amd vs. intel

  • Thread starter Thread starter Tanya
  • Start date Start date
Just an additional data point on heat generation issue...
Athlon 64 winchester (90nm) is pretty unbelievable when it
comes to heat generation/ power consumption. It generates
even less heat than old athlon xp (thoroughbred B). Figures:

Idle K7burn
3GHz P4 Prescott D0: 30W 90W
3GHz P4 Northwood: 18W 70W
2GHz AXP Barton: 38W 60W
2GHz A64 Newcastle: 15W 55W
2GHz A64 Winchester: <10W 30W

I have athlon 64 3000+, and I am often amazed that cpu fan
is at dead stop and hardly stirs even while after I've been
playing UT2004 for a while. And to confirm, yes the heat sink
is barely warm to touch, which is unbelievable for cpu of
this performance category.

-ax
 
hi Martin,
thank you for the reply and advice...

tartan_martin said:
Hi,

Hope this can help you some more....

For games then Intel has to be serious choice though AMD can't be ruled
out because of its relatively cheap price, however... for offive based
applications etc then AMD becomes the favoured choice as its a HELL of
a lot cheaper than Intel and the functionality of the chips are
slightly better than Intel.

wherever i ask, amd seems to be a better choice (for anything / everything
:-)
i guess i'll try it...
thanks for the info!
sincerely
Tanya
 
hi Tony,
thanks again for replying and for providing the information!
[...below...]

Tony Hill wrote:
The i925XE chipset adds support for the 1066MT/s processor bus, but
drops support for DDR memory in favor of DDR2. It also adds a good
$100+ to the price tag of the board. All else being equal, an i925XE
chipset should also be a tiny bit faster than an i915G chipset, but
the difference is quite small and, IMO, not worthwhile.

getting more confused:)
read that a main determinant of performance is the "clock speed" (states that this
is not the same as operating frequency of the cpu (but is related)) and is a
function of the motherboard bus (however the only speeds on the motherboards seem to
be memory speeds...)
i was looking at the abit fatl1ty and the asus p5ad2-e (both supporting 1066 fsb
speed but only ddr2 (and more expen$ive) (chipset i925xe))
None of the i9xx series of chipsets support AGP. Not really a big
loss unless you've already got an old AGP card you want to bring over
to the new system.

(i guess they are slower ...)
The integrated video of the i915G is actually quite decent. Beyond
gaming, the only other sorts of applications where it really flops are
things like high-end graphics and CAD applications (particularly if
you're doing any sort of 3D CAD).

i would assume that if i bought a pci-express graphics card, the overall performance
would not be adversely affected (if it had been affected with the onBoard chip) ?
though these days the difference is

PCI-Express slot... Or PCI I guess, though it doesn't make much sense
to use that if you've got a PCI-Express 16x slot.



This connector will be the one connected directly to the chipset and
is the only ATA connector you can use to initially load your operating
system with.


These would be fine once Windows is loaded, but won't work for loading
the OS.


This is where the important stuff is. Most likely you'll want to get
an SATA hard drive and throw it on one of these connectors. You
should be able to load the OS with no trouble on these. At least in
theory, SATA should be a bit faster than ATA as well.

yes, plan on sata / raid drive
Yup, should be plenty. Asus has decided to include a secondary ATA
controller on-board (the ITE one). As mentioned above, you can't
really use that until your OS is loaded, but that shouldn't be a big
worry, you just need a single hard drive (ATA or SATA) and a single
CD-ROM to load the OS. Additional devices can be left until later.


Putting the memory controller on-chip reduces the time it takes to get
data to/from main memory by about 30%. This translates into about a
10-20% improvement in overall system performance and is one of the key
reasons why 2.0GHz Athlon64 chips are often faster than 3.0GHz P4
chips. Basically everyone in the chip-making business other than
Intel has decided the integrated memory controllers are the way to go.

not related to l2 cache (better to get a cpu with the l2 cache *built in* (vs. on
the board))?
and the memory controller being on the chip, is this why the l2 cache and the total
system mem are less than p4's?
Beyond performance though, it shouldn't change things much.
while for Intel the memory controller is off on the chipset. For AMD
chips, any of their Athlon64s that fit into Socket 939 are
dual-channel, regardless of the chipset used, while those that fit
into Socket 754 are single channel. For Intel it's a little bit more
complicated in that, for example, the i915 is dual channel but the
i910 is single channel. Generally speaking, dual-channel is better
for performance and shouldn't change the price much, but you do have
to remember to add memory in matched pairs, which can complicate
upgrades slightly. For example, a dual-channel motherboard might have
4 memory sockets, but after you drop the first two sticks in you've
only got room for one more matched pair of memory to upgrade with down
the road. Keep in mind that two or three years after you get your
computer, upgrading the memory is almost always the best bang/buck
upgrade you can get.

there are 4 sockets...
this board supports either ddr or ddr2 (it states [under the specs] that it
supports dual, single (perhaps the ddr2 will not require dual?))
i'd be using ddr initially
also the max is 4 gb

From the specs I can see there are 4 DDR sockets and 2 DDR2 sockets on
this board. Both DDR and DDR2 are best used in a dual-channel setup,
though sometimes it is possible to use a single-channel setup with
about a 10% performance loss, ie not a very good setup.
Note that even though this board supports both DDR and DDR2, it almost
certainly does NOT support both at the same time, it's one or the
other.

not sure whether 2 ddr2 chips are enough?
(i also found that it supports 2*ddr2 OR up to 4*ddr)
with respect to the amd-based boards:
there are 3 categories; (on motherboards.org site)
(i realize that one cannot compare apples to oranges however which group would
be comparable [closest] to s775?)
socket a

Socket A is rather dated these days and is in the process of being
EOLed. It's had a good life (it was first introduced in late 2000 or
early 2001 if my memory serves me correctly) and can still be used for
a great low-budget system. However for the most part you can ignore
this one.
socket 754, 939, 940

This is where the magic is happening for AMD systems.

