Either you're very naïve or very optimist. Athlon64FX and Athlon64 have their
ups and downs and there are benchmarks that show that in some areas in spite of
their 64-bit design, they don't perform as good as Pentium4.
As for "in some areas", I said as much, so I don't quite understand
what's your point here, but never mind.
As for "in spite of their 64-bit design"...
First of all, all benchmarks are 32-bit. As is the OS.
Secondly, you seem to assume there is some inherent performance
advantage to 64-bit adressing. Well, there isn't.
That doesn't mean 64-bit operation won't be faster, it will, but more
about that later.
With the Hyper-Threading Intel Technology, AMD is no longer the king of the
jungle.
AMD stopped being percieved 'king' when published benchmark suits
started to emphasize media use and SSE2 optimization.
And eventually, when those applications benefitting from SSE2 was
optimized for the P4.
Even the old 32-bit K7 AthlonXP is still (much even) faster than even
the best P4 "in some areas" (as you put it).
Sofar, HT, is still mainly a tool for benchmark upmanship.
Increased parallellism in the same cpu package is a good idea though,
and maybe soon the best use of additional transistors on ever larger
chips. This is an initiative I like very much. Eventually, we are
going to see real performance gains from this. Software too, is going
to change, in a direction that is better in the long run.
Having said that, Intel CPUs are more stable and reliable than any AMD
CPUs. Even though AMD doesn't seem to have software incompatibility issues, all
applications on the market are written for Intel CPUs. Meaning, the Intel CPUs
are the norm for applications whereas AMD users hope not to hit the one in a
million software incompatibility jackpot.
- What a load of utter bullshit!
"the one in a million software incompatibility jackpot."
- ¤&@%€µ*!!!!!! (and other unprintable exclamations)
- You really still believe in this shit propaganda!?!?
And what's your point? Like "Don't buy AMD"?
What have this to do with anything?
Not to mention that AMD CPU tends to have overheating problem.
Actually, it's not any more severe than Intels. It's pretty much
exactly the same amount of Watts. It's just that a lot of Intel
customers are not really concerned about real performance. It's often
just percieved performance and GHz that counts. That's why P4-Celerons
are possible to sell. So it makes perfect sense to just throttle the
cpu when it gets hot. Nobody seems to notice anyway.
The motherboards for Intel CPUs performs better and
more stable.
Not an issue IMO. Buy the good ones, whatever your cpu choice. My AMD
track record is 100%. My Intel is down to about 80%, but that includes
some bad luck and a case of overheating due to a fan stopping, so I'm
not really concerned.
Intel manufactures CPUs, and knows inside out everything down to
undocumented secrets about its products, and they produce their own motherboard
chipsets for their own CPUs whereas AMD uses some chipset from a third party.
But what is really interesting is this: What is your point? Like
"Don't buy AMD"? Again?
Well, people at Intel already know that they have to provide 32-bit
compatibility in their 64-bit CPU if they want to stay in business. Intel's
previous attempts and approaches for 64-bit CPU design were nothing more than
brainstorming to find their direction in the right course. Intel has some
ongoing projects for the new Intel 64-bit CPU.
You could say that again. ...Somewhat differently worded! The Itanium
was originally planned to conquer the world from about 2000 onwards.
It's now being 'reengineered' for the third or fourth time. Each
iteration gets bigger and more expensive. ...And not enough faster to
even make up for lost time. Nobody but HP even bothers to offer
Itanium workstations anymore. With IBM power5, Apple G5, AthlonFX and
Opteron it makes less sense than ever. As I said, "...complete idiot
to buy..."
The problem with Intel is that it always stuck to its old CPU design and
architecture with only minor improvements by adding a few new features. Intel
kept bumping up the CPU speed without any difference in CPU design.
More bullshit (sorry, but it is).
Intel have done five different logic cores for their '386 class CPUs.
The first one was the '386.
