to
4800+)
a processor which can perform 3-4 times better, however they currently cost
a small fortune
probably as much as I paid for the entire PC (base unit) at the time.
However in a few
years time when I might want to upgrade I expect the prices will have fallen
considerably.
So my advice would be to look for the most upgradeable PC motherboard,
probably
AM2 in your case, get a 'basic' CPU say 3500 or whatever you feel happy
with,
you certaintly won't find it lacking in power,
and then you can upgrade cheaply in a few years time when you have had time
to
read all the 'bumf' on CPU's etc.....
http://tomshardware.co.uk/cpu/index.html

Tons of stuff there!!
It says Intel is ahead there, but the two chip makes tend to leap frog each
other
with their latest offering.

Conclusion of that article.
http://tomshardware.co.uk/2006/07/14/core2_duo_knocks_out_athlon_64_uk/page18.html

 

[Yugo]

Paul wrote:

the rest of the Athlon64 family.

Athlon64FX - enthusiast, single core, unlocked multiplier
- also some dual core versions with FX moniker

Unlocked multiplier means it's OK for overclocking? Are the Semprons "unlocked
too"?

Athlon64 X2 - dual core, like two Athlon64, with various amounts
of cache, depending on whims of AMD and marketing
pressures.

The Performance Rating system attempts to take into account the
clock rate and amount of cache. The P.R. system has a much tougher
job when evaluating dual core processors, since you cannot compare
the usage pattern of a single core to a dual core. Some people will
get more use from a dual core than others. When comparing X2, take
the clock rate and amount of cache, and compare to the single core
Athlon64, to determine roughly how each core would perform.

http://www.amdcompare.com (for clock rate and cache details)

OK Let's say I go here:

http://www.amdcompare.com/us-en/desktop/default.aspx

It seems I can get an Athlon 64 3500+ with two sockets: 939 and Am2. Bazzer,
below, says I should go for an AM2 socket. Why? Also, how come you can have
the same processor in different wattages, for the same 90 nm dye (is this the
correct term?)

What does "Stepping" mean?

 

[Yugo]

It certaintly can be hard work trying to keep up with modern processors
especially if you have not been following the market.
Anyway using the amdcheck link
http://www.amdcompare.com/us-en/desktop/Default.aspx

It seems you should look for one which fits in an AM2 socket as that
is the most upgradeable

To tell you the truth, I don't believe I'll ever upgrade a CPU. I certainly
never did until now. I figure a mobo lasts about 7 years and the CPU should
perform fast enough for me for that long. Would there be any advantage in
choosing an 939 socket over an AM2? Not the price, I suppose, since processors
fitting AM2 sockets seem to have gone down a lot very recently, an Athlon 64
3500+, for instance, having gone from $370 (CAN) to $115. But is it really
exactly the same processor or did they remove half the cache (1 MB to 512 KB) ?

Conclusion of that article.
http://tomshardware.co.uk/2006/07/14/core2_duo_knocks_out_athlon_64_uk/page18.html

I had already read this conclusion yesterday. I'm not very much into reading
18 page on a new kind of CPU, mainly that I'd never pay for a dual core now.

See my questions to Paul at the top.

 

[Paul]

Paul wrote:

the rest of the Athlon64 family.

Unlocked multiplier means it's OK for overclocking? Are the Semprons "unlocked

OK Let's say I go here:

http://www.amdcompare.com/us-en/desktop/default.aspx

It seems I can get an Athlon 64 3500+ with two sockets: 939 and Am2. Bazzer,
below, says I should go for an AM2 socket. Why? Also, how come you can have
the same processor in different wattages, for the same 90 nm dye (is this the
correct term?)

What does "Stepping" mean?

"Athlon64 FX = Unlocked Multipliers"
http://www.pcstats.com/ArtVNL.cfm?articleid=1918&page=4

The difference - AM2 socket uses DDR2 RAM, and the older S939
uses DDR RAM. AM2 boards are still having issues with various
brands of enthusiast DDR2 RAM, but AM2 is also the wave of
the future. If you never plan on upgrading the S939 motherboard,
then it would also be a viable alternative. S939 is a mature
technology.

