Intel had a great period the months before AMD’s launch of the K8 core where AMD had troubles catching up with Intel at the top with their somewhat aged K7 architecture. But AMD’s wholehearted bet on the K8 architecture has undeniably shown itself to be just right. Since the introduction of Athlon64 and Opteron which were the first K8-based processors the interest and popularity has increased steady for AMD.
AMD has since the launch of the first Athlon models been a force to recognize within the CPU market. Before the ”Athlon era” AMD was a relatively quiet processor manufacturer which fed from crumbs of Intel’s table. Even if AMD’s K6 series wasn’t bad it lacked that extra which could have made it a threat against Intel’s Pentium series.
But then came AMD’s launch of the K7 core and the Athlon name was in use. AMD delivered very powerful processors which in addition to good performance could be bought at good prices, furthermore it didn’t hurt that the overclocking potential was often very good, not at least because of its unlocked nature. The rest is as we say history and if there is something negative that we can say about AMD’s K7 architecture it might be that they tried to live on it a little bit too long.
Intel had a great period the months before AMD’s launch of the K8 core where AMD had troubles catching up with Intel at the top with their somewhat aged K7 architecture. But AMD’s wholehearted bet on the K8 architecture has undeniably shown itself to be just right. Since the introduction of Athlon64 and Opteron which were the first K8-based processors the interest and popularity has increased steady for AMD.
Despite that AMD’s maybe largest drawing card, the 64-bit support, still is more or less unused because of Microsoft’s delayed operating system (Windows XP 64bit Edition) the K8-architecture (also called AMD64) has been praised by most people and not at least consumers.
Except some complicated platform changes AMD has just begun its real evolution within the AMD64-architecture. For they have begun to manufacture their processors with 90nm manufacturing technology which is a great step against today’s 130nm technology. The transition to 90nm is not a simple process though and even if the great step is taken it will be a running-in period for the new manufacturing process and AMD has also taken the chance to launch two new flagships on the desktop market which are actually based on 130nm manufacturing process.
It is these two monsters for processors that we shall look closer on today, and for you who have read our earlier processor reviews you might guess that it’s about Athlon64 and Athlon64 FX. More detailed, Athlon 64 4000+ and Athlon 64 FX-55.
We shall inspect these beasts in their seams and look closer on what has happened since last we looked at these processor series, which shows to be quite a lot.
With that said we move on directly to the main character in the drama.
For those who want to read on about the AMD64-architecture itself we can recommend that you go back to our review of the Athlon 64 3200+ where we go deep into that subject.
When AMD launched their desktop models of AMD64 it was in two different series, Athlon 64 for consumers with great demands and good finances and Athlon 64 FX for consumers/enthusiasts with very good finances.
But it was also much more than just the cost that separated the two processor series. This is what it looked like at the very first beginning.
| Athlon 64 | Athlon 64 FX | |
|---|---|---|
| Processor platform: | Socket 754 | Socket 940 |
| L2 cache : | 1 MB | 1 MB |
| Memory support: | DDR-SDRAM | Registered DDR-SDRAM |
| Memory controller: | DDR 64-bit (single channel) | DDR 128-bit (dual channel) |
| Unlocked multiplier: | Yes. but only lower than original | Yes, completely unlocked |
Thus the processors had a number of physical differences and it was not difficult to tell them apart. This way it looked when we looked at Athlon 64 and Athlon 64 FX the last time and then Athlon 64 3400+ was compared to Athlon 64 FX-51.
Though, in 2004 AMD did a prominal change in their Athlon 64-venture, they aimed for a completely new processor platform. They locked Socket 940 in a server room which most of people has wanted since day one, since Socket 940 basically never was a good solution for the desktop market. And Socket 754 was deported to AMD’s budget processors, Newcastle and Sempron.
To additionally increase the platform rave for Athlon 64 they introduced a completely new processor format, Socket 939. The new platform took the positive from both the predecessors and consolidated these into one package, support for dual memory channels (heritage from S940) and support for ordinary DDR-SDRAM modules (heritage from S754).
When they moved over the processors to Socket 939 they also cut the L2-cache on Athlon 64 in half so that the Athlon 64 FX would have something more in advance other than just the clock frequency performance-wise. After the move to Socket 939 it looked like this.
| Athlon 64 | Athlon 64 FX | |
|---|---|---|
| Processor platform: | Socket 939 | Socket 939 |
| L2 cache : | 512 KB | 1 MB |
| Memory support: | DDR-SDRAM | DDR-SDRAM |
| Memory controller : | DDR 128-bit (dual channel) | DDR 128-bit (dual channel) |
| Unlocked multiplier: | Yes. but only lower than original | Yes, completely unlocked |
Now it was only the L2-cache’s size and the multiplier lock that separated the two architectures.
But now AMD has, through the launch of Athlon 64 4000+, closed the circle more or less completely. Athlon 64 4000+ has regained its complete L2-cache (1MB) and the only thing that separates 4000+ against Athlon 64 FX-55 is that the latter is unlocked and is being delivered with a higher clock frequency. In other words the Athlon 64 4000+ is an exact copy of Athlon 64 FX-53 (disregarding the multiplier lock) since both processors use the same basic architecture and are being delivered with a clock frequency of 2.4GHz.
It still remains to see if AMD will implement 1MB L2-cache even on the lower clocked Athlon 64-processors. But it doesn’t look that way today which we soon will explain further. It is not impossible though that they choose to launch a new model of Athlon 64 4000+ with 512KB L2-cache and 2.6GHz clock frequency, but then with 90nm manufacturing technology.
Now it’s time to look closer at the two processors we test today, Athlon 64 4000+ and Athlon 64 FX-55.