The 386 defines the basic instruction sets and memory modes for all
later Wintel CPUs.
The second was the '486 family.
The 486 introduced pipelining, and started to dispense with microcode.
Third are the Pentium and PentiumMMX.
These did pipelining in earnest and also introduced 64-bit memorybus.
Also rumoured to be mostly based on technology stolen from DEC.
(DEC sued for billions, but I don't know how that resolved. Doesn't
matter much. Basically Intel is one of those rich enough companies,
for which crime actually always pays. No matter what.)
Fourth are the PentiumPro, PentiumII, PentiumIII, PentiumIIIe, Xeon,
Celeron, Tualatin.
These too are rumoured to incorporate lots of tech stolen from DEC.
Introduced codefission (technique pioneered by AMD) in order to be
able to use (DEC?) RISC logic. Better prefetch, branchprediction and
speculative execution. Also introduced closely connected L2 cache, in
package and eventually onchip. An excellent cpu. Possibly one reason I
got so bitterly disappointed with the P4.
Fifth are the current P4 family and P5/Prescott.
This wasn't planned. Intel originally intended PentiumIII to be their
last 32-bit cpu. AMD and the Itanium failure forced Intel to do
another. Intel identified media as the performance critical use and
bottleneck of the future mainstream desktop PC. This typically
involves performing repetetive simple nonconditional operations on
large blocks of data. Major concern was that the new core 'Willamette'
should be fast on that. Traditional GP integer operations came
secondhand. And indeed, the P4 is much less efficient and slower than
the earlier PIII on that. I've heard that out of two competing design
approaches, the slower, but higher clockable design was chosen on the
insistance of the marketing department, that specifically wanted 'GHz'
to fool customers. Another reason that may well be stronger, was that
it was the simplest design and offered less 'technological risc'.
After the Itanium experience and considering the immediate threat from
AMD, that makes lots of sense.
Major features of the P4 are very deep pipeline and very short and
narrow transistor chain trees, making it possible to clock fast. It
also, in praxis, almost dispenses entirely with effective
branchhandling. Which have effects on performance, when criteria
traversing datastructures, pathfinding and AI.
And all these five cores are radically different from each other, and
are each designed from scratch.
... but on the other hand, people at AMD did something so radical by
totally building a new CPU from scratch with its uniquely architectural design.
Such uniquely architectural design allowed people at AMD to expand such CPU
technology to 64-bit. Their expansion from 32-bit to 64-bit CPU is the same
concept of expansion from 16-bit to 32-bit CPU.
Well. The logic design is all new. As is the functional architecture.
Has to be, since it supports new instructions and a new adressing
scheme. But the hardware execution concept in the K8 is very similar
to the K7. Except for the 'HyperTransport' memory controler.
There are some fundamental differences between the 16/32 bit
instructionset expansion of the '386 and the 32/64 extension of the
A-64.
With the '386 it was mainly a question of finally, long overdue,
letting the Wintel PC use linear adressing, just like everybody else
did.
With 32/64, the linear adressing isn't an issue, it stays. Instead
it's about creating a larger process space and provide a newer more
modern instructionset using more registers. And that is equally true
for Intels 64-bit efforts (though Intel also tried a new
branchprediction scheme).
The idea to let the memory mapping handle both instructionsets in a
single cpu, is the only thing that is similar to the '386's 16/32 bit
expansion.
And it wasn't 'allowed'. It was the intended design goal from start.
The expansion to 32-bit was not revolutionary as the same thing applies to
64-bit expansion.
Uh? 32-bit was definitly a revolution! (for the Wintel PC).
64-bits are necessary since we're running out of space. And in a
number of ways it's going to be interesting and exciting, but it's not
a revolution like getting linear, protected mode adressing.
In the ways it is going to be a revolution, it's more due to more RAM
and large HDs, provided by Moore's law. 64-bit is of course necessary
to make use of it, but it's the RAM providing the 'revolution'.
