by Ilya Gavrichenkov
07/31/2005 | 10:39 PM
The users welcomed the dual-core desktop processors with great enthusiasm. New architectures that allow combining two processor cores within a single semi-conductor die boosted the performance of contemporary CPUs quite tangibly. Since it has become extremely difficult to continue increasing the clock frequencies of the CPUs nowadays, the processor developers are truly excited about the potential of the dual-core solutions.
However, dual-core CPUs, like any other new products out there, turned out pretty expensive and thus could hardly become really mass products. This is primarily the case for dual-core AMD Athlon 64 X2 CPUs, which have been initially positioned as higher-end solutions than the single-core Athlon 64. As a result, the price of a single Athlon 64 X2 rested somewhere between $500 and $1,000.
Intel proved to be more democratic in setting the prices on its new dual-core solutions. The price of Pentium D processors starts at around $241, which gives them green light on the way to mainstream systems. However, this tremendous price difference is absolutely justified: the dual-core processors from AMD that have been in the market so far are significantly faster than the dual-core Pentium D solutions.
No matter what, this was definitely not what AMD wanted. The fact that Intel’s dual-core CPUs were selling at a more affordable price was a pain for AMD marketing people. Therefore, following the announcement of the first dual-core processors, AMD started working real hard on the ways of improving their dual-core solutions pricing. And this task has finally been solved. Today, on August 1, 2005 (EDC) AMD announces the youngest model in the Athlon 64 X2 processor family with the 3800+ rating, which price is set at $354, according to the company’s official price-list. Another pleasant thing is that this is not a “paper” launch, it is absolutely real: AMD Athlon 64 X2 3800+ should start appearing in stores any minute now.
The price of the new Athlon 64 X2 has been reduced to $354 in a pretty standard way. Firstly, the working frequency of this processor is lower than the working frequency of all other dual-core AMD solutions. Secondly, this processor features smaller L2 cache. By reducing the L2 cache size AMD could make the die size smaller, which certainly affects the production cost of the CPU. The first Athlon 64 X2 processors were based on Toledo core consisting of 233.2 million transistors and featuring the die size of 199sq.mm. The new Athlon 64 X2 3800+ is based on the new core known under the codename of Manchester (this core has also been used in other younger processor models of the family), which is 147sq.mm big and features only 154 million transistors. This is certainly more than a single-core AMD processor would have, but it nevertheless allows increasing the production of dies per single 200mm wafer by the good 38%. By the way, due to smaller L2 cache, the new Athlon 64 X2 on Manchester core turned out almost as small as the Pentium 4 6XX processors from Intel.
So, the new Athlon 64 X2 3800+ appears a very interesting solution for our detailed study and testing. This dual-core processor from AMD falls into a slightly different price group than all of its predecessors, which may theoretically turn it into a true sales hit. Of course, this can only happen if its performance turns out adequate. So, in our today’s article we decided to take a closer look at the performance of the new comer and thus evaluate its future prospects for the processor market.
We have already discussed the peculiarities of dual-core AMD processors in our article called AMD Athlon 64 X2 4800+ Dual-Core Processor Review . The new Athlon 64 X2 3800+ differs from its faster counterparts by smaller L2 cache, which is 512KB for each of the processor cores (Athlon 64 X2 4600+ and Athlon 64 X2 4200+ feature the L2 cache of the same size), and by lower clock frequency, which is 2.0GHz. This way the entire AMD Athlon 64 X2 processor family now looks as follows:
L2 cache size
Athlon 64 X2 4800+
1 MB + 1 MB
Athlon 64 X2 4600+
512 KB + 512 KB
Athlon 64 X2 4400+
1 MB + 1 MB
Athlon 64 X2 4200+
512 KB + 512 KB
Athlon 64 X2 3800+
512 KB + 512 KB
The full specifications list for the new AMD Athlon 64 X2 3800+ is given in the table below:
Athlon 64 X2 3800+
939-pin organic micro-PGA
512 KB + 512 KB
Supported memory types
HyperTransport bus frequency
Typical heat dissipation
Max. case temperature
I would like to draw your attention to the fact that the Thermal Design Power for AMD Athlon 64 X2 3800+ is set to 89W. It means that the CPU is compatible with all those mainboards and cooling systems that suit for regular Athlon 64 processors. The remarkable thing here is that all the previous Athlon 64 X2 processor models except 4200+ featured typical heat dissipation of 110W.
