As promised in the first article of the series dedicated to dual core processors, today we shall review the second CPU from this new series: AMD Athlon 64 X2 4800+.

AMD Athlon 64 X2 Inside

Like the dual core processor from Intel, Athlon 64 X2 actually comprises two almost equal processors, integrated into a single die. But if the Pentium XE / Pentium D package theoretically must turn out a tad larger than the doubled original dimensions (due to the integrated arbiter for the dual core bus), in case of Athlon 64 X2 it must be a tad smaller than the total of two processors, because memory and Hyper Transport bus controllers are not replicated in the dual core processor from AMD, both cores share them. Thus, there appears a counterpart of the bus arbiter (Hello, Intel!) that organizes interactions between the cores and the shared buses. However, this arbiter (aka Crossbar Switch in AMD terms) can sort of organize interaction between the cores as well (to all appearances, it means that one core can fetch necessary data from L2 Cache of the other core, saving on long queries to main RAM and lowering the memory controller load, as well as ensuring cache coherence). That's how it looks like:

Dual Core AMD Processor Flowchart



...And here is the die

You can easily notice that the differences between the dual core concepts from Intel and AMD are not very significant: this way or another, we have two processing cores, each one is equipped with its own L1 and L2 Caches. We could have spoken of different approaches, if one of the processors had used common L2 Cache for two cores, or, for example, if it had used separate L2 Caches supplemented with a common L3 Cache. But as it is now... on the whole everything is the same. Even the direct interaction engine between two cores (without involving a memory controller) has been copied by AMD from the ancient Athlon MP (two such processors in an SMP system could communicate directly as well). So, each company just took advantage of all its previous designs, which could be applied in this case. Frankly speaking, AMD had a tad more designs in stock...

Another bonus of the new AMD processor is that, unlike Pentium XE and Pentium D, it can operate (according to AMD; we have no reasons to distrust it) not only on specially designed motherboards, but on any motherboard for Socket 939, which can accommodate Athlon 64 FX, for example. But BIOS will have to be updated anyway, because if it doesn't support Athlon 64 X2, it will not provide the long-awaited dual core functionality: the operating system will detect only one core instead of both. I wouldn't like to croak disaster, but some motherboard manufacturers may be allured into selling the same models for the second time, simply by refusing to release updated BIOS versions for their old products and launching "special modifications" for dual core processors instead. However, let's hope that such manufacturers will be in minority...

Comparative technical characteristics

We shall just publish a table with the most important characteristics of dual core and top single core processors from both companies. In our opinion, comment is superfluous in this section: if you are interested in technical details, you should be able to interpret them on your own. We can note that dual core processors from Intel and AMD have two and three important and unobvious changes correspondingly (compared to top single core processors): Pentium XE 840 is now equipped with the EIST technology that reduces power consumption, but its bus rolled back — from 1066 MHz to 800 MHz, Athlon 64 X2 4800+ acquired SSE3 support and improved memory controller. Besides, it's manufactured by the 90nm process technology. We shouldn't bother mentioning the increase in die dimensions, number of transistors, and power consumption: that's obvious.

Testing

Testbed configurations

• Processors
  • Intel Pentium eXtreme Edition 840 (2 x 1 MB L2, 800 MHz FSB, 2 x 3.2 GHz core)
  • Intel Pentium 4 540J (1 MB L2, 800 MHz FSB, 3.2 GHz)
  • AMD Athlon 64 FX-55 (1 MB L2, 2.6 GHz)
  • AMD Athlon 64 4000+ (1 MB L2, 2.4 GHz)
  • AMD Athlon 64 X2 4800+ (2 x 1 MB L2, 2 x 2.4 GHz core)

• Motherboards
  • Intel D955XBK Desktop Board (Intel 955X)
  • Albatron K8SLI (NVIDIA nForce4 SLI)
  • ASUS A8N-SLI Deluxe (NVIDIA nForce4 SLI)

• Memory
  • 2x512 MB PC3200 (DDR400) DDR SDRAM DIMM 2-2-2-5 (Corsair TwinX)
  • 2x512 MB PC5400 (DDR2-533) DDR2 SDRAM DIMM 3-3-3-8 (Corsair)

• Video card: ATI Radeon X800 (256 MB)
• HDD: Samsung SP1614C (SATA), 7200 rpm, 8 MB Cache
• AC power adapter: FSP 550-60PLN (500-550W)
• Windows XP Professional SP2, DirectX 9.0c
• ATI CATALYST 5.4 (Display Driver 6.14.10.6525)

Did you notice that we used the most progressive and (presumably) the fastest chipsets, which is quite appropriate for the status of processors under review? They are all top solutions, so the slogan "A bas the economy!" looks quite justified in this case.