Socket 754 - Older (single-channel memory) Athlon64 chips and Sempron
chips. Mainly a low-cost solution these days.

Socket 940 - Used pretty much exclusively for the AMD Opteron for
servers and workstations. This socket includes support for large
amounts of memory and multiple processors, but probably not of much
interest for most home users (err.. except for Keith, but he doesn't
count :> ). Socket 940 would be AMD's equivalent to Intel's Socket
604 for their Xeon processors.

Socket 939 - This is definitely the closest equivalent to Intel's
Socket 775 and would be what you would probably want to concentrate on
when looking at AMD-based systems. It is used for pretty much all the
new AMD Athlon64 based systems. It adds dual-channel memory when
compared to Socket 754 and lower costs when compared to Socket 940.
socket 939 (64-bit)

This one hasn't exactly showed up yet, but it's expected to replace
Socket 754 for low-cost systems eventually.

thanks for the above...i'll look into the socket 939......
It does generate a fair bit of heat, though the new 600 series P4
chips have made some improvements in this regard. Generally speaking
it is unavoidable, though a good heatsink and decent airflow in your
case can take care of it.

Speaking of those 600 series P4 chips, I would HIGHLY recommend
getting one of those if you're looking at a P4 system. Performance
should be about the same as a 500 series P4 of one higher model number
(ie the P4 640/3.2GHz should perform about the same as a 550/3.4GHz
chip). On top of that they do reduce the power consumption with a few
new tricks, but most importantly, they add 64-bit support. While
Microsoft still hasn't quite got around to releasing 64-bit Windows
(1.5 years late and counting) it is coming and this is going to be the
standard going forward.

aren't these the ones that have the 1066 mhz front side bus speed?
if they are not, i could still get the i915g board.......

For AMD Athlon64 chips you've already got 64-bit support and better
power consumption than any of the Pentium4 chips, so not such a big
worry.

thanks a lot again!
very much appreciated!
sincerely
Tanya
 
hi Kieth, thanks for replying...
confused but...
hi Yousuf,
thanks for the reply...
[...below...]

Yousuf said:
Tanya wrote:
hi
i'm building a system (pentium 4 based) from scratch
i keep hearing how great amds are and want to make sure that i am not
missing something
thanks

These days there's no reason to not consider AMD, but that applies
equally back to Intel. At one time, people who were looking to put
together a system on a budget used to look automatically to AMD,
nowadays some of Intel's high-end chips are as cheap as AMD's mid-price
chips. In fact, I'd say the Intels are a good bargain these days.

i've only owned intels (and find them reliable)
i heard somewhere that amd is not as reliable....

You heard wrong, though FUD has been the rule in this business since its
conception.
maybe you can explain this: the categories (of ranking amds) are socket a;
socket 754, 939, 940; and socket 939)
which would be *comparible* to the lga 775 socket?

It depends on what you mean by "comparable". We've had the LGA discussion
here before. I'm not a fan of LGAs in this market at all. They're great
if the chip is permanently mounted, but call me skeptical of LGA sockets.

comparable i guess is not possible... from what i've read, the operating
frequency, the fsb speed and the l2 cache size are the important determinants
(the amd is less in each category) HOWEVER, i read on amd's Web site that the
amd does more per cycle than the pentium also they (the amd Web site) compares a
p4 with an amd and the "overall performance" is better for the amd (i don't know
what they mean by overall performance.)

Socket-939 is likely the best choice for a "performance" system now.
Socket-940 requires registered DRAM, which is somewhat more expensive and
a (very) little lower performance. Socket-940 is really intended for
servers. IMO, socket-754 is dead-end and I don't see any reason to go
there. Socket-A is used for Athlons and a good choice for "value" systems.

One can build a rather impressive socket-A system on the cheap. I built
quite an impressive system for a friend for $400 (display, keyboard, and
OS recycled from one that had the magic smoke let out).

The big advantage AMD has (sockets 939 and 940) is the memory
architecture. The integrated DRAM controllers (vs. controller on the
northbridge) are a performance advantage, but also cost a lot of pins on the
processor chip.

is that why the boards hold less ram? and why the cpu performs better even with
a smaller l2 cache?

i appreciate the above info about the sockets a lot -- thanks
In short, I'd go with a P4 if I were going to do a lot of streaming video,
P3M for a laptop, and Athlon64 for everything else. I do "everything
else" so have an Opteron (socket-940), which I bought before 939 was
generally available).

even though the *numbers* (values) don't add up, (i.e. lower fsb speed, l2
cache, operating freq) your (and others) opinion are helpful -- i appreciate it
and now will look at (learn about) amd's.


sincerely,
Tanya
 
Just an additional data point on heat generation issue...
Athlon 64 winchester (90nm) is pretty unbelievable when it
comes to heat generation/ power consumption. It generates
even less heat than old athlon xp (thoroughbred B). Figures:

Idle K7burn
3GHz P4 Prescott D0: 30W 90W
3GHz P4 Northwood: 18W 70W
2GHz AXP Barton: 38W 60W
2GHz A64 Newcastle: 15W 55W
2GHz A64 Winchester: <10W 30W

Interesting. Do you have benchmarks for these. Say, Widget-ops/watt?
I have athlon 64 3000+, and I am often amazed that cpu fan
is at dead stop and hardly stirs even while after I've been
playing UT2004 for a while. And to confirm, yes the heat sink
is barely warm to touch, which is unbelievable for cpu of
this performance category.