Another interesting thing is that Athlon 64 3800+ has set processor Vcore of 1.35V. Of course, it means that they didn’t have to increase the Vcore to 1.4V for their youngest CPU model production.
The diagnostic CPU-Z utility reports the following about Athlon 64 X2 3800+:
No surprises here, as you may see. The utility recognizes the Manchester core working at 2GHz frequency.
The practical power consumption that we measured for the reviewed CPU reached the value of 65.1W. The workload was created by the special S&M 1.7.2 utility. Let’s compare this result with the power consumption of other processors:
As we see, the typical heat dissipation value set for Athlon 64 X2 3800+ is absolutely justified. Although it consumes more power than its single-core fellows from Athlon 64 family (on Venice core), it is still a way below Athlon 64 FX-57 with the Thermal Design Power of 104W. As for the competing solutions from Intel, any comparison here would be absolutely pointless, because all Intel CPUs consume about twice as much as their AMD rivals anyway.
Of course, I should say a few words about Cool’n’Quiet technology, which the dual-core processors inherited from their single-core buddies. Athlon 64 X2 3800+ does support this technology, and its only peculiarity is the synchronous frequency and voltage reduction of the two cores in case the CPU is not loaded heavily.
In case of lower power consumption, Athlon 64 X2 3800+ drops its working frequency to 1GHz and core voltage to 1.1V. As a result the power consumption drops down to 5.8W in idle mode, which makes Athlon 64 X2 3800+ a very economical solution. In fact, they could have achieved even higher power saving efficiency if the cores could turn to economy mode independently. However, it looks like this feature will only be implemented in the dual-core processors designed for the mobile segment.
We tested the performance of AMD Athlon 64 X2 3800+ and compared it against the results shown by the processors falling into the same price category. Among them are: Athlon 64 3800+ ($373), Pentium 4 650 ($401) and Pentium D 830 ($316).
So, we assembled a few systems with the following hardware components:
The peculiarity of our today’s test session is the use of two different operating systems: 32-bit and 64-bit Windows XP versions. When we tested the processors in 64-bit mode we tried to use mostly native 64-bit applications, which have already become quite numerous by now. This way we will be able to evaluate the processors performance and efficiency not only in 32-bit mode but also in case AMD64 and EMT64 technologies kick in.
To be fair we should definitely mention that many 64-bit applications available today are none other but Open Source program ported developed by computer enthusiasts. So, these programs boast very unique specifics. Unfortunately, there are very few really 64-bit versions of any large commercial products from big-name companies.
The new revision of the PCMark test doesn’t differ that greatly from the previous test versions. The CPU test from this benchmark set is based on real-life data encoding and compression algorithms and uses multi-threading a lot. Therefore, the result obtained is quite natural. Dual-core processors show better results in applications of this type than single-core processors, and CPUs with NetBurst architecture that are traditionally faster in PCMark again boast the best performance results in this test.
I would also like to point out that the results of PCMark05 test are the same in 32-bit operating system and in 64-bit OS where AMD64 and EMT64 technologies start working. This is another illustrative evidence of high efficiency of x86-64 architecture: the 32-bit applications performed in the compatibility mode in 64-bit operating system work at the same speed as in their authentic 32-bit environment.
The same is true for 3DMark05 results. The 64-bit Microsoft Windows XP Professional x64 Edition OS doesn’t cause any performance drop in 32-bit DirectX programs. So, the gamers shouldn’t fear the migration to 64-bit era supported by AMD processors with AMD64 technology and Intel processors with EMT64 technology.
The 3DMark05 benchmark itself, like many games out there doesn’t support multi-threading. That is why dual-core processors do not stand out here in any way. However, this benchmark set includes specific CPU tests, which use multi-threading for shader calculations and simultaneous modeling of the gaming environment.
The new Athlon 64 X2 3800+ processor performs quite up to its price. In the first gaming test it outpaces its single-core competitors being just slightly behind Pentium D 830 working at 3GHz core clock. However, in the second test it really dashes forward, so that all the rivals from the same price group fall far behind.