We have already revealed that the processor architecture of the Athlon 54
4000+ and Athlon 64FX-55 are identical and here is the proof.
 |
 |
 |
 |
As seen is it only the effective clockfrequency that separates the two combatants
apart. The FX55 has been enriched with a slighty higher voltage because of its
higher clockfrequency but otherwise is it only the name and cpu ID that separates
them. As like as two peas.
We can also determine that AMD also has chosen their well tried 130nm manufacturing
technology for both Athlon 64 FX55 and Athlon 64 4000+. Their 90nm technology
is as mentioned before is on the move but there is still more to get from the
130nm technology, as they choose to introduce their first 2.6 Ghz cpu with this
coore.
The thing that seperates the two is as said only the multiplierlock and thats
something we will look into later in, but now its time to focus on the physics
of the cpus.
|
Athlon XP 1700+ : Athlon 64 FX-55 : Athlon 64 4000+ : Pentium 4 E 3.2GHz (S775)
|
Above we have a comparison of a couple
of cpu’s with clearly differnt physic. The older Athlon XP series with an ”bare”
core that does look awkward in comparison to AMD more slimlined Athlon 64 series.
Smallest however is the Intel Pentium 4-cpu for socket 775. Something to note
is that the Socket 775-cpus doesn’t have any pins on the cpu, they are placed
in the motherboards socket.
 |
 |
|
Athlon 64 4000+
|
Athlon 64 FX-55
|
AMD introduced its own heatspreader (metalplate that conceal the cpu core)
in connection with its introduction of Athlon 64. We don’t think anyone regrets
this, neither AMD, consumers or retailers.
Far to many Athlon XP-cpus has been killed because of its bare and sensitive
core (se picture above).
On the bottom side alot has happend to, Socket 939 stands for the 939 pins that
connect the cpu with the mainboard. On the Athlon XP there was 462 pins så
they have drasticly increased in numbers, much because of the integrated memorycontroller
which for Socket 939 is 128-bit wide. The 754- series only support one 64-bit
memorycontroller and accordingly there no need to send as much information between
the cpu and mainboard, so the pins are fewer.
Before we move on with the review we can determine that nothing has changed
in the basic AMD64-architecture. AMD has however played around with clockfrequencys,
memorycontroller and even the cachememory. In a matter of fact they have increased
the clockfrequency on the HyperTransport-bus which connects the cpu to the rest
of the system. When AMD introduced the Athlon 64 series the HT-bus clockfrequency
was 1.6GHz(800 Mhz DDR) but todays Athlon 64-cpus can brag with a 2.0GHz HT-bus
(1GHz DDR). This increases the cpu bandwith from 6.4GB/s to 8.0GB/s, along with
the memorycontroller 6.4GB/s it add up to a total bandwith of 14.4 GB/s which
is quite impressive. Whether the faster HT-bus make any difference from a performance
point of view is another matter, since you rarely press the HT-bus that heavy.
Physicly not much has happend since our last view on the Athlon 64-series. The
number of pins has changed and thereby also the plattform that the cpus use.
In our review of Athlon 64 3200+ we took a peek at the VIA K8T800 chipset that
was used in our Socket 754-testsystem at the time. In our Socket 939-system
its once again VIA that supplys that chipset, but this time it is an uppgraded
variant of K8T800, the K8T800 Pro.
Before we move on the the performance testing we shall inspect this chipset
a little closer, and also analyse other details of the testsystems we used for
this review.
Included in the testsystem we got from AMD was a mainboard from the manufacturer
MSI. The board is based on the VIA K8T800 Pro chipset and the name is MSI K8T
Neo2.
Honestly speaking there are no big changes in K8T800 Pro compared to its predecessor
K8T800, but a run-through of the chipset, and of course the new socket (939),
can hardly do any damage.
First and foremost, the VIA K8T800 Pro supports Athlon 64 processors with
the 2GHz HyperTransport bus, which makes communication between the CPU and
the
chipset’s northbridge very fast. The K8T800 Pro then has its own data bus
to tie the north-
and southbridge together in the chipset, called VIA Ultra V-Link. VIA Ultra
V-Link has a bandwidth of 1.06 GB/s, obviously not comparable to the processor’s
HT
bus at 8.0 GB/s, but still an improvement from the V-Link bus used in the
previous K8T800 chipset, which had a bandwidth of 533 MB/s.
In addition to the optimized data buses, the K8T800 Pro also supports AGP-
and PCI-lock. This enables the user to lock the clock frequencies of AGP
and PCI components, which often allows higher CPU bus speeds and thereby
better overclocking properties. This feature did not exist on the K8T800, and was missed by many overclocking enthusiasts.
We will probably not have any great use for this feature in our review, but
it is worth considering in future main board reviews.
Something we undoubtedly will be using, however, is the new socket on MSI
K8T Neo2. Below is a picture of how the socket itself looks on the main board.