I find this amazing too.
 
hi Tony,
thanks again for replying and for providing the information!
[...below...]

Tony Hill wrote:
The i925XE chipset adds support for the 1066MT/s processor bus, but
drops support for DDR memory in favor of DDR2. It also adds a good
$100+ to the price tag of the board. All else being equal, an i925XE
chipset should also be a tiny bit faster than an i915G chipset, but
the difference is quite small and, IMO, not worthwhile.

getting more confused:)
read that a main determinant of performance is the "clock speed" (states that this
is not the same as operating frequency of the cpu (but is related)) and is a
function of the motherboard bus (however the only speeds on the motherboards seem to
be memory speeds...)

Oh, no! Clock speed is only important when comparing (very) like
processors. A 2GHz Opteron will choke a 3GHz P4 in almost any test you
wish to throw at it (video encoding is the possible exception).

I'm not sure what you mean by "operating frequency". The "front-side bus"
frequency matters, though not so much with the AMD64s because they don't
use the FSB for memory. The memory controller is integrated into the
processor so the "bus" is only for I/O stuff (on a single processor system).
i was looking at the abit fatl1ty and the asus p5ad2-e (both supporting
1066 fsb speed but only ddr2 (and more expen$ive) (chipset i925xe))

DDR2 is a waste of money. You'll do *much* better with an AMD processor
with its integrated memory controller and DDR400, or some such. As has
been suggested here before PCI-E is a good idea, if only because AGP has
died an early death. A *very* surprising event, IMO.
(i guess they are slower ...)

A newer graphics card will likely be faster. The difference between AGP
and PCI-E will likly never be seen. Given that PCI-E has rather quicly
swept AGP under the door-mat, it's the way to go.
discrete graphics



i would assume that if i bought a pci-express graphics card, the overall
performance would not be adversely affected (if it had been affected
with the onBoard chip) ?

The integrated graphics controller can be disabled, or used to drive
another monitor. Dual monitors are the way to go, IMO (productivity
gains are incredible). Though most decent graphics cards support them
natively.

^
+--- Good idea!

yes, plan on sata / raid drive

I've been burnt. Software RAID doesn't really interest me (went through
these exercises moons ago when the Promise controllers hit the market).
SATA seemed like a good idea, and is. ...in the long run. I'd gladly trade
my SATA drive for a pATA. It's been doing nothing for almost a year.
Again, in the long run...

not related to l2 cache (better to get a cpu with the l2 cache *built
in* (vs. on the board))?

No L2's are on-board anymore. They're all integrated into the processor.
That said, the integrated DRAM controller is still very important. I
find it amazing that AMD still is the only one (in this market) who's
figured this out.
and the memory controller being on the chip, is this why the l2 cache
and the total system mem are less than p4's?

You'd better look at your figures again. I'm not sure where you're
getting this information from.

not sure whether 2 ddr2 chips are enough? (i also found that it supports
2*ddr2 OR up to 4*ddr)

Why are you concerned with DDR2? It's a waste of money. Don't go there.

thanks for the above...i'll look into the socket 939......

DO it! You won't be dissapointed. Were it available last year I'd have
gone that way.
aren't these the ones that have the 1066 mhz front side bus speed? if
they are not, i could still get the i915g board.......

The FSB speed is next to irrelevant if system memory isn't forced to hang
off it. ;-)
 
hi Kieth, thanks for replying...
confused but...

Fire away! BS about differing processors/systems is what we do here. ;-)

comparable i guess is not possible... from what i've read, the operating
frequency, the fsb speed and the l2 cache size are the important determinants
(the amd is less in each category)

Have you looked at the D and I caches on the P4? The cache latencies?
No, like life, it's not the size that's important, but what you do with it.
;-)

Frequency is next to irrelevant for processors with differing
micro-architectures.
HOWEVER, i read on amd's Web site that the
amd does more per cycle than the pentium also they (the amd Web site)
compares a p4 with an amd and the "overall performance" is better for
the amd (i don't know what they mean by overall performance.)

Benchmarks. YEs, AMD has regularly kicked Intel's but in instructions per
clock. The P4 is particularly bad in this regard, as AMD will gladly
point out. IMO, the P4 was a horrible design, not suited for the market
it ended up in. The P3 is a better choice and AMD out-does even that.
is that why the boards hold less ram? and why the cpu performs better
even with a smaller l2 cache?

Less RAM? My Opteron has 1.5GB and it (as opposed to my CFO) would easily
go to 8GB, which would be a problem on any P4. ;-)
i appreciate the above info about the sockets a lot -- thanks

Hang around. Lots of fun stuff happens around here. ;-)

even though the *numbers* (values) don't add up, (i.e. lower fsb speed,
l2 cache, operating freq) your (and others) opinion are helpful -- i
appreciate it and now will look at (learn about) amd's.

The FSB is meaningless when comparing the Intel and AMD offerings. I'd
like to see what L2's you're comparing. ...notto mention the L1s, whichh
I believe you're ignoring. The bottom line is that you have to compare
*performance*. The microarchitectures of modern processors are just too
different to compare raw numbers.
 