Contemporary games do not use multi-threading that is why dual-core processors cannot boast significant improvements in applications of this type. Athlon 64 X2 3800+ performs at about the same fps rate as a single-core Athlon 64 3200+:
However, due to the efficiency of K8 architecture in games, Athlon 64 X2 3800+ doesn’t yield that greatly to the single-core Pentium 4 processor in games. Moreover, we can once again point out that the shift to 64-bit mode hardly affects the processor performance in gaming applications.
Even though game developers do not please us with too many advantages of multi-core processor architectures, 64-bit extensions start little by little finding their way into this type of tasks. Not so long ago a new patch came out for the popular Far Cry game, so that now you can play this game from Microsoft Windows XP Professional x64 Edition in 64-bit mode. Of course, we couldn’t leave out this fact and tested the processors performance not only in the standard 32-bit version of the game, but also in the 64-bit version.
As we see, the fps rate is somewhat higher in the 64-bit Far Cry version. So, 64-bit operating system and 64-bit version of the game will guarantee you about 3-5% performance increase.
The popular WinRAR archiving utility doesn’t support multi-threading that is why the results shown by our today’s hero, Athlon 64 X2 3800+, are not that high at all. Namely, it yields to the single-core CPUs from the same price group. However, if we compare the results of our Athlon 64 X2 3800+ with those of the dual-core Pentium D 830 processor, the situation will look not so bad at all: both these processors perform almost equally fast.
You should also pay attention to the fact that when the 32-bit WinRAR utility works in 64-bit operation system, it works overall slower. The reason for that might be the WoW64 interpreter, which makes it possible for 32-bit software to work in Microsoft Windows XP Professional x64 Edition.
Some archiving utilities actually do support multi-threading. For example 7zip. 7zip not only works very efficiently with multi-core processors, but there is also a 64-bit version of it already available. That is why we thought it would be very interesting to use it in our today’s test session.
The data compression algorithm in 7zip uses Hyper-Threading technology pretty actively. Nevertheless, Pentium D 830 working at 3GHz frequency runs almost as fast as Pentium 4 650 working at 3.4GHz core clock frequency. The single-core Athlon 64 3800+ yields to Intel CPUs here. And as for the Athlon 64 X2 3800+, it fails to catch up with the competitors from Intel’s camp, Pentium 4 and Pentium D, even though its performance in this test got 22% higher than that of Athlon 64 3800+.
Everything we have just said is true only for the 32-bit version of 7zip archiving utility. The situation changes noticeably once we switch to 64-bit version. The thing is that 64-bit registers do a lot of good to Athlon 64 processors improving their performance significantly, while Pentium 4 processors cannot benefit that much from 64-bit registers. As we can see from our benchmark results, CPUs with NetBurst architecture may run slower in 64-bit mode than they would run in 32-bit mode. That is why the 64-bit version of 7zip utility makes Athlon 64 X2 3800+ processor the winner.
During data decompression, both - Athlon 64 and Pentium 4 work faster in 64-bit mode. However, in our particular case K8 processors appear overall faster. The leader here is a single-core Athlon 64 3800+, while the dual-core Athlon 64 X2 3800+ is the second fastest falling by about 18% behind the leader.
First of all we will take a look at the audio encoding into mp3 format with a popular LAME codec. For our tests we used an unofficial codec version 3.97, which supports multi-threading and exists in 64-bit version.
During audio encoding, dual-core processors prove faster than their single-core fellows even though they work at lower clock frequencies. During audio encoding with a 32-bit codec, the laurels will be won by the dual-core Intel Pentium D 830. in case of a 64-bit version the situation will change. Strange as it might seem, but the 64-bit version of LAME is slower than the 32-bit one. Moreover, if the Athlon 64 processors get only 10% slower, then the Pentium 4 CPUs lose about 20% of their speed. As a result, the best performance in 64-bit LAM version belongs to Athlon 64 X2 3800+.
This relatively strange situation with the 64-bit version of LAME is most likely connected with the Microsoft compiler, which was used for the final code processing. However, in this “critical” situation, when the 64-bit program version appears slower than the 32-bit one, you can easily use the faster version in the 64-bit OS, even though it will require activating the compatibility mode.