With the number of pins on the processors in mind, it’s no surprise that the
socket more or less consists of tiny holes.
 |
After a short exposition of the main boards and chipsets, it is time to go
through the rest of the test system, where we also explain a few peculiarities.
|
Test system
|
|
|
AMD Athlon64 – Socket 939
|
|
| Processors: |
AMD Athlon 64 FX55 (2.6GHz, 1MB L2-cache, 2GHz HT)
AMD Athlon 64 FX53 (2.4GHz, 1MB L2-cache, 2GHz HT) AMD Athlon 64 4000+ (2.4GHz, 1MB L2-cache, 2GHz HT) AMD Athlon 64 3800+ (2.4GHz, 512KB L2-cache, 2GHz HT) AMD Athlon 64 3500+ (2.2GHz, 512KB L2-cache, 2GHz HT) |
| Mainboard: |
MSI K8T Neo2 (VIA K8T800 Pro) |
| Memory: |
1GB Corsair XMS3200 DDR-SDRAM |
| Graphics card: |
Sapphire Radeon X800 XT AGP 500MHz/500MHz(1GHz DDR) |
|
Intel Pentium 4 Prescott – Socket 775
|
|
| Processor: |
Intel Pentium 4 560 @ 3.8GHz (844MHz FSB, 211×18)
Intel Pentium 4 560 (3.6GHz, 1MB L2-cache, 800MHz FSB) Intel Pentium 4 540 (3.2GHz, 1MB L2-cache, 800MHz FSB) |
| Mainboard: |
ABIT AA8-DuraMAX (Intel 925X) |
| Memory: |
1GB Corsair XMS4200 DDR2-SDRAM |
| Graphics card: |
Sapphire Radeon X800 XT PCI Express 500MHz/500MHz(1GHz DDR) |
|
&Misc. hardware found in all test systems
|
|
|
Harddrive:
|
60GB IDE Seagate Barracuda IV
|
|
Power supply:
|
Enermax EG365AX-VE (350W)
|
|
Mjukvara
|
|
| Operating system: |
Windows XP Professional SP1
|
| Resolution: |
1024x768x32bit, 85Hz
|
| Drivers: |
ATI Catalyst v4.9
DirectX 9.0c VIA 4in1 v Intel Chipset Driver 6.001 DivX 5.1.1 Xvid 1.0.2 |
|
Benchmarking software:
|
Sisoftware Sandra 2004 |
As you can see, we have a fine collection of processors in this lineup, including
the whole top tier of AMD Socket 939 processors. From the Athlon 64 3500+ to
the new flagships 4000+ and FX55.
We are really looking forward to comparing Athlon 4000+ to Athlon 3800+, as they
work at the same clock frequency, but with different amounts of L2-cache. Athlon
64 FX53, which is most likely to be [phased out] by AMD shortly, is only in the
test to serve as a point of reference. It should provide identical performance
to the Athlon 64 4000+, but we can not know for sure without testing.
The Athlon 64 system used 1GB Corsair XMS3200 memory using 2-2-2-10 timings at
DDR400 speed, which is the optimal setting for the Athlon 64 platform.
If we take a closer look at the Intel Pentium 4 platform, we have focused
solely on the newer Socket 775 interface, and as the base of this testing system
we
have Intel’s own chipset, i925X, which supports DDR2-SDRAM.
Our board comes from ABIT and goes by the name ABIT AA8-DuraMAX. The system
was equipped with DDR2 memory from Corsair, which we gave as aggressive timings
as
we could (4-3-3-9 at 533MHz).
The processors tested on Intel’s 775 platform are Pentium 4 540 and Pentium
4 560. The latter is Intel’s official flagship and clocks in at 3.66GHz.
However, the Pentium 4 570 has started appearing in stores lately, which is an
Intel Prescott running at 3.8Ghz. This model has hardly been advertised at all
by Intel, and is yet quite difficult to come by. We couldn’t get a P4 560
to our test lab, but we have simulated this processor by overclocking the P4560
to 3.8GHz. This makes the bus speed of the processor slightly higher than the
stock P4 570, however. Therefore, the results from our 3.8GHz processor aren’t
completely accurate, but should still give us a hint on how the Pentium 4 570
performs compared to the other test subjects.
HyperThreading is activated in all tests unless stated otherwise.
First out among the
performance is as usual the synthetic memory tests. Here we shall take a look
at the processors’ memory bandwidth and also the memory’s own latency, waiting
time. The former is simply how much data the CPU can transfer between itself
and the memory in a given time span, while the latter is a measurement of how
quickly the CPU can start acquiring data from the memory.
Sisoftware Sandra 2004
First in the test suite we have the familiar Sisoft Sandra 2004, which focuses
on measuring the system’s memory bandwidth. Starting out, we have two diagrams
of buffered tests, where the maximum memory bandwidth is measured. Here all
performance enhancing processor specific technologies are activated, such as
MMX and SSE2 instructions.
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
We can clearly see hear
that AMD’s 128-bit memory controller (2 x 64-bit for two channels) for the 939
platform really pays off, and it doesn’t hurt that the memory controller is
integrated in the CPU core, working at the same frequency. The memory bandwidth
is very high on all systems as Intel too uses dual memory channels. Despite the
DDR2-533 memory having a theoretical bandwidth of 8.5 GB/s, the Intel system is
far from reaching these results. This is simply because of the fact that the
CPU’s own bus can only transfer 6.4 GB/s, thus causing a bottleneck in the system.
Consequently, AMD clearly wins this round but we move on with additional memory
tests, this time deactivating MMX and SSE support in Sisoft Sandra.
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
The results we get here are
actually rather confusing as this unbuffered test usually relies more on memory
timings than pure bandwidth. The DDR2 technology is far from efficient when it
comes to timings, but despite this Intel is the one dominating this test. With
this in mind it is with great anticipation we move on to Sciencemark 2.9 where
we will measure both memory bandwidth and latency.
ScienceMark 2.0
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
ScienceMark 2.0
First out is the memory bandwidth test where AMD Athlon 64 once again dominates
with its efficient memory controller, which now supports dual memory channels
for all processors in Socket 939. Intel slips behind despite the high bandwidth
of the DDR2 technology and it is obvious that Intel’s external memory
controller is not as efficient as AMD’s integrated one.