Paul

 

[Yugo]

"Athlon64 FX = Unlocked Multipliers"
http://www.pcstats.com/ArtVNL.cfm?articleid=1918&page=4

The difference - AM2 socket uses DDR2 RAM, and the older S939
uses DDR RAM. AM2 boards are still having issues with various
brands of enthusiast DDR2 RAM, but AM2 is also the wave of
the future. If you never plan on upgrading the S939 motherboard,
then it would also be a viable alternative. S939 is a mature
technology.

Thanks Paul for those late evening answers. So two questions are left:

Also, how come you can have the same processor in different wattages, for the
same 90 nm dye (is this the correct term?)

What does "Stepping" mean?

 

[Bazzer Smith]

To tell you the truth, I don't believe I'll ever upgrade a CPU. I
certainly never did until now. I figure a mobo lasts about 7 years and the
CPU should perform fast enough for me for that long.

Well I had a computer which lasted me about 7 years, it was not possible to
upgrade the CPU as it was the max CPU for the mobo, if i could have put in a
CPU
2 or 3 times as fast, I might have entended it's life a little, or at least
it might have
been of more use to me today. But you are probably riight in a way, but
there is
no harm in having some upgradeability in the computer, especially if it's
actually
costs you nothing.

Would there be any advantage in choosing an 939 socket over an AM2?

No but there were no reasonably priced AM2 machines when I purchased.

Not the price, I suppose, since processors fitting AM2 sockets seem to
have gone down a lot very recently, an Athlon 64 3500+, for instance,
having gone from $370 (CAN) to $115. But is it really exactly the same
processor or did they remove half the cache (1 MB to 512 KB) ?

I had already read this conclusion yesterday. I'm not very much into
reading 18 page on a new kind of CPU, mainly that I'd never pay for a dual
core now.

But I could use a dual core in mine when the price drops, it would extend
the machines
life for a few years perhaps.

 

[Paul]

Thanks Paul for those late evening answers. So two questions are left:

Also, how come you can have the same processor in different wattages, for the
same 90 nm dye (is this the correct term?)

What does "Stepping" mean?

Generally, companies do not like to make a ton of different
silicon dies, for any chip product. To make a processor die might
cost $1 million for a mask set for the fab (probably even more
than that, because I haven't seen figures for what it actually
costs now).

What you can do at the factory, is sort the processors into "bins".
The processors can be tested at various voltages and frequencies.
The results give what is called a SHMOO plot. For example, with
AMD Cool and Quiet, all the FID/VID values used by the device
must be tested on the chip tester, in order for the chip to be
suitable for its "bin". (That is so AMD can guarantee the chip
will work, when the frequency is changed dynamically while the
system is running.)

http://en.wikipedia.org/wiki/Schmoo_plot

So, to make processors with different wattages, you can sort through
all the chips, and take the ones with low power consumption, and
make those your "low power" devices. If the difference between
the regular and low power devices is only one wattage step, then
that will work out OK. But it also means that the regular devices
will all tend to be the warmer ones, so you won't get "lucky" and
get a lower power one by accident, when binning is being used.
Since the low power ones might be harder to find, that might be
reflected in the asking price - but it is also equally likely that
the demand for the low power ones, is what makes the marketing
department set the price a little bit higher.

I don't think it is too practical to make production changes
to the wafers, to make some of the devices use less power. At
90 and 65nm, there cannot be a lot of room to "mess around"
when it comes to making working parts. It has been many years
since I've talked to anyone involved with fabrication, and a
lot of geometry reductions have happened since then.