There also exists a 64-bit version of the XviD codec. With this codec we tested the performance of our processors during video encoding in 64-bit and 32-bit operating system.
Well, no surprises here, not like in the previous case. The 64-bit codec version is evidently faster than the 32-bit one, although you will not be able to benefit from dual-core architecture in any way here. So, in the price group in question the best result was demonstrated by Athlon 64 3800+ processor.
Now let’s take a look at our testing participants in those codecs that do not have 64-bit revisions yet.
Despite its dual-core architecture and SSE3 instructions support, Athlon 64 X2 3800+ cannot win this race. Pentium D 830 appears the fastest here. Note that in this codec the dual-core AMD processor runs somewhat slower than its single-core fellow from the same price group, while the situation in Intel’s camp is just the contrary: single-core Pentium 4 650 gets defeated by the dual-core Pentium D 830.
The results obtained with the DivX codec are quite predictable. NetBurst architecture works better in this case than K8. Besides that, higher core clock frequency of the single-core processors appears more efficient here than the dual-core architecture, even though the application does support multi-threading. So, AMD Athlon 64 X2 and Pentium D processors cannot benefit from their second core that much here.
It is also very interesting that the 32-bit DivX codec works slightly faster than the 64-bit version of it in Microsoft Windows XP Professional x64 Edition OS. The advantage of the 32-bit program version makes about 3-5%.
During our previous dual-core processor testing sessions we have already mentioned a few times that Windows Media Encoder is an excellent example of an application using two processor cores. So, Athlon 64 X2 3800+ appears over 30% faster than Athlon 64 3800+, even though the dual-core processor works at 17% lower clock frequency. All in all, Athlon 64 X2 3800+ manages to defeat even Pentium D 830, although NetBurst architecture proves pretty efficient during media data encoding.
The popular SuperPi benchmark doesn’t support multi-threading. That is why dual-core processors turn out slower here that single-core ones.
ScienceMark 2.0 benchmark is quite an interesting one, I should say. Firstly, it supports all contemporary instructions sets and multi-threading, and secondly, there is a revision for Microsoft Windows XP Professional x64 Edition. And the remarkable thing is that the use of 64-bit code for math1ematical modeling tasks performed within this benchmark ensures a pretty significant performance growth. In Molecular Dynamics test it even exceeds 100%.
This test loads the processor computational resources to the full extent and AMD processors run faster in this test than their competitors from Intel. The new Athlon 64 X2 3800+ outperforms its single-core brother in both subtests and becomes an indisputable leader.
Adobe Photoshop CS2 supports multi-threading and Athlon 64 X2 3800+ appears the fastest here leaving all the CPUs from the same price group far behind, even the Pentium D 830.
Athlon 64 X2 3800+ also wins in 3ds max test during scene final rendering. As you know tasks like that can be split into a few parallel threads very easily that is why Athlon 64 X2 3800+ appears 49% faster than the single-core Athlon 64 3800+, which an even greater advantage than in case of Windows Media Encoder 9 that we have just discussed a few moment ago.
And as we turn to working in Viewports, single-core CPUs take back the leadership.
By the way, I would like to draw your attention to a significant performance drop once we switch to the 64-bit OS version. It looks like the drives haven’t been optimized completely.
Photoshop and 3ds max are 32-bit applications. Unfortunately, there are no versions of these software packages yet that would work in Microsoft Windows XP Professional x64 Edition. Luckily, however, one of the professional 3D graphics packages is already available for x86-64. It is CINEMA 4D from MAXON. Of course, we couldn’t miss this great opportunity and tested our CPUs in this application with the help of a CINEBENCH 2003 test.
Like in 3ds max, the dual-core processor works faster than all the others during final rendering in CINEMA 4D too. Here you have to note that the final rendering speed in 64-bit work modes gets even higher, so it is definitely worth it to use dual-core 64-bit processors for tasks of this kind.
In OpenGL we can see the same effect as we have just seen in 3ds max. In this case, however, it shows itself in the native 64-bit application. The Microsoft Windows XP Professional x64 Edition and the application using processor Long Mode leads to a certain performance drop. It looks we will have to blame the drivers for that again. As for the performance of our today’s hero in OpenGL tests, single-core CPUs again prove the best out here.