Now, if we look closer at the latency of the memory we see that AMD completely
crushes Intel with its integrated memory controller. As the memory controller
is integrated directly into the processor and working at the same clock
frequency, the access times become significantly shorter than on Intel’s
system.
We can also see that in the memory tests, very little separates the different
CPU speeds in both systems, and this is of course due to the fact that the processor
bus and memory bus still have the same clock frequencies.
It is obvious that AMD
should have a lead over Intel in memory-heavy tests thanks to its efficient
memory handling, but as we all know theory and reality may differ more than one
might think.
To spin further along this
line, we move on to a file compression test where memory is a key component.
In our file compression benchmark we’ve chosen to use the popular file compression software WinRAR. We simply took the latest version of WinRAR, which was WinRAR 3.4, and then ran WinRAR’s built-in benchmarking test that measures the systems ability to compress data per second.
This means that you can compare your own results to this test. You can find it under ”tools -> benchmark and hardware test”.
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
In our first test, which isn’t entirely synthetic, AMD once again turns out for the better. The integrated memory controller together with dual DDR-SDRAM’s channels offers the Athlon 64-processorers uninhibited performance in this test and Intel has to has to face the fact of almost being crushed beyond recognition. Even the overclocked 560-processor, that actually has both higher processor and memory bus, is left far behind.
You can get approximately two Athlon 64 3500+ processors for the price of one Pentium 4 570 (3.8GHz) and that is despite the fact that that the first one has about 18% better performance in this test.
Athlon 64 FX55 crushes the overclocked Intel processor with more than 34% and Pentium 4 560 is left behind with a 42% margin.
AMD couldn’t have gotten a better start and it’s pretty clear that AMD’s two new flagships are playing in a league of their own with the FX55 coming out on top. We can also observe that Athlon 64 4000+ and the FX-53 performs quite even, but even the 3800+ keeps up pretty well despite its smaller L2-cache. There is a 5% difference in rating between 4000+ and 3800+, but so far the 3800+ performs within a 3% margin thus the 4000+-rating might seem overrated at the moment.
Now it’s time for something that Intel usually dominates, multimedia.
We start off with some MP3 encoding, which means that we take a WAV file and compress it to MP3 format. This is done in a program called Audioactive Production Studio using the Fraunhofer codec.
Audioactive Production Studio – MP3 encodning
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Somewhat surprising AMD wins this test as well, but that could be due to the choice of application and codec. With a newer application it is quite possible that Intel’s NetBurst architecture might have gotten more scope. One thing that is clear, however, is the fact that clock frequency is more or less the only thing that matters here, as all three A64 processors at 2.4GHz finished the task in the same amount of time. Athlon FX55 once again struts its stuff, finishing the encoding in almost ten percent less time than the closest opponent. That is actually a marginally greater difference than that of clock frequency between the two processors, which is truly remarkable. There is no doubt that the Athlon 64 architecture scales well when increasing the clock frequency.
AutoGordian Knot – DivX, Xvid encoding
Now it is time for the next
multimedia test, video encoding. Here we used AutoGordian Knot, which
compressed a 34MB MPEG2 file with DivX 5.1.1 and Xvid 1.0.2. The file can
actually be obtained from our server, as it is a clip from Band of Brothers,
used in our DTV tuner review.
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Finally Intel has the opportunity to dominate and takes the top spot in both the DivX and Xvid encoding. Pentium 4 560, together with its overclocked self, wins both tests and even P4 540 sneaks up to a spot behind AMD’s flagship FX55 in the DivX test.
Intel’s NetBurst architecture is very rewarding at multimedia encoding, but for the sake of it we also tried deactivating HyperThreading in the video compression tests, as you can see in the diagrams. Without HyperThreading things are completely different and we cannot enough recommend all Intel users to activate HyperThreading (if it is not already activated) if you work a lot with multimedia, it gives plenty of free performance.
Another thing to notice is that the amount of cache memory apparently has different effects in the two tests. In the DivX test 4000+ has a lead of 4 percent, while 3800+ catches up in the Xvid test as the difference is only 1 percent here.
But in other words, Intel is still the king of video encoding, which doesn’t come as any surprise.
Futuremark PCMark04
We move on with a more synthetic test which still relies quite a bit on multimedia tests, namely Futuremark PCMark04. Here we will test both the whole system and the processor itself in two different test suites where, among other things, compression tests of audio and video are present (a complete list of the all the tests can be found at Futuremark’s home page). In other words, PCMark04 is supposed to give a good image of the system’s general performance. But please note that this is a synthetic test, nothing else.
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Once again we see Intel stepping up when it comes to multimedia and video compression. Intel takes all three top places in the processor test, while only taking the first two spots in the system test. AMD has problems keeping up but Athlon 64 FX55 is on Pentium 4 560’s heels, albeit at double the price.
Despite AMD’s victory in the audio compression test, it is plain to see that Intel takes the multimedia round, and now that we are in on the more synthetic tests we head on to a multitude of those with our first 3D graphics tests.
The first 3D tests we publish in this review are all from Futuremark, and the reason we show them separately is that they are entirely synthetic tests and should be viewed as such. Simply good indicators of performance, but with no direct connection to regular applications.
However, they are frequently used in the enthusiast community as performance tests, even by us here at NordicHardware in our performance championship, and so we shall certainly give you results from these tests as well.
3DMark2001 SE – DirectX 8.1
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
3Dmark2001 SE is not the video card dependent test it once was, but more of a decent indicator of system performance, and we can see that AMD more or less plays around with Intel here. All Athlon 64 processors, from 3500+ and up, are ahead of Intel’s equivalence. One thing worth notice is the performance difference between 3500+ and 4000+, where the latter gives a whopping 7.5% better performance, which is more than the model name would indicate. 3DMark2001 SE simply loves cache memory, so there is no surprise when AMD takes the victory with its integrated memory controller.