To save a lot of power, would require some change to the design.
Intel has given some public information about some of the techniques
they have used, to reduce power consumption. So there are ways to
make logic gates "leak" less power when not doing anything. But
there would have to be a significant payback, to design a separate
device with a lower power (like maybe chips designed for laptops
for example). Just making a few low power parts for desktop users
would not justify a separate design.

What does "Stepping" mean ?

For some examples, see page 7 here:
http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/25759.pdf

Steppings are the different revisions of chips. If one chip is made
at 130nm and another at 90nm and yet another at 65nm, they would all
be different revisions. If the process used was changed, like say
silicon on insulator or strained silicon or whatever, those would
likely result in a different revision as well. Since every change
to a chip can result in different bugs being introduced, each
revision should be digitally distinguishable, and the CPUID
function returns a unique value for each revision. (Then you can use
microcode patches to fix the more serious bugs.)

For some of the chips in the AMD table, it might be possible to only
change certain layers in the chip, and change the chip revision. If
you put the CPUID information in a metal layer, you might only have
to change a metal mask to make a different processor revision. So
it may not have cost $16 million to make those 16 different revisions.
And you are not likely to find a friendly text file that explains
why they made all of those. We know some of them were geometry
changes (like going from 130nm to 90nm) but 16 is a lot of different
versions.

Paul

 

[JohnS]

See my questions to Paul at the top.

You know 1-2 years is a lifetime in PC terms. If you buy a decent
system it can be stretched if things dont change a lot over several
cycles of PC hardware and software doesnt increase demands on hardware
radically --- like switch from DOS to WIN95 etc, maybe 3-4 years but
more than that gets a bit iffy.

The best thing to do if you live in the US or other place that has
reasonable access to hardware and OK prices, buy the stuff thats not
cutting edge , the best bargain at the time and sell on ebay etc and
rebuy the same level of stuff.

You know something like a 2500 barton OCed to 3200 when nforce2
systems fell fairly cheap, then go to a 3000 AMD 64 or so when they
started falling and then now a 3800 X2 AM2 or 939 --- can be cheap
with prices in flux now.

You dont even have to worry about OCing . But if you do want to get
into it you should read up on the specific techniques for that CPU.
People will have recommended approaches, memory, boards to get and you
will want to get a feel for what the odds are at what cost and how
high can get easily with a certain CPU+MB combo.

If you really dont upgrade that much a AM2 may not be the best choice.
It all depends on price. If you can get a 754 socket MB and processor
for $80 or something and a 939 for 130 and a AM2 combo for 170 ,
sure you have more upgrade option with a 939 and even more with AM2
but it might not be that big a difference to just sell your 754 dirt
cheap 1-2 years from now and totally upgrade to something else.

 

[Yugo]

No but there were no reasonably priced AM2 machines when I purchased.

I saw the prices drop tremendously but, as I said, I wonder if the new
processors selling under the same name have as much cache.

 

[Yugo]

You know 1-2 years is a lifetime in PC terms.

It depends for who. I write a few emails, write to the news, browse the net,
sometimes crop a picture for a background. What would I do with a dual core?
When I build my present computer with old parts around a Celeron 800 five
years ago in november, I thought I'd throw it away after 3 years. But it works
OK. On Linux, at least. The big problem was dial-up, now I have an ADSL modem
and I hear I'll be upgraded to 5 mbps for free. Hey, I'll take it, but I don't
really need it.

It's really a weird feeling to have technology offer you more than you really
need when, for years, you've been craving for updates.

 

[Yugo]

What you can do at the factory, is sort the processors into "bins".
The processors can be tested at various voltages and frequencies.
The results give what is called a SHMOO plot. For example, with
AMD Cool and Quiet, all the FID/VID values used by the device
must be tested on the chip tester, in order for the chip to be
suitable for its "bin". (That is so AMD can guarantee the chip
will work, when the frequency is changed dynamically while the
system is running.)

You mean when overclocking? How is th efrequency changed dynamically? I
suppose it doesn't require an FX processor, but can be done on the mobo. Can
software do that? (Excuse my complete ignorance of overclocking.)