A separate note should be made of intelligent rpm control on testbeds with dual core processors: first of all it really works — when in idle mode, fans dropped their speed rather fast and rotated almost noiselessly. But the second argument for the Pentium XE testbed looks less optimistic: its stock boxed cooler was much noisier at maximum rpm than the Athlon 64 X2 system. Besides, we have a subjective impression that it took less time for the fan in the Intel system to arouse to increased rpms (from the start of the test).

Test results

Diagrams with all test results (65 all in all) are published on a separate web page — without comments, just as is. The article provides only summary diagrams that calculate the results of entire test groups into average scores. This approach appeases curiosity of the most inquisitive readers, who are against cutting down the number of test results published in our articles, and still makes the article less motley and graphics-intense. What concerns our comments, real professionals (who are interested in details) are expected to need none of them.

SPECapc for 3ds max 6 + 3ds max 7.0

A complete set of diagrams

In this case Athlon 64 X2 4800+ outperforms the dual core Pentium XE in all tests (you may refer to the complete set of diagrams with test results, if you wish). It could have been predicted by the results of comparing Pentium XE 840 with Athlon 64 FX-55: even FX-55 was superior to XE 840 in total score due to its higher points in Interactive test. And now the top AMD processor improved its results in the rendering test due to its dual cores (it's well known that render engines are very good at multithreading; in some cases the second processor may even yield 100% of gain).

The same results of the single core Athlon 64 4000+ and Pentium XE 840 are also indicative. It's crystal clear that the identical results are conditioned by absolutely different merits: in case of AMD it's due to the high result in Interactive test, in case of Intel it's up to Rendering test. But it's still very illustrative: "the younger brother" of Athlon 64 X2 4800+ (single core Athlon 64 4000+, its only core runs at the same clock as in X2 4800+) goes on a par with the fastest dual core processors from Intel in the total score...

Maya 6.5

We have finally repaired the main defect of SPECapc for Maya and introduced our own rendering test. So from now on the Maya section will feature two diagrams: SPEC total score and rendering speed. That's how the rendered scene looks like:

A complete set of diagrams

The alignment of forces is rather complicated: single core processors turn out the fastest in SPEC tests, top extreme processors from both manufacturers being almost on a par. The dual core processor from AMD slightly outperforms the dual core processor from Intel, but that's because the AMD core running at 2.4 GHz is faster than the Intel core operating at 3.2 GHz rather than because of the better implementation of dual cores. The same situation in rendering: extreme modifications of single core processors are on a par, the dual core processor from AMD is just a tad faster than the dual core processor from Intel. On the whole, we can say that AMD wins on a decision, but its breakaway is minimal.

Lightwave 8.2, rendering

In general, the situation resembles Maya, but it's more complicated: Intel is victorious in single core contest (Pentium 4 eXtreme Edition 3.73 GHz), but the dual core CPU group is dominated by AMD Athlon 64 X2 4800+. However, in both cases the advantage of the winners is not that large.

SPECapc for SolidWorks 2003

A complete set of diagrams

That's the demesne of AMD, its processors take all the three top places. But whether SolidWorks 2003 (or the SPEC test for this program, to be more exact) can use the second core is dubious. There is practically no difference between Pentium 4 540J and Pentium XE 840. However, in case of Athlon 64 4000+ and Athlon 64 X2 4800+ the difference is about 4%, which can hardly be written off to the measurement error! Will SolidWorks take advantage of the second core or no? We don't think so. The evidence is the almost complete lack of differences between the test results of single and dual core Intel processors. And the victory of AMD dual core can be explained by an improved memory controller and (provided this set is utilized) SSE3 support. Considering all the above said, that will probably be the most likely explanation.