The FSB is meaningless when comparing the Intel and AMD offerings. I'd
like to see what L2's you're comparing. ...notto mention the L1s, whichh
I believe you're ignoring. The bottom line is that you have to compare
*performance*. The microarchitectures of modern processors are just too
different to compare raw numbers.

I think she means the larger L2 cache in P4s - the 600s have 2MB... the
message being that Intel had to go there to err, catch up and try to hide
its latency... with umm, bandwidth?<titter>
 
getting more confused:)
read that a main determinant of performance is the "clock speed" (states that this
is not the same as operating frequency of the cpu (but is related)) and is a
function of the motherboard bus (however the only speeds on the motherboards seem to
be memory speeds...)

Sadly there are MANY factors that can improve performance, so we can't
just throw one number at them and have it explain everything.

The i925XE chipset does include some slightly better timings when it
comes to the memory controller, so it can very slightly reduce the
memory latency when compared to the i915. How much does this amount
to? Typically it's a question of a clock cycle or two here and there
and it works out to about 0-2% overall system performance.

For the extra $100 that the motherboards cost an average improvement
of only about 1% really isn't worth IMO.
(i guess they are slower ...)

Somewhat, though it's more a question of cost. In theory PCI-Express
is the One Bus to Unit Them All, which should reduce the cost. With
some older systems you could easily have 4 or 5 different buses for a
variety of different cards and connectors. Supporting extra buses
means more connectors, more wires, more transistors and just generally
more $$$. Dropping support for AGP cuts the costs of a fairly
expensive bus that can most easily be replaced. As such, it was one
of the first to hit the chopping block (ACR, CNR and CSA also aren't
likely to make the cut, but they were rather obscure to begin with and
can safely be ignored).
i would assume that if i bought a pci-express graphics card, the overall performance
would not be adversely affected (if it had been affected with the onBoard chip) ?

That is correct. I wouldn't sweat any performance loss due to the
on-board chip either, it's really quite minimal these days. 5 years
ago it was quite a different story, but these days the performance
difference is mostly lost in the noise (ie less than 2%).
yes, plan on sata / raid drive

Sounds like a good plan, though RAID can be a whole other can of
worms. My personally opinion on it is to stick with RAID-1
(mirroring) due to reliability concerns. I've seen just WAY too many
hard drives die to consider RAID-0 (stripping).
not related to l2 cache (better to get a cpu with the l2 cache *built in* (vs. on
the board))?

Similar idea, though you'll have a HELL of time finding any chip that
doesn't have L2 cache built-in. The last x86 chips I know of that
used external L2 cache were the original Athlon chips, discontinued in
late 2000, and these chips had their L2 in a little cartridge
alongside the processor, not on the system board.

All current processors have their L2 cache on-chip.
and the memory controller being on the chip, is this why the l2 cache and the total
system mem are less than p4's?

They are partly related, though it's a bit more complicated than that.
The integrated memory controller and lower memory latency time of the
Athlon64 mean that it's less dependant on getting data from it's L2
cache, so it can get by with less cache (or conversely, the higher
memory latency of the P4 means that it's more dependent on L2).

As for total system memory, there isn't much reason why that should
change one way or the other. Actually the Athlon64 supports MORE
memory than the P4 (8GB vs. 4GB), but in practical purposes the
difference doesn't amount to much.
there are 4 sockets...
this board supports either ddr or ddr2 (it states [under the specs] that it
supports dual, single (perhaps the ddr2 will not require dual?))
i'd be using ddr initially
also the max is 4 gb

From the specs I can see there are 4 DDR sockets and 2 DDR2 sockets on
this board. Both DDR and DDR2 are best used in a dual-channel setup,
though sometimes it is possible to use a single-channel setup with
about a 10% performance loss, ie not a very good setup.
Note that even though this board supports both DDR and DDR2, it almost
certainly does NOT support both at the same time, it's one or the
other.

not sure whether 2 ddr2 chips are enough?
(i also found that it supports 2*ddr2 OR up to 4*ddr)

Probably. DDR2 still hasn't really made a big impact on things and
likely isn't going to for a while yet. Personally I would mostly just
ignore the DDR2 for the time-being, it costs more but is no faster and
DDR will be widely available for a long time (in computer terms, ie
3-5 years) to come.
aren't these the ones that have the 1066 mhz front side bus speed?
if they are not, i could still get the i915g board.......

Nope, the only chips with the 1066MT/s bus speed are the P4 Extremely
Expensive Edition chips. The 600 series P4 chips have the same
800MT/s bus speeds as the older 500 series P4 chips before them. they
should work on most/all i915G boards. They definitely will work on
that Asus board you had mentioned earlier.

Ohh.. speaking of that board, the Asus P5GDC-V Deluxe (I think that
was the one you were thinking of?) does NOT support integrated video.
Even though it comes up under their i915G chipsets, it's actually
using the i915P chipset. I don't know if this is a flaw in their
website or just the way they have things setup, just a word of warning
though. It's rather confusing and I definitely missed this one the
first read through it.
 
keith said:
Interesting. Do you have benchmarks for these. Say, Widget-ops/watt?

I find this amazing too.

Another good comparison of power consumption is given by anandtech:

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2275&p=13

Under load (media encoder 9 test which stress cpu sub-system mostly),
3.0 ghz p4 530 consumes 81 watts more than athlon 64 winchester 3500+.
3.8 ghz p4 570 consumes 111 watts more than athlon 64 3500+. So the
actual difference in athlon 64 winchester vs. p4 power consumptions may
be even more exaggerated than the k7burn figures given above.