In order to maintain 32-bit application compatibility in
64-bit CPU, AMD people followed the same path that has been followed 16-bit
application compatibility in 32-bit CPU.
Yes, in a way they did. But you're not thinking this is about
registers, are you? It's about the length of the byte adress used by
cpu instructions to refer to data. It's not width. In that sense, our
32-bit cpus are already '64-bit' and the A-64 '128-bit'.
Depending on what kind of direction Intel takes in its 64-bit CPU design, there
might be fork on the road, and serious issues of incompatibility among 64-bit
AMD and Intel CPUs. If Intel doesn't adopt such 64-bit register expansion that
is similar to 64-bit AMD CPUs with such compatibility, every 64bit application
has to be written specifically for both of these CPUs, and no software company
will write two different versions of the same application. They will stick to
the CPU that is compatible with 32-bit applications because that's what all
consumers want.
Software companies might adopt 64-bit Athlon as a norm for their future
applications if Intel fails to provide 32-bit application compatibility in
their 64-bit CPU. Under such threat and pressure, Intel might have to go along
with the similar adoption of the AMD technology in their 64-bit CPU.
Yes, you're touching some areas of worry here. Yes, we would like to
have a single 32/64 bit way ahead. Intel seem to still want that to be
the Itanium. Only way that is going to happen is AMD dies.
Intel are rumoured to have a simple AMD compatible 64-bit cpu design,
as a fallback. My belief is that this project is pretty much dead, and
that the stopgap solution right now, is trying to slap on some crude
64-bit shell around the P4/P5. Anyway, Intel holds their cards close.
They don't want to give AMD credibility.
AMD is in a precarious financial situation and Intel would dearly
like to see AMD go belly up now. It's a good time since Intel doesn't
have it's act together. So Intels PR department pushes on with dissin'
and downplaying AMD with media, and by controling benchmarks.
We see lots of smoke for the clueless here, like P4EE and like
suggesting 32-bit paged adressing to form 36-bit adresses. (It's
insane. It's incompatible with everything, OS'es, drivers, AGP
videocards. And it would be as bad as 16-bit. So it's never going to
happen. It's just BS.) The idea is to shift awareness and focus away
from AMD and 64-bit and create the illusion that P5 and P4EE are great
for the future.
I think AMD, without a doubt, has already provided the best means of 32-bit
application compatibility in their new 64-bit CPUs in a very convenient and
practical way.
It is such a big loss that AMD for some reason is not interested in
manufacturing motherboard chipsets for their own CPUs.
- Aah, but then you have missed it! Much of the 'chipset' have moved
inside the Athlon64. So they are indeed making their own. In a way.
Athlon64 pulls a little ahead of Pentium 4 in "Direct3D Gaming" mostly due to
the ATI Radeon 9800 Pro video card. Even with the use of the best video card,
Athlon64 pulls a little bit ahead which is not anything so significant.
Err... While the P4 cannot use 9800pro?...
Sorry, but you're BS'ing again. Games are an A64 advantage. Don't let
that P4EE blind you (it's the intention, but it's all a trick. In the
real word, larger caches gives singleusers & singleCPU diminishing
returns. You can convince yourself about that by comparing some
3.2GHzP4 vs P4EE applications. Now many benchmarks, on the other
hand...) You know there _IS_ a reason cpu manufacturers haven't ever
bothered to slap on gigantic caches before.
Let's look at the benchmark test at
www.extremetech.com
1- In "Content Creation Winstone 2003" benchmark test, Pentium4 pulls ahead of
all Athlon64 CPUs.
2- In "Quicktime 6.1 Encoding (MPEG-4)" benchmark test, Pentium4 pulls ahead of
all Athlon64 CPUs.
3- In "XMPEG5 / DiVX 5.1 Video Encode" benchmark test, Pentium4 pulls ahead of
all Athlon64 CPUs.
4- In "Windows Media Encoder 9" benchmark test, Pentium4 pulls ahead of all
Athlon64 CPUs.