Adobe Photoshop CS (8)

A complete set of diagrams

Intel and AMD processors of the same rank usually take the first and the second places correspondingly in our test for Adobe Photoshop 8, that is a small advantage is enjoyed by Intel NetBurst. However, in case of dual core CPUs, this advantage is purely nominal (1 second!), and can be written off to the measurement error (that is if we run this test another time, the dual core processor from AMD may turn out faster by a couple of seconds). Thus, we can see approximate parity. If you look into detailed diagrams, you will see that the dual core processor from Intel wins more or less significantly in such operations as color conversion and Unsharp Mask, while Athlon 64 X2 4800+ turns out faster in lighting effects and Resize. And Pentium 4 XE 3.73 GHz is still faster than Athlon 64 FX-55.

Adobe Acrobat 6.0

Strange as it may seem, Adobe Acrobat 6 Distiller is evidently optimized for SMP architectures, because dual core processors (both from Intel and AMD) demonstrate a significant advantage over single core ones with the same clocks. The total advantage in this test belongs to AMD processors.

All-purpose data compression (archiving)

A complete set of diagrams

In all previous tests Intel processors got a noticeable compensation for their traditional defeat in WinRAR test thanks to the results demonstrated in the other archiver — 7-Zip. The latter can use both processors, so Intel CPUs with Hyper-Threading support turn out better as a rule. However in this case one of the test participants is an AMD processor with two sterling cores and all Intel processors were outperformed in 7-Zip. As a result, the advantage of AMD group on the summary diagrams gets even more prominent.

Multimedia lossy compression (MP3/MPEG2-4)

A complete set of diagrams

In this section we recommend referring to the detailed diagrams, because the defeat (even if very insignificant) of the dual core processor from AMD to the Intel representative in the same group (Pentium XE 840) can be explained by a really significant lag of Athlon 64 X2 4800+, but... only in two tests only in two test! ! Here is the situation: Athlon 64 X2 4800+ was outperformed by Pentium XE 840 in MP3 encoding with average quality, it was heavily outscored in MP3 encoding with maximum quality (Q=0), and... it was victorious in all other tests! That is the dual core processor from AMD encodes video (MPEG2/MPEG4) faster, despite its nominally worse result on the summary diagram.

CPU RightMark 2004B

A complete set of diagrams

Yep, today's tests can be truly called (though it may seem too emotional) the downfall of the last Intel's bastions. And now CPU RightMark... AMD processors were traditionally victorious in the Solver module and defeated in rendering because it's multithreaded. But the dual core Athlon 64 X2 changed the situation radically: now the top AMD processor wins in the render module as well and thus it demonstrates the highest total score!

Let's dwell on the effect of "super performance gain" (over 100%!) in dual core processors (both from Intel and AMD). As a matter of fact, this seemingly impossible thing is quite possible and even easy to explain. The fact is that in case of a single CPU, it spends some of the processing time on solving the physical task (Solver) and some time on rendering (Render). Besides, you should take into account that Solver performance is dozen times higher than the Render module performance. It turns out that the slowest process (Render) is sometimes interrupted for some processing time slots to calculate the next frame in Solver.

What happens when the second core is added? Solver module operates with a single thread only, so nothing happens here. Render module is multi-threaded, so all resources of the second core are claimed by the rendering module! And considering that it's the major bottleneck in CPU RightMark, it becomes clear why the second core increases the general performance more than twofold: the second processor (the second core) devotes all its processing time to rendering, while a single (the first) CPU also spends its time on switching between Solver and Render, and on the Solver module itself.

3D games and graphics visualization
in professional packages

DOOM 3

Far Cry

Painkiller

Unreal Tournament 2004

Total score in games

A complete set of diagrams

Nothing interesting can be seen in this group of tests: games still cannot use the second core, so single-core processors are almost on a par with dual core ones. The dual core AMD processor is a tad faster (compared to the single core CPU with the same clock), but it can be explained by an improved memory controller (and probably the introduction of SSE3 support). And the general advantage of AMD processors in games is nothing new...