-ax
 
Another good comparison of power consumption is given by anandtech:

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2275&p=13

Under load (media encoder 9 test which stress cpu sub-system mostly),
3.0 ghz p4 530 consumes 81 watts more than athlon 64 winchester 3500+.
3.8 ghz p4 570 consumes 111 watts more than athlon 64 3500+. So the
actual difference in athlon 64 winchester vs. p4 power consumptions may
be even more exaggerated than the k7burn figures given above.

A point AMD recently made about power consumption comparisons is
that the AMD64 processors have an integrated memory controller.
By not having to also be memory controllers, chipsets for AMD64
apparently save about 20W over chipsets for Intel.

So if your AMD64 cpu saves you 29 W (29 W makes the math easy
below ) and your NF4 chipset saves you 20W, and if you have a
great PSU with 70% efficiency, then your actual saving at the
wall outlet should be along the lines of
(29W + 20W)/0.70 = 70 Watts
which compares reasonably well with the AnandTech numbers.
 
A point AMD recently made about power consumption comparisons is
that the AMD64 processors have an integrated memory controller.
By not having to also be memory controllers, chipsets for AMD64
apparently save about 20W over chipsets for Intel.

So if your AMD64 cpu saves you 29 W (29 W makes the math easy
below ) and your NF4 chipset saves you 20W, and if you have a
great PSU with 70% efficiency, then your actual saving at the
wall outlet should be along the lines of
(29W + 20W)/0.70 = 70 Watts
which compares reasonably well with the AnandTech numbers.

20W seems high to me for just the memory controller, but it's a very good
point.
 
Rob said:
A point AMD recently made about power consumption comparisons is
that the AMD64 processors have an integrated memory controller.
By not having to also be memory controllers, chipsets for AMD64
apparently save about 20W over chipsets for Intel.

So if your AMD64 cpu saves you 29 W (29 W makes the math easy
below ) and your NF4 chipset saves you 20W, and if you have a
great PSU with 70% efficiency, then your actual saving at the
wall outlet should be along the lines of
(29W + 20W)/0.70 = 70 Watts
which compares reasonably well with the AnandTech numbers.

That's a good point about external memory controllers. But
according to some direct cpu power consumption measurements,
70 Watts seem to be the actual power savings of athlon 64 winchester
over p4 prescott, and are more in line with anandtech numbers:

power consumption of a64 winchester range from 40watts ~ 51 watts under
load:
http://www.xbitlabs.com/articles/cpu/display/athlon64-90nm_5.html

power consumption of prescott under load range from 97 watts ~ 118
watts:
http://www.xbitlabs.com/articles/cpu/display/pentium4-570_4.html

These measurements were currents of 12 volts line supplying the cpu,
so there is no PSU power conversion factor involved. If, as you
mentioned athlon 64 processors add additional 20W of power savings to
account for external memory controller for prescott, overall power
savings would be as large as 90 Watts which is huge difference. After
PSU inefficiency, of course, it would be even bigger, up to numbers
mentioned from anandtech article.

-ax
 
hi and thanks again for answering!
[...below...]

<snip>
sorry -- read further that the internal architecture of a cpu is a major determinant of
speed (as you mention)
Oh, no! Clock speed is only important when comparing (very) like
processors. A 2GHz Opteron will choke a 3GHz P4 in almost any test you
wish to throw at it (video encoding is the possible exception).

I'm not sure what you mean by "operating frequency".

what i mean is the number of ghz. eg. p4 (@ 2.8 ghz); amd (@2 ghz)
The "front-side bus"
frequency matters, though not so much with the AMD64s because they don't
use the FSB for memory. The memory controller is integrated into the
processor so the "bus" is only for I/O stuff (on a single processor system).

(isn't the communication between the cpu and memory (which is determined via the memory
controller (integrated for amd; onMotherBoard for pentium4) called the fsb? i.e. the
"path" : cpu <-> memory))
(i.e. the mem controller determines what to get for the cpu -- right? well doesn't the
memory need to get to the cpu? (it is still on the motherboard so needs a bus to get to
the cpu?)
?
DDR2 is a waste of money.

i hear that plus there's a relatively long latency...
You'll do *much* better with an AMD processor
with its integrated memory controller and DDR400, or some such. As has
been suggested here before PCI-E is a good idea



A newer graphics card will likely be faster. The difference between AGP
and PCI-E will likly never be seen. Given that PCI-E has rather quicly
swept AGP under the door-mat, it's the way to go.

isn't it more expen$ive?
The integrated graphics controller can be disabled, or used to drive
another monitor. Dual monitors are the way to go, IMO (productivity
gains are incredible). Though most decent graphics cards support them
natively.

i wouldn't even know what to use dual monitors for (this is where the human bottleneck is
having to look at 2 monitors simultaneously said:
^
+--- Good idea!


I've been burnt. Software RAID doesn't really interest me (went through
these exercises moons ago when the Promise controllers hit the market).

i got a question here (i cannot find the article though -- i read that there are several
raid levels and some are great in keeping *copies* soToSpeak of the drive's data...)
SATA seemed like a good idea, and is. ...in the long run. I'd gladly trade
my SATA drive for a pATA.