Is a processor with higher wattage rating more apt to be overclocked or should
everybody look for lower wattage processors?

http://en.wikipedia.org/wiki/Schmoo_plot

If the difference between
the regular and low power devices is only one wattage step, then
that will work out OK. But it also means that the regular devices
will all tend to be the warmer ones, so you won't get "lucky" and
get a lower power one by accident, when binning is being used.

Since an Athlon 64 3500+, a rather cheap processor theses days, is offered in
different voltages, it means that it's been binned. So, it's in a very general
practice. Aren't pretty much all processors binned?

Since the low power ones might be harder to find, that might be
reflected in the asking price - but it is also equally likely that
the demand for the low power ones, is what makes the marketing
department set the price a little bit higher.

How much higher 10 - 15 % ?

I don't think it is too practical to make production changes
to the wafers, to make some of the devices use less power. At
90 and 65nm, there cannot be a lot of room to "mess around"

Isn't Intel testing 50 or 55nm technology? Does this look like the last step
before the quantum theory prevents any further reduction, hence all this
multicore rage to further increase speed?

What does "Stepping" mean ?

For some examples, see page 7 here:
http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/25759.pdf

Steppings are the different revisions of chips. If one chip is made
at 130nm and another at 90nm and yet another at 65nm, they would all
be different revisions. If the process used was changed, like say
silicon on insulator or strained silicon or whatever, those would
likely result in a different revision as well.

I wonder if we're talking about the same thing. At:
http://www.amdcompare.com/us-en/desktop/Default.aspx

the different values for "stepping" for an an Athlon 64 3500+ are C0 CG D0 E3
E4 E6 and F2. I don't see why they would describe the different processes used
in hex. Oops, CG is not hex , but still.

 

[Paul]

What you can do at the factory, is sort the processors into "bins".
The processors can be tested at various voltages and frequencies.
The results give what is called a SHMOO plot. For example, with
AMD Cool and Quiet, all the FID/VID values used by the device
must be tested on the chip tester, in order for the chip to be
suitable for its "bin". (That is so AMD can guarantee the chip
will work, when the frequency is changed dynamically while the
system is running.)

You mean when overclocking? How is th efrequency changed dynamically? I
suppose it doesn't require an FX processor, but can be done on the mobo. Can
software do that? (Excuse my complete ignorance of overclocking.)[/QUOTE]

Cool N' Quiet is a feature of Athlon64 processors. When Windows
is idle, the "CPU driver" can write a FID (frequency) control
register, and drop the processor from 2400MHz, to 1000MHz.
Since in many interactive (web surfing, email) situations, the
processor is idle most of the time, this saves electricity
and makes the processor run cooler. It immediately speeds up
as soon as you move the mouse.

When the processor is running at full speed (say you are shrinking
a DVD before writing a movie to a blank DVD), then the power
difference between the high power and low power processors takes
effect. Maybe if you kept your processor busy all the time,
the lower power rated processor would give you a cooler running
computer. But for interactive style usage, the Cool N' Quiet feature
saves plenty of power.

Is a processor with higher wattage rating more apt to be overclocked or should
everybody look for lower wattage processors?

Overclocking is when a processor is used outside its intended
frequency range. In all the examples I've used so far, the
processor has been used within its intended range.

As an example, look at PDF page 17 of this document. It shows the processor
ADA3500DEP4AS, a 3500+.

http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/30430.pdf

The allowed operating conditions, tested at the factory are:

Name Power Frequency
Max Pstate 89W 2200MHz
Intermediate Pstate #1 69W 2000MHz
Intermediate Pstate #2 50W 1800MHz
Min Pstate 22W 1000MHz

If you were to operate that processor at 2400MHz, _that_ would be
overclocking. If the operating system writes the frequency control
with a value between 1000MHz and 2200MHz, that is all part of the
normal operating range. If the computer is idle in Windows, then
the frequency can drop down to 1000MHz, and your processor only uses
22W of power.