SPEC viewperf

A complete set of diagrams

As before, we cannot but establish a fact that 3D graphics visualization in professional packages is no different from the preferences of 3D graphics visualization in games: AMD takes the lead, there is almost no effect from the second core. However, in this case the new memory controller played into the hands of AMD Athlon 64 X2 4800+.

The drop of the main process performance
with several background processes

It's an improved version of the new test, which was used in our previous article. Today we have decided to apply maximum load to all processors, available on testbeds, their total number reaching four (together with virtual ones): the system detects the dual core Pentium XE 840 with Hyper-Threading support as four CPUs. So, in order to overload the 4-CPU system, we had to employ five active processes (by one process more than the number of CPUs). That's why we carried out our tests in the following way:

1. At first we started Far Cry (Volcano) benchmark in high quality mode (1024x768x32, corresponding graphics settings) and measured performance, when nothing interfered with the main process. This result was assumed as a basis (the first point on the graph).
2. Then we started MP3 encoding in LAME in minimized window (it's important the OS understands that it is a background process). Then we started Far Cry (Volcano) benchmark again and measured performance (in frames per second). Thus we could see how much the active background process (that spends CPU resources) interferes with the main process.
3. Then we made the task more difficult: Far Cry benchmark was started on the background of MP3 encoding in LAME together with archiving in 7-Zip (they are all our standard tests, but in this case they ran in the background).
4. Another background process was added to MP3 encoding in LAME and archiving in 7-Zip: converting PostScript files into PDF format in Adobe Distiller.
5. The number of background processes was increased to four by adding MPEG4 (XviD) encoding in VirtualDub.

Besides, in order not to overcrowd the diagram (and not to degrade its readability), we decided to discard less interesting processors that have nothing to do with the object of our tests - Athlon 64 FX-55 and Pentium 4 XE 3.73 GHz. Thus, we came up with dual core processors and single core CPUs operating on the same clock as the dual core ones.

It goes without saying that this test has nothing to do with real operations. We cannot imagine a user (compos mentis), who would archive files, encode MP3 and MPEG4, and distill PostScript at the same time. But on the other hand, we didn't undertake the task of simulating real user operations. The task was purely synthetic: to see how the performance of the main process would drop when an increasing number of background processes were started on various processors. Our contenders included a variety of real/virtual multiprocessing:

1. Sterling single-core Athlon 64 4000+.
2. Virtual dual core Pentium 4 540J. However, we could transform it into a sterling single core processor by disabling Hyper-Threading.
3. Sterling dual core Athlon 64 X2 4800+.
4. Virtual quarter core Pentium XE 840. It could be easily transformed into a sterling dual core by disabling Hyper-Threading.

And now let's have a look at the diagrams. Note that the Y axis starts from the lowest fps value in a given test instead of zero, in order to zoom in the diagram.

Remember that we measured performance drop of the main process depending on the number of active background processes. That is we are not interested in the performance of background processes, so we don't measure it. They just provide background load on the CPU. So, what do we see?

1. Sterling single core processors (AMD Athlon 64 4000+, deep green line; Intel Pentium 4 540J with disabled Hyper-Threading, deep brown line) — they just slump. In fact, performance of the main process instantly drops more than twofold, thus rendering the main process (Far Cry) impossible to use: playing THIS (judging from the demo) would have been almost impossible, because the game runs in jerks to say nothing of real slow, only Terminator with his nerves of steel could have stand it.
2. Pentium 4 540J with enabled Hyper-Threading (red line) slumps from a single background process much less noticeably. Besides, we didn't notice any jerks when the benchmark was running: its speed just slowed down a little.
3. But a still funnier reaction is demonstrated by Pentium 4 + HT to two or three background processes: its performance was faster in case of three background processes than with two of them. We could come up with only one explanation, which we called the effect of bouncing main process. Here is what we think: each of two background processes, started before the main process, occupies a single CPU. As a result, the third process (the main one!) finds itself in a very unpleasant situation: this very process has to bounce from CPU 0 to CPU 1 and back, because the dispatcher cannot assign it permanently to only one processor due to the odd total number of active processes.
4. Sterling dual core processors (AMD Athlon 64 X2 4800+, bright green line; Intel Pentium XE 840 with disabled Hyper-Threading, navy blue line) behave in the same manner: a single background process has almost no effect on the performance of the foreground process, the second background process added slumps its performance significantly, the third background process — even accelerates the main process (to all appearances, we again witness the deliverance from the effect of a bouncing main process).
5. The smoothest performance drop curve is demonstrated by the virtual quarter core processor (Intel Pentium XE 840 with enabled Hyper-Threading, blue line). Each time another background process is started, performance of the main process always drops, but not very much.