No L2's are on-board anymore. They're all integrated into the processor.
That said, the integrated DRAM controller is still very important.

i believe that it would be. the way i understand it (which is likely more fiction:) is
that it calls data from ram as per the cpu's direction HOWEVER if it is integrated it can
*arrange* to get / send memory data on it's own being the cpu???)
I
find it amazing that AMD still is the only one (in this market) who's
figured this out.

i actually asked amd for some tech articles on why their cpu is <almost> always
recommended / preferred and i have some interesting info from them which definitely
prooves their cpus are better performers.
i believe that amd is not in <as much> flux (at least not as much as intel is) i.e. is not
between chip architecture -- i still like intel however, their 1066 mhz fsb is new (and
EXPEN$IVE) and likely needs fine tuning --
You'd better look at your figures again. I'm not sure where you're
getting this information from.

i'm getting the "max memory" from amd-based motherboards: for example: aopen n250a-fr "max
mem" 3 gb.
asus k8n-e dx "max mem" 3gb.
msi k8n neo ms7030 "max mem" 3 gb
but for other boards i see 4 gb max............
<more snippage>

Why are you concerned with DDR2? It's a waste of money. Don't go there.

when i was considering an intel-based system with support for 1066 mhz fsb speed... WAY
back...
:)
DO it! You won't be dissapointed. Were it available last year I'd have
gone that way.

only 1 problem: which board??? and i guess the nforce4 chipset... is asus good? (it is
supposed to be for p4 systems.)
The FSB speed is next to irrelevant if system memory isn't forced to hang
off it. ;-)


thank you
sincerely
Tanya

p.s. 1 more question: i read an article on cpu's (in general) which states that the more
transistors, the better (they list intel based cpus) in some of the info on amd's site,
the amd athlon 64 has 105.9 * 10^6 and the p4 has 125 * 10^6
perhaps it is the way transistors are used not their absolute number?
 
hi Tony,
thanks for the response!
[...below...]

Tony said:
Sadly there are MANY factors that can improve performance, so we can't
just throw one number at them and have it explain everything.

even in an article that defines clock speed it states that given the same clock speed 2
cpu's of differing architecture will perform differently.
The i925XE chipset does include some slightly better timings when it
comes to the memory controller, so it can very slightly reduce the
memory latency when compared to the i915. How much does this amount
to? Typically it's a question of a clock cycle or two here and there
and it works out to about 0-2% overall system performance.

For the extra $100 that the motherboards cost an average improvement
of only about 1% really isn't worth IMO.

and i'm realizing that amd might be the best choice for now -- i like intel however, they
are just introducing the p4 (with the 1066 mhz fsb speed) so they're expensive and
possibly have issues that'll need to be cleared up...
Somewhat, though it's more a question of cost. In theory PCI-Express
is the One Bus to Unit Them All, which should reduce the cost. With
some older systems you could easily have 4 or 5 different buses for a
variety of different cards and connectors.

i'd read that there is 1 speed for the pci bus, and the agp is 2* the pci bus (this is an
older article) i don't know whether isa had its own bus speed?
Supporting extra buses
means more connectors, more wires, more transistors and just generally
more $$$. Dropping support for AGP cuts the costs of a fairly
expensive bus that can most easily be replaced. As such, it was one
of the first to hit the chopping block (ACR, CNR and CSA also aren't
likely to make the cut, but they were rather obscure to begin with and
can safely be ignored).




That is correct. I wouldn't sweat any performance loss due to the
on-board chip either, it's really quite minimal these days. 5 years
ago it was quite a different story, but these days the performance
difference is mostly lost in the noise (ie less than 2%).


Sounds like a good plan, though RAID can be a whole other can of
worms. My personally opinion on it is to stick with RAID-1
(mirroring) due to reliability concerns. I've seen just WAY too many
hard drives die to consider RAID-0 (stripping).

read that the raid levels are what are important...for example level 5 (block interleaved
distributed parity) is supposed to be the best

isn't the time it takes memory data <-> cpu the same (same bus speed) but the total time
is reduced b/c the controller is in the cpu and likely *knows* what the cpu will need /
send reducing the time of cpu-controller communication....i hope this is the case:)
This translates into about a

Similar idea, though you'll have a HELL of time finding any chip that
doesn't have L2 cache built-in.

except for my house:)
(256 kb cache ram)
The last x86 chips I know of that
used external L2 cache were the original Athlon chips, discontinued in
late 2000, and these chips had their L2 in a little cartridge
alongside the processor, not on the system board.

All current processors have their L2 cache on-chip.


They are partly related, though it's a bit more complicated than that.
The integrated memory controller and lower memory latency time of the
Athlon64 mean that it's less dependant on getting data from it's L2
cache, so it can get by with less cache (or conversely, the higher
memory latency of the P4 means that it's more dependent on L2).

As for total system memory, there isn't much reason why that should
change one way or the other. Actually the Athlon64 supports MORE
memory than the P4 (8GB vs. 4GB), but in practical purposes the
difference doesn't amount to much.

the ones i read about (nforce3) hold max 3 gbs...

this board supports either ddr or ddr2 (it states [under the specs] that it
supports dual, single )
i'd be using ddr initially
also the max is 4 gb
not sure whether 2 ddr2 chips are enough?
(i also found that it supports 2*ddr2 OR up to 4*ddr)

Probably. DDR2 still hasn't really made a big impact on things and
likely isn't going to for a while yet. Personally I would mostly just
ignore the DDR2 for the time-being, it costs more but is no faster and
DDR will be widely available for a long time (in computer terms, ie
3-5 years) to come.
aren't these the ones that have the 1066 mhz front side bus speed?
if they are not, i could still get the i915g board.......