5- In "AfterEffects 5.5 Rendering" benchmark test, Pentium4 pulls ahead of all
Athlon64 CPUs.
6- In "Music Encoding" benchmark test, Pentium4 pulls ahead of all Athlon64
CPUs.
7- In "3D Studio Max 5.0" benchmark test, Pentium4 pulls ahead of all Athlon64
CPUs.
8- In "Cinebench 2003 CPU Rendering" benchmark test, Pentium4 pulls ahead of
all Athlon64 CPUs.
9- In "LightWave 7.5 Render" benchmark test, Pentium4 pulls ahead of all
Athlon64 CPUs.
10- In "PCMark 2002 CPU" benchmark test, Pentium4 pulls ahead of all Athlon64
CPUs.
11- In "PCMark 2002 Memory" benchmark test, Pentium4 pulls ahead of all
Athlon64 CPUs.
What are you intending to prove with this list? Prove that I was right
when I said the P4 remains competitive on mediaencoding? Thankyou very
much then. But one mpeg benchmark suffice, the rest are redundant.
(And yes, unfortunatly, PCmark 2002 CPU is just yet another P4
optimized mediaencoding benchmark. It would have been nice if they had
measured other types of cpu performance instead, but alas.)
But hey, have a look at this, for what's in store when the A-64 does
mpeg encoding in 64-bit environment:
Linux 64-bit, Lame 3.93 MP3 encoder.
400MB encode, lower is better. A-FX
64-bit op 2.02
32-bit op 3.07
....And here's some examples of also optimizing 32-bit mediaencoding
for the A-64, instead of just running the P4 benchmark, as is soo
traditional...
DivX encode, lower is better.
A-FX(64) 7.5
A-FX(32) 8.8
3.2 P4C 9.2
1024-bit RSA encrypt, lower is better.
A-FX(64) 1.6
A-FX(32) 3.1
3.2 P4C 4.9
(these did came from AMD, so you might want to be a bit cautious
here...)
(7&9): As for the 3D image renditions: Optimizing these for speed and
the P4 comes at a penalty. Consider raising quality configurations:
Lightwave 7.5 Raytrace, lower is better.
A-FX 87.9
A-64 2.2GHz 88.3
3.2 P4C 93.1
A-64 96.4
(1): As for "Content Creation Winstone 2003"
higher is better.
3.2 P4C 51.5
A-FX 48.7
A-64 48.1
"Pulls ahead"? You said? By a nose, maybe. But this benchmark is
optimized for the P4 but not for AMD. There is an AMD mediaencoder
patch that changes this...
(11): As for "PCMark 2002 Memory" your claim is false. This benchmark
does not correctly measure P4EE memory performance as it runs against
the cache. The rest of the story is:
higher is better.
A-FX 11599
3.2 P4C 9989
A-64 8733
Obviously I saw these benchmarks long before you ever did, and they
didn't stop me from saying 'overall' fastest. Do you wan't me to
compile some 'similar lists'?
Sciencemark 2.0
Memory Bandwidth, higher is better:
A-FX 5315
3.2 P4C 4025
A-64 2954
Sciencemark 2.0
Memory latency, lower is better:
A-FX 57.35
A-64 61.54
3.2 P4C 72.48
Business Winstone 2002:
Business application&general performance, higher is better.
A-FX 43.5
A-64 43.5
3.2 P4C 34.7
Visual Studio 6.0 compile test, lower is better.
A-FX 14.1
A-64 14.9
3.2 P4C 17.3
Mathematica 5, lower is better
A-FX 5.7
A-64 6.2
3.2 P4C 7.4
Raw FPU power, Sciencemark 2.0 BLAS double precision.
(Matrix multiplication, btw AMD advantage on solving systems, is even
much greater)
'387 FP, higher is better
A-XP3200 3178
A-FX 3112
A-64 2803
3.2 P4C 2015
(-And what about that ol' AthlonXP! Well, I've always been nag'n about
the AthlonXP packing a mean '387 FPU!)