The main conclusion from the above said is as follows: Hyper-Threading technology makes sense, even though it cannot replace sterling multiple cores. Virtually doubling the number of processors mostly has an effect on how comfortable it is to work in an application rather than on its performance, given it runs on the background of several active processes. Dual cores without Hyper-Threading are less flexible in this respect: they are characterized by deeper and sharper slumps at each new background process started. These slumps are especially prominent when the even number of processes (including the foreground process!) grows into the odd number.

Conclusion

A surprise is what happens,
when people have been stubborn not to notice
something obvious for a long time.
(I read it in a forum)

So, certain parity has been demonstrated before the launch of dual core processors: from time to time Intel or AMD would release another monster, a couple percents faster than a top competing processor, that would enjoy (for a couple of months, or even less) the title of the fastest desktop x86 CPU. This situation was generally accepted. Power consumption (and consequently heat dissipation) of Intel processors had been growing a tad faster than in AMD processors, but it "worked" so far — frankly speaking, the majority of users were not at all concerned by this growth (especially as it had grown gradually). In the end, even the scariest power consumption of state-of-the-art CPUs will hardly surpass some chandelier with a lot of lamps in a large room. Many of us use such things and don't even think about it, it doesn't scare us away... Bottom line: the above mentioned tendency (Intel processors consume increasingly more power and dissipate more and more heat) was no secret.

The other tendency is not hushed either: since the first Prescott core CPUs, the number of transistors in Intel processors has started to noticeably surpass the number of transistors in AMD processors (just compare the humble 106 millions in the top Athlon 64 FX-55 and 125 millions in the relatively mainstream Pentium 4 5xx series). Theoretically, AMD processors based on K8 core should have had more transistors: they have the same cache size (1 MB), but Athlon 64 has an integrated dual channel memory controller and a Hyper Transport bus controller at that. C'est la vie...

And now I want to ask a provocative but hackneyed question: considering the above said, who should have designed a better dual core processor?

It remains for us to establish the fact: the one, who should have designed it, has been successful. What was an annoying but not that significant hindrance in case of single core Intel processors has become the main stumbling block for designing a dual core CPU. It's now popular to say that AMD "talked about multiple cores" long ago and to make a conclusion that its processors "were specifically designed" with regard for integrating two cores into a single CPU. Of course, that's a probability. But I didn't take part in the design process, so I have nothing to say. But personally, I don't see any specifically multi-core properties in the AMD K8 core. Except for one vital thing: if designing a processor for multiple cores means designing a CPU with relatively low power consumption and low (for desktops) heat dissipation — then, you are right! In this case "the initial orientation towards multiple cores" is obvious.

Summing it all up: AMD Athlon X2 4800+ has swept in Pentium XE 840 essentially because AMD was ready to design two cores on a single die. Alas, Intel turned out not ready for this concept de facto: the current mainstream core based on NetBurst architecture (another reincarnation of basic Prescott) consumes too much power and dissipates too much heat. That would do for a regular CPU (top Pentium 4 processors work all right!), but that's too much for a dual core processor with the same clock as top single core CPUs. That's actually the reason of Pentium eXtreme Edition 840 running at a rather low clock (compared to top Pentium 4 processors). It's the low clock of the Intel dual core processor that left it no chances to victory: if it had been 3.8 GHz instead of 3.2 GHz, everything could have been different. Speaking of ifs and ans, let's try and fantasize a little. So, what will the two rivals do tomorrow, considering the present day situation?