Nope, the only chips with the 1066MT/s bus speed are the P4 Extremely
Expensive Edition chips. The 600 series P4 chips have the same
800MT/s bus speeds as the older 500 series P4 chips before them. they
should work on most/all i915G boards. They definitely will work on
that Asus board you had mentioned earlier.

i'll look at them...
Ohh.. speaking of that board, the Asus P5GDC-V Deluxe (I think that
was the one you were thinking of?) does NOT support integrated video.
Even though it comes up under their i915G chipsets, it's actually
using the i915P chipset. I don't know if this is a flaw in their
website or just the way they have things setup, just a word of warning
though. It's rather confusing and I definitely missed this one the
first read through it.

all the reviews i've read state it has onBoard video chip...

thanks very much again for answering and for the info!
sincerely
Tanya
 
hi keith,
[...below...]
On Mon, 28 Feb 2005 21:42:46 -0500, Tanya wrote:


Fire away! BS about differing processors/systems is what we do here. ;-)



Have you looked at the D and I caches on the P4? The cache latencies?
No, like life, it's not the size that's important, but what you do with it.
;-)

no i have not...are you talking about the l1 instruction cache and the l1 data
cache?
Frequency is next to irrelevant for processors with differing
micro-architectures.

i now know that......
Benchmarks. YEs, AMD has regularly kicked Intel's but in instructions per
clock. The P4 is particularly bad in this regard, as AMD will gladly
point out. IMO, the P4 was a horrible design, not suited for the market
it ended up in. The P3 is a better choice and AMD out-does even that.


Less RAM? My Opteron has 1.5GB and it (as opposed to my CFO) would easily
go to 8GB, which would be a problem on any P4. ;-)

i posted what i read in the other post... which board(s) hold the 8gb?
Hang around. Lots of fun stuff happens around here. ;-)


The FSB is meaningless when comparing the Intel and AMD offerings. I'd
like to see what L2's you're comparing. ...notto mention the L1s, whichh
I believe you're ignoring.

i stopped ignoring them however i still don't understand why the integration of the
mem controller reduces the need for cached mem... (i won't repeat the question but
basically, since the ram is in the board, the data from ram still needs to get to
the cpu -- (i.e. it is still separate from the processor).
The bottom line is that you have to compare
*performance*. The microarchitectures of modern processors are just too
different to compare raw numbers.

thanks alot,
sincerely
Tanya
 
hi George
[...below...]

George said:
I think she means the larger L2 cache in P4s - the 600s have 2MB... the
message being that Intel had to go there to err, catch up and try to hide
its latency... with umm, bandwidth?<titter>

i do mean that
i read that amd's have larger bandwidth; also that amd's use hyperTransport (up
to 2000mhz (system bus technology) (which is compared to the p4's fsb speed

thanks,
sincerely
Tanya
 
hi George
[...below...]

George said:
I think she means the larger L2 cache in P4s - the 600s have 2MB... the
message being that Intel had to go there to err, catch up and try to hide
its latency... with umm, bandwidth?<titter>

i do mean that
i read that amd's have larger bandwidth; also that amd's use hyperTransport (up
to 2000mhz (system bus technology) (which is compared to the p4's fsb speed

Well the Hypertransport speed is not really comparable to Intel's FSB as to
the data carried on them in a uniprocessor system. In an Intel system, the
FSB is only a "bus" in that it can have more than one CPU on it in a
multiprocessor system and it connects the CPU(s) to the MCH (Memory
Controller HUB... what we used to call the North Bridge); the MCH in turn
connects to the memory channel, to fast I/O devices, namely PCI Express and
to the ICH (IO Controller Hub) which handles all the other I/O devices.

Since the AMD Athlon64 systems have the Memory Controller on the CPU die,
CPU <-> Memory transfers do not have to travel on the Hypertransport Bus
*but* all I/O device <-> memory transfers do. Hypertransport is really two
buses twinned together, one for the up-stream and the other with the
down-stream; with current clock speeds, their aggregate bandwidth is close
to the same as Intel's FSB.

Take a look at the Data Sheets from both Intel and AMD - you'll find some
diagrams which illustrate how the data lanes fan out system-wise better
than any description I can give.

There are pros & cons to both ways of doing things but the big gain for AMD
is in the low latency access to main memory since addresses/data don't have
to cross external clock domains through an MCH. Intel tries to get around
this with their Hyper Threading and by agressive prefetching of data from
memory to the CPU's (usually larger) L2 Cache The bottom line is that in
overall system performance you're not going to see a lot of difference -
some apps wll favor one approach and others the alternative method. If you
get a decent system with either CPU -- I usually buy one notch down from
the leading/bleeding edge -- you're not going to fret over the difference,
which in most cases will be barely measurable.
 
hi and thanks again for answering!
[...below...]
The "front-side bus"
frequency matters, though not so much with the AMD64s because they don't
use the FSB for memory. The memory controller is integrated into the
processor so the "bus" is only for I/O stuff (on a single processor system).

(isn't the communication between the cpu and memory (which is determined via the memory
controller (integrated for amd; onMotherBoard for pentium4) called the fsb? i.e. the
"path" : cpu <-> memory))
(i.e. the mem controller determines what to get for the cpu -- right? well doesn't the
memory need to get to the cpu? (it is still on the motherboard so needs a bus to get to
the cpu?)
?

The classical definition of the front-side bus is the bus from the
processor to the north-bridge (chipset). It became known as the FSB
when the L2 cache was removed from the processor bus and moved to the
"back-side" of the processor.