If Cool N' Quiet is disabled, by either not using the CPU driver, or
by disabling the feature in the BIOS, the processor would run at a
constant 2200MHz, and draw _up to_ 89W. All I can say, is the power
used would be more than 22W and less than or equal to 89W, with CNQ
disabled. The power drawn is still proportional to the amount of
work being done, but doesn't save quite as much power when the
processor is idle, since the chip clock is still 2200MHz. And
that is why enabling CNQ and installing the CPU driver is important
for your electric bill.

How much higher 10 - 15 % ?

According to amdcompare.com, the difference between the regular 4400+
and the "energy efficient" is 89W versus 65W. The savings are
most evident when the computer is kept busy 100% of the time.
With Cool N' Quiet enabled, we need a document like 30430.pdf
from AMD, to tell us how the two compare when the computer is
completely idle.

Isn't Intel testing 50 or 55nm technology? Does this look like the last step
before the quantum theory prevents any further reduction, hence all this
multicore rage to further increase speed?

They are working on 45nm. Who needs quantum theory

The transistors are getting more complicated, with mechanisms needed to
keep the transistors from "leaking" when they are not doing anything.
That is one secret to success. Other things are not improving, like
the copper wire used inside.

I wonder if we're talking about the same thing. At:
http://www.amdcompare.com/us-en/desktop/Default.aspx

the different values for "stepping" for an an Athlon 64 3500+ are C0 CG D0 E3
E4 E6 and F2. I don't see why they would describe the different processes used
in hex. Oops, CG is not hex , but still.

As far as I know, CG is 130nm and E4 is 90nm. The transistor size is
smaller on the more modern processor. The processors are cheaper to
make, as the silicon die can be smaller. Some improvements are
for profitability, others are for speed improvements or power savings.
If the projected savings exceed the engineering cost, then voila,
they make a new stepping. There is a large staff in the fab, who
do nothing but work on improvements. Some of those people are
amazingly clever.

Paul

 

[Bazzer Smith]

It depends for who. I write a few emails, write to the news, browse the
net, sometimes crop a picture for a background. What would I do with a
dual core? When I build my present computer with old parts around a
Celeron 800 five years ago in november, I thought I'd throw it away after
3 years. But it works OK. On Linux, at least. The big problem was dial-up,
now I have an ADSL modem and I hear I'll be upgraded to 5 mbps for free.
Hey, I'll take it, but I don't really need it.

It's really a weird feeling to have technology offer you more than you
really need when, for years, you've been craving for updates.

The system I replaced recently was a Cyrix MII 300, but realistically it
needed to
be replaced a few years earlier, there were some things it was just not
capable
of doing, like watching streaming video, or playing some .wmv video clips,
but
I could get by without them. What I did find annoying was that I could not
upgrade my CPU because ithe motherboard socket supported nothing faster,
thats why I
say go for the most advanced motherboard socket you can get, which in your
case would be AM2.
Had I bought a machine with a more modern motherboard I could have picked up
a faster CPU like your celeron 800 for about £10, and then my computer would
have
had a much more active 'old age'.
It would have just given me a bit of breathing space whilst I researched the
latest
technology, which, as you have probably discovered, is no where near as
simple
as it used to be.

Also you never know what your future needs may be, I recently bought a USB
'digital televison stick' (DVB-T for) about £40 so now my PC can act as both
a TV and
a digital TV recorder (PVR). If I were to buy a PVR in the shops it would
cost me
£150-£200 alone, almost as much as I paid for my PC!
I probably won't ever need to buy a TV again now, all I need is a big
computer
monitor to replace it with. That means I can do away with a digital set-top
box,
PVR and all the asociated cabling. And and if I want another TV in a
different room
I just have to buy another monitor and it can use my PC resources, I might
have to buy another DVB-T stick though, but my proccesor, or a processor
upgrade should be able to cope with all that.