AMD

1. Having reached approximate parity (let's ignore fractions of a percent) with Intel in performance of single core desktop CPUs and having scored an impressive victory in the sector of dual core solutions, the company may have... a little rest and do nothing. Just resting on its oars. But this idle life is fraught with negative consequences, because Intel will surely not stand waiting, it will hardly lump the present situation.
2. It would probably be reasonable to focus on streamlining the manufacturing process and introducing new facilities. It's no secret that the productive capacity of AMD is much lower than in Intel. If our readers had an impulse to buy Athlon 64 X2, but failed to find it in stores, it would ruin the impression from the record breaking performance...
3. It would be logical to slowly expand the nomenclature of dual core processors downwards — to lower clocks and more democratic prices. After all, we can hardly expect the forthcoming launch of considerably higher clocked dual core processors from Intel. So the success on top should be consolidated in the middle end segment.
4. Strange as it may seem, just in case of dual core CPUs (which should theoretically provide not only high performance with a single running task, but also more comfortable performance with active background processes running) the idea of Hyper-Threading counterpart in AMD processors looks quite tempting. At least I must admit that the smooth curve of Pentium XE 840 was quite impressive to me, irregardless of its performance.
5. That's about it. We can hardly expect AMD dual core clocks to grow (I mean a significant kind of growth): at least one piece of evidence in favor of this assumption is that the top single core Athlon 64 FX-55 is just 200MHz higher clocked, and FX-57 is not to be seen so far...

Intel

1. The scenario "by contraries": to bury the market under relatively inexpensive dual core processors, even if with lower clocks and performance than offered by top Pentium XE 840 / Pentium D 840 processors. Unlike AMD, Intel can easily manipulate its productive capacities, so this option does not look far fetched at all. In this case, if dual core processors were in demand, there would be a paradoxical situation: Athlon 64 X2 would still remain the fastest processors, but regular users would be able to afford only Pentium D.
2. To start (continue?...) overhauling cardinally the Prescott core (or to design a special core for multi core CPUs) and use this new core in processors, which would be able to compete with Athlon 64 X2 (including the top 4800+ CPU) in terms of pure performance. You shouldn't be a Solomon to understand that this option will be carried out anyway, because Intel has no other way in the long terms. The question is how soon this core will be designed and streamlined.
3. Swingaround (it's a term from navy tactics textbooks: at the command of a flagship, the entire fleet simultaneously sets a different course). In our case it means abandoning NetBurst (at least in multi-core solutions) and switching dual core processors to a core with low power consumption and heat dissipation. There is certainly only one idea that springs to mind: Banias/Dothan, Pentium M. However, despite the optimism of many users about the Pentium M performance on MHz basis and its low power consumption, this move looks like a two-edged weapon in the long view: one can assume that having designed a dual core processor based on the top Pentium M, Intel will manage to catch up with Athlon 64 X2 4800+ in performance right now. That is the gap will be covered at the earliest possible date, everybody is happy and laughing... But what then? There is no evidence that Pentium M will be good at scaling: its clock "has frozen" so far. Having switched most of its production facilities to the new core, Intel would run into the same wall as before. In return it would get a dubious (in terms of expenses) "prize"... it would just catch up with the main competitor in the sector of multi-core solutions. Is the game worth the candle?
4. Strange as it may seem, another option sounds like the first scenario for AMD: just do nothing. At least in the nearest future. Indeed, the main sales currently fall on regular single core CPUs, this tendency will remain for rather long. And Intel sells them like hot cakes. So the company can simply forget about the temporary failure in just a single (and not the largest) sector of the x86 CPU market and leisurely design and streamline a new architecture (Scenario 2).

But irregardless of how the events will unfold in future, we can presently say one thing for sure: we haven't seen such a situation for a long time, when Intel has practically nothing to oppose to the AMD challenge in terms of the top x86 processor performance in the most promising vector of its development (both companies have agreed upon multiple cores being this vector). We haven't seen it since Pentium 4 Willamette. But let's hope that the company will come up with something like "the second Northwood": I think that even AMD zealots wouldn't like to see this company alone in the sector of high-performance multi-core x86 CPUs.

Write a comment below. No registration needed!