Since the AMD64 processors have integrated memory controllers the
concept of the "north-bridge" and "front-side bus" is a little muddled.
There really isn't a northbridge or FSB, as such. The AMD 64
processors have the memory bus tied directly to the processor, where
Intel and earlier AMD processors have the memory bus tied to the
chipset (northbridge-half) and the front-side bus connects the
northbridge to the processor. In an Intel system the memory traverses
the FSB, thus its performance is important. In an AMD (uni-processor)
system it's less important since it isn't in the memory path. AMD64
multi-processors do have to use the bus(ses) (hypertransport) to access
memory on the other processor(s), so hypertransport isn't a slouch.

The main point here being that AMD64 processors have a memory latency
advantage because they don't have the extra trip over the FSB and
through the northbridge. System memory is hooked directly to the
processor.
i hear that plus there's a relatively long latency...

Which is why it's a waste of money. ;-)

isn't it more expen$ive?

George MacDonald tells us that PCI-E cards are cheaper (I.e. AGP has
already seen the other end of the bathtub). I don't do 3D games so
don't much care about either. ;-)
i wouldn't even know what to use dual monitors for (this is where the human bottleneck is
having to look at 2 monitors simultaneously <ASAP> <lol>)

You have two eyes don't you? ;-)

Seriously, I use two monitors both at work and home. Actually I have
three monitors in front of me now. One of the monitors is connected to
a RISC box, but I'd find room for a few more if I could. Big desktops
are very nice. Be warned though, once you try it you won't go back!
i got a question here (i cannot find the article though -- i read that there are several
raid levels and some are great in keeping *copies* soToSpeak of the drive's data...)

Sure, but none will protect you against the biggest source of data
loss; the loose nut behind the keyboard. RAID only protects against
one source of data loss - the hard disk itself.
i believe that it would be. the way i understand it (which is likely more fiction:) is
that it calls data from ram as per the cpu's direction HOWEVER if it is integrated it can
*arrange* to get / send memory data on it's own being the cpu???)

The request doesn't have to go from the processor, over the FSB,
through the northbridge, to the DRAM, and back. The FSB and
northbridge are eliminated.
i actually asked amd for some tech articles on why their cpu is <almost> always
recommended / preferred and i have some interesting info from them which definitely
prooves their cpus are better performers.
i believe that amd is not in <as much> flux (at least not as much as intel is) i.e. is not
between chip architecture -- i still like intel however, their 1066 mhz fsb is new (and
EXPEN$IVE) and likely needs fine tuning --

AMDs FSB is *infinitely* fast. ;-)
i'm getting the "max memory" from amd-based motherboards: for example: aopen n250a-fr "max
mem" 3 gb.
asus k8n-e dx "max mem" 3gb.
msi k8n neo ms7030 "max mem" 3 gb
but for other boards i see 4 gb max............

How about: http://www.tyan.com/products/html/matrix.html

Is 32GB enough? ;-) Seriously, that's for a (serious) 4-processor
board, but others are 8GB per processor too. 2GB per stick, four
slots...
only 1 problem: which board??? and i guess the nforce4 chipset... is asus good? (it is
supposed to be for p4 systems.)

Ah, there's the $64,000 question. Asus has a good reputation and I've
built several systems with them. I prefer Tyan these days, but don't
pretend they're the only manufacturer out there.
p.s. 1 more question: i read an article on cpu's (in general) which states that the more
transistors, the better (they list intel based cpus) in some of the info on amd's site,
the amd athlon 64 has 105.9 * 10^6 and the p4 has 125 * 10^6
perhaps it is the way transistors are used not their absolute number?

It's like the "23 jewel" watch a friend once showed me. It was a
standard 17 jewel swiss movement with six jewels taped inside the back
cover.

Of course more transistors isn't better. They cost (small) money to
make and they dissipate power. If they're used for something useful
they may be interesting though. Using your example above, it seems
that the P4 has ~20E6 transistors taped inside the case. ...maybe
that's what they really meant by having "taped out". ;-)
 
hi keith,
[...below...]
On Mon, 28 Feb 2005 21:42:46 -0500, Tanya wrote:


Fire away! BS about differing processors/systems is what we do here. ;-)



Have you looked at the D and I caches on the P4? The cache latencies?
No, like life, it's not the size that's important, but what you do with it.
;-)

no i have not...are you talking about the l1 instruction cache and the l1 data
cache?

Yes. The P4 has an interesting Icache, but it's tiny. You really have
to crawl through each cache to see what's going on. Like processor
clock frequency, just the size alone doesn't mean much.
i posted what i read in the other post... which board(s) hold the 8gb?

Perhaps one has to go to socket-940 (registered memory) to get to 8GB.
THe Asus socket-939 boards apparently do 4GB.

http://usa.asus.com/prog/spec.asp?m=A8V-E Deluxe&langs=09

i stopped ignoring them however i still don't understand why the integration of the
mem controller reduces the need for cached mem... (i won't repeat the question but
basically, since the ram is in the board, the data from ram still needs to get to
the cpu -- (i.e. it is still separate from the processor).

Look at it another way; to make up for the longer latency of an
external memory controller intel had to add more cache. You can try to
hide latency, but that's all you can do - try.
(i won't repeat the question but
basically, since the ram is in the board, the data from ram still needs to get to
the cpu -- (i.e. it is still separate from the processor).

With an external controller the data has to make four hops rather than
two.
 
Back
Top