 

[Bazzer Smith]

I saw the prices drop tremendously but, as I said, I wonder if the new
processors selling under the same name have as much cache.

I would assume so, I think the price drop is to make them competitive
with newer processors, so lowering the cache would be counter productive.
Anyway, in a lot of applications the cache size makes little or no
difference
beyond a certain level of cache.
My system 'only' has 128KB of cache, but even if it had 1MB of cache I doubt
I would be able to tell the difference, at least not without using a
stopwatch and that
would only apply to a few very specific applications.

 

[kony]

I saw the prices drop tremendously but, as I said, I wonder if the new
processors selling under the same name have as much cache.

Unless your particular task(s) need an especially large
cache it may not matter... the rating system is meant to
handle this, you get a higher clockspeed if all else were
equal.

 

[Yugo]

Cool N' Quiet is a feature of Athlon64 processors. When Windows
is idle, the "CPU driver" can write a FID (frequency) control
register, and drop the processor from 2400MHz, to 1000MHz.
Since in many interactive (web surfing, email) situations, the
processor is idle most of the time, this saves electricity
and makes the processor run cooler. It immediately speeds up
as soon as you move the mouse.

It seems there's also a driver for Linux 2.6.10 and later kernels:

AMD Turion™ 64 Mobile Technology Processors,
AMD Opteron™ Processors, and
AMD Athlon™ 64 Processors Driver Version 1.60.01 for Linux 2.6. - Supports all
AMD Turion™ 64 Mobile Technology Processors, AMD Opteron™ Processors, and
Athlon™ 64 Processors released through 2006. Provides support for AMD
PowerNow!™ technology and, where appropriate, AMD’s Cool-n-Quiet™ technology
for Linux systems. Works with all kernels, version 2.6.10 or later. Requires
cpufreq-1.20, cpuspeed-1.20.1, or powersaved-0.8.19 or later to support SMP
and dual-core systems.

http://www.amd.com/us-en/Processors/TechnicalResources/0,,30_182_871_9706,00.html

So, the same processor may run at different wattages. OK.

When the processor is running at full speed (say you are shrinking
a DVD before writing a movie to a blank DVD), then the power
difference between the high power and low power processors takes
effect. Maybe if you kept your processor busy all the time,
the lower power rated processor would give you a cooler running
computer. But for interactive style usage, the Cool N' Quiet feature
saves plenty of power.

I see... but I'm afraid I'll have a question below.

As an example, look at PDF page 17 of this document. It shows the processor
ADA3500DEP4AS, a 3500+.

http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/30430.pdf

The allowed operating conditions, tested at the factory are:

Name Power Frequency
Max Pstate 89W 2200MHz
Intermediate Pstate #1 69W 2000MHz
Intermediate Pstate #2 50W 1800MHz
Min Pstate 22W 1000MHz

Ok. Now, here's the SAME processor at:

http://www.amdcompare.com/us-en/desktop/default.aspx

Processor: AMD Athlon™ 64
Model Number: 3500+
Frequency (MHz): 2200
L2 Cache Size: 512KB
Socket: AM2
Stepping: F2
Manufacturing Tech (CMOS): 90nm SOI
Wattage (W):
System Bus (MHz): 2000

As you can see, there's nothing beside Wattage. At this stage, the choice is
between 35 or 62. Isn't this to operate at 2200 MHz, the maximum frequency?

If you were to operate that processor at 2400MHz, _that_ would be
overclocking.

This much I know

They are working on 45nm. Who needs quantum theory
Hum...


processes used



As far as I know, CG is 130nm and E4 is 90nm.

I see...

There is a large staff in the fab, who
do nothing but work on improvements. Some of those people are
amazingly clever.

They better be, competition is fierce. How do you think VIA will succeed in
the chinese market? Their processors are not very powerful, but they consume
less. Their random generator is, to say the least, original. Maybe they've got
some other original designs, I don't know.