3DMark03 – DirectX 9.0a
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
3DMark03 is definitely more demanding on the video card than its predecessor and therefore the differences are very small, even when comparing between platforms. But in the CPU test, where the graphics card plays a smaller role, AMD once again takes the lead and leaves Intel behind.
3DMark05 – DirectX 9.0c
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
If 3DMark03 was hard on the video card, 3DMark05 (which you can read more about here) almost exclusively depends on it. Admittedly, the Athlon 64 processors take the lead, but the difference in performance is hardly noteworthy.
Now it is time to take a look at a few more 3D tests and this time it is mainly real game tests, even though we included two semi-synthetic tests.
We move on with our real game tests. However, we begin with two pure benchmark demos, both based on 3D engines from real game titles, first out is Aquamark3 which is based on the Aquanox engine.
Aquamark3 – DirectX 9.0(a)
First of all we would like to point out that Aquamark runs 4xAF by default, and
that this is the reason we run the benchmark with this setting activated.
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Aquamark3 is not the first test in this review to show good collaboration with AMD’s Athlon 64 architecture, and most likely not the last. AMD is at the top, both in the standard benchmark and in the CPU test, and even though the margins aren’t very great it is plain to see that Athlon 64 simply is a better platform than Pentium 4 Prescott in this test.
Continuing our 4000+ vs. 3800+ comparison we notice here that the difference in the standard test is 1.7%, and in the pure CPU test it is only a measly 1.6%. In other words, no 5% difference in performance here.
Comanche 4 – DirectX 8.1
The next test is also a separate benchmarking test, but as Comanche 4 is a game
as well we do not see it as a completely synthetic test.
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
We have been using Comanche 4 for a long time in our tests here at NordicHardware, in CPU, video card and motherboard reviews and more, as Comanche 4 is very demanding on most components in the system. It also appreciates high CPU speeds and fast memory access. Very suitable for Athlon 64, one might think, and so is the case if we look at our test results.
Athlon 64 FX55 crushes the opposition, and the other Athlon 64 processors perform nicely as well. Intel could in fact not surpass Athlon 64 3500+, not even with the overclocked 560 processor.
The difference between 4000+ and 3800+ is actually exactly 5%, meaning that Comanche 4 has good use for the 512KB of extra cache that the 4000+ offers.
UnrealTournament 2004 – DirectX 8.1/9.0a
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
UT2004 is still a very popular game, and has worked well with AMD’s platforms for a long time. Our results stick to that fact and Athlon 64 simply humiliates Pentium 4 Prescott.
We see similar differences as in Comanche 4, and the Pentium 4 processors are not even close to the Athlon 64 when it comes to performance. The difference between 4000+ and 3800+ is once again almost exactly 5%; AMD doesn’t seem to have been all wrong with the naming of its newest Athlon 64 CPU.
Quake3: Arena – OpenGL
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Before we move on with some newer game titles we have taken on an old acquaintance, namely the classic Quake 3. However, looking at the results we get in this review, it feels like Quake 3 perhaps is slightly outdated. The system with the worst performance cranks out more than 370 fps at 1024×768, with all graphic settings at maximum (except for AF and AA), which is almost laughable.
Not surprisingly, Athlon 64 FX55 takes the first place and is very close to passing the 500 mark, perhaps something that could be remedied with some overclocking later on?
This test, once the great attraction point of Intel, is now, like many other game benchmarks, a pure play field for Athlon 64. Only the overclocked 560 CPU makes it past any of the Athlon 64 processors, and even that is very close. Of course, Intel’s Extreme Edition CPU (2MB L3-cache) would surely have done well in this test, but we could not get a hold of this beast for our test, and as the 3.4GHz model of this processor makes the Athlon 64 FX55 look cheap, it doesn’t seem to be of interest either.
Also, we must not forget the fight between 4000+ and 3800+ where we once again see a difference of about 5% in performance.
Now the time has come for a few newer game titles, so jump to the next page for more tests.
The last tests in this review all use DirectX 9.0 in a certain manner and are
therefore a little more demanding on the graphics cards than the earlier tests
we’ve seen.
But equipped with Sapphire’s X800 XT card, we should be able to (sprida fältet?)
some in these tests aswell.
Halo – DirectX 9.0a
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Halo hardly shows any larger graphics card bottleneck
and we begin this page where we ended the other, with total Athlon 64 dominance.
The Athlon 64 processors are relatively close to eachother but they have really
got an advantage over the Pentium 4 processors. When even the Athlon 3500+ whips
the Pentium 560 with 17% it’s hardly worthwhile to comment the performance with
the FX55 and the 4000+ in comparison to Intel’s contribution.
So, the Athlon 64 processors are relatively close to eachother in this test
and this time there’s a ~3% performance differential between the 4000+ and the
3800+, a measureable difference but hardly noticeable in the game.
FarCry – DirectX 9.0b
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Most of the PC enthusiasts have tested and probably fallen for FarCry. The
game’s graphics is one of the reasons of this and it’s a title that demands
a good system.
Undeniably, Intel doesn’t seem to deliver this system where they once again
get a lesson from the Athlon 64 processors. Internally, AMD’s processors are
very close to eachother but then there’s a long way down to the nearest Intel
contribution.
Even if you get a good gaming experience with the Intel systems we’ve tested
here, you’ll have a lot more fun when we turn to Athlon 64.