SSE2 scalar FP, higher is better
A-FX 3093
A-64 2776
3.2 P4C 1350
(on single precision and vectorized SSE2, the P4 is better though.
Which shows off in media encoding. This is another thing that might
change with 64-bit ops.)
Diep chess, (highly tuned for the P4. And here's an app that really
uses HT well!) higher is better
A-FX 134
A-64 124
3.2 P4 (x2 HT) 123
A-XP3200 112
3.2 P4 101
More serious numbercrunching.
ScienceMark Molecular Dynamics
(P4-optimized, by the way), lower is better
A-FX 73.8
A-XP3200 77.9
3.2 P4C 78.9
A-64 82.8
More serious scientific simulation (well, if PC-rags and websites can
do pages after pages of virtually identical mpeg/jpeg benchmarks,
giving the impression that Intel and P4 rules, why can't I milk math &
physics a bit? Eeh ;-))
CPU Plasma
(again, P4 optimized), lower is better
A-FX 406
A-64 508
3.2 P4C 588
A-XP3200 616
ScienceMark 2.0 Cypher AES.
(good ol' integer code, no chance for the P4 to hide it's ugly side.
This is how the P4 performs in between mpeg and data shuffling.
Disgusting isn't it! This is a hint to as how the P4 performs on AI
and pathfinding, something few "game" -benchmark shows off.)
lower is better.
A-FX 13.1
A-64 14.4
A-XP3200 14.7
3.2 P4C 23.9
Now all gamers.
3DMark 2003, higher is better
A-FX 876
A-64 760
3.2 P4C 743
Gunmetal game framerate, higher is better
A-FX 71.5
A-64 65.5
3.2 P4C 61.5
Unreal 2, higher is better
A-FX 76
A-64 68.8
3.2 P4C 61.2
Geoff Grammonds GP4, higher is better.
A-FX 46
A-64 45
3.2 P4C 36
Wolfenstein ET, Ace Hawski, higher is better.
A-FX 85.3
A-64 81.4
3.2 P4C 79
Let's do a Database too. Got to try to be complete.
MS Access, lower is better
A-FX best 5.1 worst 5.5
A-64 best 5.9 worst 6.4
3.2 P4C best 6 worst 6.5
A-XP3200 best 6.3 worst 6.7
Frankly, I'm not sure why I bothered...
The benchmarks are there, on the websites, for anyone to read for
themselves. Yes, the P4s does well on many of them, for various
reasons, not all of them 'valid'. There's an enormous amount of
mediaencoding benchmarks there, all measuring basically the same
thing. So unless you take out the redundancy, the P4 probably wins any
'count'. Anyone will have to decide what is relevant for him/her.
The A-FX is in all cases the Athlon64FX 51. The A-64 is the Athlon64
3200+. The 3.2 P4C is of course the 3.2GHz@800FSB Pentium4.
I have culled the Athlon64FX 53 and P4EE. Both because of their
expense and because I believe many current benchmarks will tend to
inflate the performance one can expect in practice, from the P4EE.
The big L3 cache simply hides the fact that the Athlon64 is a
fundamentally faster cpu. Which is of course, exactly Intels
intention.
Finally, I think such a 32-bit development is greatly undesirable. Why
pay so much money on 32-bits and a large cache? It doesn't make any
sense. It's certainly not what we want for the future instead of
64-bit.
Well, yes and no! No immediate performance gain! You will notice significant
performance gain once software companies start utilizing the features of the
64-bit CPU.
Yes, that is probable. Because it's a new instructionset with new
registers. And because of the 128-bit memorybus. And possibly, because
memory mapping might be faster. Significant, yes. Major, like 16 to
32-bit, no.
And this is not the interesting part. That is the process space, to do
things not possible with 32-bit, ...at any performance.
ancra