Since the new AMD processor, Athlon 64 X2 3800+, appears the slowest model in the dual-core processor product line from AMD, it will be of our primary interest in terms of its overclocking potential. To test the overclocking abilities of this processor we assembled the same testbed as we have just used for performance benchmarks, i.e. based on DFI NF4 Ultra-D mainboard. For CPU cooling we used a Thermaltake CL-P0200 air cooler.
The default processor clock frequency multiplier of our Athlon 64 X2 3800+ equals 10x, so the only thing you can do with it is to make it lower (due to Cool’n’Quiet support). As a result the CPU can only be overclocked by increasing the clock generator frequency. In order to avoid being limited by the potential of the other system components, we locked the PCI and PCI Express bus frequencies at their nominal values for the entire test session, and the multiplier for Hyper-Transport bus was reduced to 4x. The memory frequency was also set with a reduced divider value, which would guarantee that the DIMM modules would work for sure even when the clock generator frequency grows up.
During our tests we found the maximum clock generator frequency when the processor remained stable. It was 240MHz. To go beyond this limit we had to slightly increase the processor Vcore: we raised it to 1.45V. The resulting CPU frequency in this case made 2.4GHz.
So, during our overclocking experiments we managed to raise the frequency of our Athlon 64 X2 3800+ on Manchester core by 20%. I have to stress that it is not that much at all, as dual-core Athlon 64 X2 4800+ and Athlon 64 X2 4600+ processors work at exactly the same frequency. And the latter is based exactly on the same Manchester core. In other words, we managed to overclock our Athlon 64 X2 3800+ only to the level of Athlon 64 X2 4600+. Looks like AMD doesn’t use the best cores for the youngest processors in its dual-core family. For example, when we tested Athlon 64 X2 4800+ processor, though this one was based on Toledo core, we managed to conquer the 2.7GHz bar.
Well, anyway, be happy with what you’ve got. In order to give you a better idea of how fast the overclocked Athlon 64 X2 3800+ actually is compared with the top AMD processors, we ran a few specific tests. We compared our today’s main hero against the Athlon 64 FX-57 and Athlon 64 X2 4800+. For a more illustrative result we set the memory at 200MHz in all benchmarks and adjusted its timings to 2-2-2-10.
As we see, the Athlon 64 X2 3800+ overclocked to 2.4GHz fails to win the leadership in all benchmarks. Although we have to admit that it nevertheless demonstrates very nice performance. For example in applications supporting multi-threading it manages to outpace Athlon 64 FX-57. As for its lag behind the Athlon 64 X2 4800+ equipped with 1MB L2 cache for each of the cores, it is not that big at all: only 1-2% on the average.
However, there were a few applications that turned out very critical to the cache-memory size. In applications of this time the performance difference between the overclocked Athlon 64 X2 3800+ and the Athlon 64 X2 4800+ can reach up to 10%. Although this can hardly upset the happy Athlon 64 X2 3800+ owners, as their solution costs half as much as Athlon 64 X2 4800+ and Athlon 64 FX-57.
The launch of Athlon 64 X2 3800+ processor indicated that AMD dropped down the price bar for dual-core processor systems. Now mainstream platforms can easily acquire not only dual-core solutions from Intel, but also dual-core solutions from AMD. This way, the release of Athlon 64 X2 3800+ balanced out the situation somehow: now both companies offer not only extremely expensive dual-core CPUs, but also similar processors for the mainstream segment.
We will not repeat what has already been said about the application of dual-core architectures in general. I would only like to say that according to the benchmark results, Athlon 64 X2 3800+ appeared a faster processor than its competitor from Intel, the Pentium D 830. So, it looks like this new AMD solution has pretty promising future ahead. Especially, if we take into account the compatibility of dual-core AMD processors with the existing infrastructure, their low heat dissipation, Cool’n’Quiet technology support and the ability to switch to 64-bit operating systems and corresponding applications.
As for the drawbacks of the newcomer, we have to say that for some reason we were not at all impressed with its overclocking potential, as it only reached 2.4GHz. However, even in this mode its performance is high enough so that it yields just a tiny bit to the faster models in Athlon 64 X2 and Athlon 64 FX processor families.