The performance differential between the 4000+ and the 3800+ is almost non-existent
and it seems more like the graphics card is bottlenecking Athlon 64’s advance.
Doom3 – DirectX 9.0c
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
One of the newest and hottest gaming titles on the market is Doom3 and there’s
a lot of focus laid on the graphics card but as our results show, the system
processor plays a major role aswell. Once again we have the exact same placements
for the different processors with Intel on the behind. Athlon 64 FX55 takes
the lead in front of FX53/4000+ and then we have even paces downwards. Once
again we see little performance differentials between 4000+ and 3800+ where
the former gives about 3% higher performance thanks to its larger L2 cache.
Now the performance tests have ended, if we disregard from the ”bonus
tests” later in the review. It’s time to sum up the impressions from our
performance tests.
With today’s launch of the Athlon 64 FX55 and the Athlon 4000+ AMD takes another
step onto the performance throne and actually looks like the unthreatened current
performance king . Intel has gotten stuck on the second to last step when it’s
been recently announced that Pentium 4 Prescott won’t be launched at 4.0 Ghz.
AMD wins this review in 20 of 26 tests and in the majority of the tests not
only the flagships FX55 and 4000+ are in front of Intel’s flagships but also
the Athlon 3800+ and the 3500+ are seen before Intel’s Prescott processors in
the majority of the tests. We couldn’t get a hold of Intel’s monster processor,
3.4 Ghz Extreme Edition but we gave Intel an honest chance by overclocking our
Pentium 4 560 processor to 3.8 Ghz. Intel doesn’t have a Pentium 4 570 (3.8
Ghz) processor on their website but since the model is in retail stores it felt
worthy of at least trying to simulate this model, which we did by overclocking
it. Despite the overclocking giving an extra push in performance, higher bus
speed for CPU and memory, it was impossible to budge AMD in most of the tests.
Memory system:
The first sign of AMD’s good performance came already in the tests of the memory
system where AMD showed a very good memory bandwidth and an impressive memory
access (very low latency). The integrated memory controller and the transition
to double memory channels proved to be a winning combination and Intel’s DDR2
memory combined with an in coherence low clocked processor bus proved to be
the opposite.
Here AMD stood as the victor in 4 out of 6 tests with the Athlon 64 FX55 at
the top. The test that the Pentium 4 Prescott won was the unbuffered test in
Sisoft Sandra 2004 which we actually expected AMD to win. Why this happened
we cannot explain.
Athlon 64 vs. Pentium 4 Prescott: 4 – 2
File compression:
We started our non-synthetic tests with a simple file compression test where
the Athlon 64 processors used their memory system performance in full and took
a comfortable victory before Pentium 4 Prescott. All Athlon 64 processors were
noticeably in front of Intel’s contributions in this test.
Athlon 64 vs. Pentium 4 Prescott: 1 – 0
Multimedia, Audio/Video encoding:
Pentium 4 Prescott is a proven capable processor in multimedia management and
in spite of a slow start with MP3 encoding, Intel won 4 out of 5 tests by Pentium
4 560 and its overclocked ditto. Athlon 64 was close to Pentium 4 Prescott with
FX55 at the front. By testing video encoding with HyperThreading disabled we
could see that Intel’s simulation of two processor cores give a good merit at
this occation. Here we can see AMD’s Achilles’ heel that they hope to protect
by launching processors with double cores.
But until then, Intel is the way to go in multimedia applications.
Athlon 64 vs. Pentium 4 Prescott: 1 – 4
3D tests, synthetic and gaming titles:
In 10 3D tests (14 results and diagrams) Intel’s Pentium 4 Prescott
didn’t win a single one. In 14 out of 14 tests AMD came out as victor and that’s
quite impressive but if you look closer on how the victories came to be it gets
even more impressive. In the majority of the tests it was more or less of a
display by Athlon 64 and in 10 of the tests, ALL of the Athlon 64 processors
were in front of Intel’s Pentium 4 processors. Athlon 64 3500+ was in other
words in front of Intel’s Prescott processor at 3.8 Ghz in 10 tests.
It’s easy to sum up. AMD Athlon 64 is the undisputable master of games and 3D
performance and with the Athlon 64 FX55, new heights of performance are reached.
Athlon 64 vs. Pentium 4 Prescott: 14 – 0
AMD is clearly the winner in performance and even if Intel’s Pentium 4 Extreme
Edition maybe would have caused some more trouble it was impossible for us to
get a hold on one of these. It’s just as impossible as buying it right now.
AMD has the performance throne and we have it in clear numbers here below.
Final result: Athlon 64 vs. P4 Prescott: 20 – 6
When we’ve completed the performance tests it’s time to look closer on how
the Athlon 64 architecture was developed concerning heat dissipation and overclocking
potential. But we will also spend some time in underclocking, believe it or
not, but more about that later on. Now it’s about enormously fast and extremely
hot processors.
Athlon 64 FX55 and 4000+ give least said respectable performance in their original
models but as you say, he who has much would like more.
Since we had two of the fastest processors on the market in our possession it
didn’t seem as a bad idea at all to push the limits even further. If not to
find the 130nm technology’s limits then at least to squeeze out some more performance
out of these poor processors.
The overclocking tests were nothing in particular. We used the K8T800 Pro-based
motherboard we received with the processors and the standard cooler that was
also delivered in the same package.
 |
 |
The only "modification" we made was to
lean a 80mm fan toward the CPU cooler to increase the performance of the cooling.
The system was tested in an open case and the ambient temperature was at 22
degrees celcius during the tests.
First out we had the Athlon 64 4000+ that in its original model has a clock
frequency of 2.4GHz and by launching a new Athlon 64 processor at 2.4GHz (the
same clock frequency as 3800+), AMD has already given us a hint on where the
limit for the 130nm technology is. At least at acceptable production volumes.
In other words, we had no greater expectations for our overclocking tests other
than reaching at least 2.6GHz, the original frequency for Athlon 64 FX55.
As the Athlon 64 4000+ doesn’t allow changing the multiplication factor upwards
(downwards is OK and we’ll look further at this later on) we have to rely ourselves
on overclocking the processor bus. Lucky as we were, our testing motherboard
managed high bus speed without any problems, at least at such high speed that
the processor became the bottleneck.
After some tinkering we succeeded in finding a stabile clock frequency at no
less than 2.8GHz, or to be more exact, 2808MHz. This was achieved with a noticeable
increase in voltage, as we at this occasion were feeding the processor with
1.75v, a lot more than 1.5v which is the original voltage.
We even managed to boot Windows at 2.86GHz but it was almost enough for the
system to crash instantly. We managed to take a screenshot though.
The final stabile clock frequency was 2.8GHz and you can see proof of this below,
something that is also a little preview of the extra performance tests that
we are going to offer.
An overclock of 16% isn’t that bad
if we look at the circumstances and this is probably the last flagship model
in AMD’s Athlon 64 program that is based on 130nm technology.
But before we tie the bag up and move on to the bonus tests, we will look
closer on overclocking the Athlon 64 FX55.
With the overclock tests of the Athlon 64 FX55, we had the same setup
as with the 4000+, the only thing switched was the processor. But with the switch
of processors we gained new overclock possibilities since the Athlon 64 FX55
is completely unlocked. We could now increase the multiplier instead of messing
around with different bus speeds which produced problems for the RAM as we’ll
soon see.
 |
We didn’t use a 25x, but a 14x multiplier (original
13x). This setup let us reach 2.8GHz, and with a little tweaking of the processor
bus we managed to obtain a stable 2.82GHz. That’s one step more than the 4000+,
in other words. This gives us a good indication of the 130nm core’s limits.
We again fed the processor 1.75 V. The FX55 has the same 1.50 V original
voltage as the 4000+.
Naturally, we also proceeded to push the FX55 to
the limit in a screen dump. And even though we were close we didn’t quite reach
2.9 GHz.
Just as for the Athlon 64 4000+ we
have proof of our stable clock frequency below.
 |
During our tests of the processors
we also took the liberty of taking some simple measurements using the
internal temperature diode. Unfortunately, MSI’s temperature surveillance felt
a bit unstable, so we’ll take the readings with a grain of salt.
For these tests we removed the extra
80mm fan and used SiSoft Sandra’s CPU-burn to put a load on the processors for
30 minutes. We didn’t take any measurements without load since the values jumped
back and forth during idle, but the full-load temperatures are the important
ones. The room temperature was still 22 degrees celsius and here are the results.
|
|||||||||||||||||||||||||||||||||||||||||||||||
We’re not exactly dealing with cool
processors here, considering that the Athlon 64 4000+ and FX55 require a maximum
89 and 104 (!) watts respectively, there’s a tremendous amount of energy in
the form of heat that needs to be removed. The only competitor with even higher
numbers is Intel who, with their Prescott core, can produce up to 115W at 3.6GHz,
Pentium 4 560. We weren’t able to find information regarding the 3.8GHz model.
In other words, AMD’s flag ship requires some serious cooling, and when we increase
the core voltage for the overclocking we also require high quality cooling.
Now it’s time to see if our overclock
efforts let us gain anything performance-wise.
And now it’s time for some more performance tests, and this time we’ll let the
numbers speak for themselves, comments aren’t really necessary. But before we
give you the charts we’ll give the systems’ specs since the overclocking has
changed a few things.
|
The Athlon 64 4000+ has a higher bus speed on the
processor and memory, but then we had to use less aggressive memory timings
to keep the system stable. Something that could prove to make a difference in
the performance tests.
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Basically, we have really good performance! Now
we’ll move on to something that doesn’t have to do with performance.
Before we leave AMD and its Athlon 64 processors this time around we’ll take
a closer look at a feature we haven’t had time for earlier, Cool’n’Quiet.
It is basically a technique descending from the mobile market where low power
and low heat are vital ingredients for a successful processor. The mobile market has
seen several solutions both from AMD and Intel, but AMD is the forerunner in
bringing it to the desk-top market first.
The main concept is to balance the power consumtion
– thus also heat dissipation – with the load on the system. There’s no sense
inmaking the processor work at full speed when you’re only using 20% of its
capacity. This is the foundation for Cool’n’Quiet which simply lets the system
have full control of the processor clock frequency during usage, enabling a
change in both frequency and voltage on the fly.
What you need to make this work is an Athlon 64
processor that supports Cool’n’Quiet, a mainboard with the same support, and
a special driver for Windows, AMD
CPU Driver.
You can find more information on which processors and mainboards support the
feature here.
In our case the feature was supported through a
BIOS update of the board, our processors supported Cool’n’Quiet. However,
we only tested it on the Athlon 64 4000+.
To test the feature we put different loads on the
processor by using Sisoft Sandra’s burn-in wizard and obtained the following
results.
| Athlon 64 4000+ with Cool’n’Quiet |
||
|---|---|---|
| Processor load | Clock frequency | Voltage |
| ~65-100% | 2.4GHz | 1.50v |
| ~35-65% | 1.8GHz | 1.22v |
| ~0-35% | 1.0GHz | 1.12v |
In other words, a drastic decrease in both clock
frequency and voltage. This greatly reduces the power consumption and thus contributes
to a lower processor temperature. For those who don’t constantly utilize the
processor 100%, this is an excellent way to lower the temperature during regular
usage. If Cool’n’Quiet is used in conjunction with thermo-controlled fans, one
can produce a really quiet system, one superb area for this is the HTPC-system
where noise levels are critical.
We also tested Cool’n’Quiet’s possible influence on the performance by running
3Dmark2001 and there was no difference whatsoever compared to running without Cool’n’Quiet, so no negative influence on the performance.
Below are three screen dumps on the different settings
for Cool’n’Quiet on the processor.
 |
 |
 |
We’ve reached the end of the testing and wrap things
up with a conclusion.
And so finally time to wrap up everything
we’ve concluded in this review, and as usual there are a lot of impressions
and experiences to convey. Starting from the beginning we look at the development
of AMD’s Athlon 64 series.
Athlon 64 och Athlon 64 FX are, with the launch of the 4000+ and FX-55, architechtually
identical. The latter is of course completely unlocked and ships at a higher
frequency, but our overclocking tests show that both seem capable of the same
frequencies in the long run. For those who aren’t planning on doing any overclocking
there may not be any difference, but who’d buy an FX-55 and not plan to increase the
frequency? There might be a few out there, but the FX-55 isn’t called en enthusiast
processor for nothing.
The question is if AMD will continue with both processor series. If they do,
we think we’ll see a return of the 512kB L2-cache for the Athlon 64 series.
The reason for launching the 4000+ with 1MB was probably due to the low yield
of 2.6GHz cores, the frequency the 4000+ would have required if given a mere
512kB L2-cache like its predecessor.
Those cores that do manage to run at 2.6GHZ are probably just enough to produce
the low-volume FX-55’s, it would have been difficult to produce enough for both
the FX and A64 series.
This way AMD has managed to buy some time and can start introducing the new
90nm wafers into the lower frequency processors. Later on, even the high-end
models will be equipped with the new efficient core. So there’s a chance that
we’ll see a 4000+ running at 2.6GHz with 512kB L2-cache, the future will tell.
Moving on with the performance, we already summed
it up and the results were very clear. Athlon 64 is by far the superior competitor,
and the Athlon 64 FX-55 and 4000+ only pull farther away from Intel’s Pentium
4 Prescott. We can safely say that the Athlon 64 FX-55 is the market’s fastest
processor, and the Athlon 64 4000+ isn’t far behind. The only bottleneck for
the Athlon 64 series is the lack of quality multitasking. Intel’s HyperThreading
works miracles when used correctly, but that’s not quite enough to make up for
the Prescot architechture’s obvious shortcomings.
AMD’s introducing multi-core processor should be an optimal solution to this
problem, but then Intel will also be doing the same, giving them more processors
by again using their excellent HyperThreading. Only time will tell how this turns
out.
What we haven’t cleared up yet though, is the difference
between the Athlon 64 4000+ and 3800+, both running at the same frequency, but
having different size L2-caches. AMD has doubled the L2-cache and given a 5%
higher model number. In our tests we’ve seen anything between 0 and 7.5% increase
in performance for the 4000+. The scores average 3.5-4% which we find acceptable.
But this performance increase has to be related to the price. The Athlon 64
4000+ and FX-55 are very impressive performance-wise, but being on top comes
with a high price. You will find a price list below with all the processors
that were part of the review. We even threw in Intel’s Pentium 4 Extreme Edition
at 3.4GHz.
| Processor | MSRP in Swedish stores:
|
|---|---|
| AMD Athlon 64 FX-55: | 8000 SEK |
| AMD Athlon 64 FX-53: | 7800 SEK |
| AMD Athlon 64 4000+: | 7100 SEK |
| AMD Athlon 64 3800+: | 5700 SEK |
| AMD Athlon 64 3500+: | 2850 SEK |
| Intel Pentium 4 EE 3.4 GHz (S478) | 9800 SEK |
| Intel Pentium 4 E 570 (3.8GHz) | 6400 SEK |
| Intel Pentium 4 E 560 (3.6GHz) | 3900 SEK |
| Intel Pentium 4 E 540 (3.2GHz) | 2100 SEK |
The majority of these processors are without a doubt untouchable
for most buyers. The two new processors from AMD, the Athlon 64 4000+ and FX-55,
are the most questionable when considering price and performance.
Only Intel’s P4 Extreme Edition helps put a more humane perspective on
the prices.
Sure, the Athlon 64 FX-55 is the best money will buy, so if you happen to have
your own sugar mama then why not? You’ll be the head honcho in the neighborhood
for sure. For this segment, the Athlon 64 FX-55 definitely has a slot to fill,
but the 4000+ will find it difficult to fit in. It’s only the second
fastest processor on the market and still incredibly expensive. It will give
an average 3-4% performance increase compared to its predecessor, a predecessor
with a 25% lower price tag. The Price/Perfromance equation
simply doesn’t compute for the Athlon 64 4000+.
We can’t really recommend the Athlon 64 4000+ with a clean conscience, and
for AMD it becomes more of a status symbol when considering that
Intel doesn’t have a 4.0GHz equivalent of the P4 Presott.
AMD Athlon 64 is an excellent processor series with
extraordinary performance and great technological innovations (64-bit support,
Cool’n’Quiet). Now that AMD has strengthened its place as the performance leader
on the market, we can only sit back and wait for the market to fill up with
the 90nm Athlon 64 processors..
We would like to thank ABIT,
and Spire for contributing with products
for this review.
 |
 |
 |
|
 |
||
😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😛 😥 😥 😥 😥 😥 😥 😥 😥 😥 😥 😥 😥 😥 😥